animal research papers

Journal of Animal Science and Biotechnology

Featured articles.

High expression circRALGPS2 in atretic follicle induces chicken granulosa cell apoptosis and autophagy via encoding a new protein (Haorong He et al.)

Staphylococcus aureus and biofilms: transmission, threats, and promising strategies in animal husbandry (Mengda Song et al.)

Featured series.

Interplay between feed, additives, and gut health of monogastric animals (Edited by: Rajesh Jha)

Nutritional modulation of embryonic development and potential in livestock and poultry (Edited by: Xiangfang Zeng)

• Most accessed
• Collections

Dietary eubiotics of microbial muramidase and glycan improve intestinal villi, ileum microbiota composition and production trait of broiler

Authors: Sungbo Cho, Shanmugam Suresh Kumar, Santiago Ramirez, Rolando Valientes and In Ho Kim

Hesperidin ameliorates H 2 O 2 -induced bovine mammary epithelial cell oxidative stress via the Nrf2 signaling pathway

Authors: Qi Huang, Jiashuo Liu, Can Peng, Xuefeng Han and Zhiliang Tan

Dietary fat supplementation relieves cold temperature-induced energy stress through AMPK-mediated mitochondrial homeostasis in pigs

Authors: Wei He, Xinyu Liu, Ye Feng, Hongwei Ding, Haoyang Sun, Zhongyu Li and Baoming Shi

Probiotic Pediococcus pentosaceus restored gossypol-induced intestinal barrier injury by increasing propionate content in Nile tilapia

Authors: Feifei Ding, Nannan Zhou, Yuan Luo, Tong Wang, Weijie Li, Fang Qiao, Zhenyu Du and Meiling Zhang

Endotoxin-induced alterations of adipose tissue function: a pathway to bovine metabolic stress

Authors: Miguel Chirivi and G. Andres Contreras

Most recent articles RSS

View all articles

Structures and characteristics of carbohydrates in diets fed to pigs: a review

Authors: Diego M. D. L. Navarro, Jerubella J. Abelilla and Hans H. Stein

Lysozyme as an alternative to growth promoting antibiotics in swine production

Authors: W. T. Oliver and J. E. Wells

GMOs in animal agriculture: time to consider both costs and benefits in regulatory evaluations

Authors: Alison L Van Eenennaam

Clostridium species as probiotics: potentials and challenges

Authors: Pingting Guo, Ke Zhang, Xi Ma and Pingli He

Microbial synthesis of zinc oxide nanoparticles and their potential application as an antimicrobial agent and a feed supplement in animal industry: a review

Authors: Hidayat Mohd Yusof, Rosfarizan Mohamad, Uswatun Hasanah Zaidan and Nor’ Aini Abdul Rahman

Most accessed articles RSS

Thematic series Advances in Silage Research Edited by: Prof. Xusheng Guo Collection published: 22 January 2024

Thematic series Feed Safety Risk Assessment and Related Detection Technology  (upcoming) Edited by: Prof. Yiqiang Chen

Cross-journal collection Gut microbiome and animal health Collection published: 16 November 2022

Thematic series Poultry genetics and genomics Edited by: Prof. Jie Wen Collection published: 11 October 2022

Thematic series Interplay between feed, additives, and gut health of monogastric animals Edited by: Prof. Rajesh Jha Collection published: 6 October 2022

Thematic series Nutritional modulation of embryonic development and potential in livestock and poultry Edited by: Prof. Xiangfang Zeng Collection published: 2 March 2022

Thematic series Biological feed and animal gut health Edited by: Prof. Haifeng Wang Collection published: 12 October 2021

Thematic series Mycotoxin Control in Animal Feed Edited by: Prof. Lv-Hui Sun Collection published: 11 October 2021

Thematic series The progress of low protein diet in poultry Edited by: Prof. Jianmin Yuan Collection published: 6 January 2021

Thematic series Advances on the interactions among nutrition, genomics, and physiology in dairy cattle Edited by: Prof Juan J. Loor Collection published: 29 June 2020 Special Issue Special Issue for JASB 10th Anniversary Edited by: Prof Dorian Garrick Collection published: 17 April 2020

Thematic series Gut microbiota of poultry Edited by: Prof Guolong Zhang Collection published: 13 March 2020

Thematic series Pig gut microbiota: Challenges and opportunities to improve the pig Edited by: Prof Paolo Trevisi and Prof Jürgen Zentek Collection published: 24 May 2019

Thematic series Frontier research in intestinal health of pigs and chickens Edited by: Prof Sung Woo Kim Collection published: 14 January 2019

Thematic series Genome editing in domestic animals Edited by: Prof Jae Yong Han Collection published: 29 January 2018

Thematic series Gastrointestinal microbial ecology and functionality Edited by: Prof Jianxin Liu, Prof Weiyun Zhu Collection published: 8 November 2016

Thematic series Alternatives to antibiotics as growth promoters for use in swine production Edited by: Prof Phil Thacker Collection published: 21 October 2015

Thematic series Fat nutrition and metabolism in food animals Edited by: Prof Jack Odle Collection published: 21 May 2015

Thematic series Animal reproductive biology Edited by: Prof Fuller Bazer Collection published: 26 March 2015

Thematic series Special Issue for WCAP 2013 Edited by: Prof Defa Li Collection published: 30 August 2013

Thematic series Special Issue for Chinese Swine Industry Symposium Collection published: 8 July 2013

Thematic series Animal Production Collection published: 14 March 2013

Thematic series Reproduction and physiology Collection published: 14 March 2013

Thematic series Animal Nutrition Collection published: 14 March 2013

Thematic series Animal Genetics Collection published: 14 March 2013

Call for papers

We invite you to submit an article to the following thematic series:

Mucosal immunology for nutritional impacts on young animals Submission deadline: August 31, 2024

Thematic series

Advances in Silage Research Edited by: Xusheng Guo

Gut microbiome and animal health

Poultry genetics and genomics Edited by: Jie Wen

Interplay between feed, additives, and gut health of monogastric animals Edited by: Rajesh Jha

Nutritional modulation of embryonic development and potential in livestock and poultry Edited by: Xiangfang Zeng

View all article collections

Aims and scope

Journal of Animal Science and Biotechnology  is an open access, peer-reviewed journal that encompasses a wide range of research areas including animal genetics, reproduction, nutrition, physiology, biochemistry, biotechnology, feedstuffs and animal products. The journal publishes original and novel research articles and reviews mainly involved in pigs, poultry, beef cattle, cows, goats and sheep, but the studies involving aquatic and laboratory animal species that address fundamental questions related to livestock are also welcome.

• Editorial Board
• Manuscript editing services
• Instructions for Editors
• Sign up for article alerts and news from this journal
• Follow us on Twitter

Annual Journal Metrics

2022 Citation Impact 7.0 - 2-year Impact Factor 7.3 - 5-year Impact Factor 1.821 - SNIP (Source Normalized Impact per Paper) 1.423 - SJR (SCImago Journal Rank)

2023 Speed 7 days submission to first editorial decision for all manuscripts (Median) 119 days submission to accept (Median)

2023 Usage 1,188,041 downloads 583 Altmetric mentions

• More about our metrics

ISSN: 2049-1891

• Submission enquiries: Access here and click Contact Us
• General enquiries: [email protected]

Click through the PLOS taxonomy to find articles in your field.

For more information about PLOS Subject Areas, click here .

Animal Research

Author requirements.

• Studies involving animals must be conducted according to internationally-accepted standards.
• Authors must obtain prior approval from their Institutional Animal Care and Use Committee (IACUC) or equivalent ethics committee(s).
• The name of the IACUC or equivalent ethics committee, as well as relevant permit numbers, in addition to any other pertinent experimental details, must be provided at submission.
• The journals’ editorial teams reserve the right to request additional information in relation to experiments on vertebrates or higher invertebrates as necessary for the evaluation of the manuscript e.g. in the context of appropriate animal welfare or studies that involve death as an experimental endpoint.

Reporting Guidelines

We encourage authors using vertebrates or cephalopods in their research to comply with the ARRIVE guidelines (see also the publications on the guidelines and elaboration document ). The ARRIVE guidelines aim to improve standards of reporting to ensure that the data can be adequately interpreted, reproduced, and utilized. Where research could be confused as pertaining to human clinical research, the animal model should also be noted in the article title.

Non-Human Primates

Non-human primate studies must be performed in accordance with the recommendations of the Weatherall report, The use of non-human primates in research . Manuscripts describing research involving non-human primates must include details of animal welfare, including information about housing, feeding, and environmental enrichment, and steps taken to minimize suffering, including use of anesthesia and method of sacrifice if appropriate.

Unregulated Research

Where unregulated animals are used or ethics approval is not required by a specific committee, authors should include a clear statement of this fact and the reasons why ethical approval is not required. The journal staff and editors will assess these situations on a case-by-case basis.

Paleontology and Archaeology Research

PLOS journals require that manuscripts reporting paleontology and archaeology research include descriptions of methods and specimens in sufficient detail to allow the work to be reproduced. Data sets supporting statistical and phylogenetic analyses should be provided, preferably in a format that allows easy re-use.

Under the PLOS data availability policy , any specimen that is erected as a new species, described, or figured must be deposited in an accessible, permanent repository (i.e., public museum or similar institution). If study conclusions depend on specimens that do not fit these criteria, the article will be rejected.

Specimen numbers and complete repository information, including museum name and geographic location, are required for publication. Locality information should be provided in the manuscript as legally allowable, or a statement should be included giving details of the availability of such information to qualified researchers.

If permits were required for any aspect of the work, details should be given of all permits that were obtained, including the full name of the issuing authority. PLOS journals will not publish research on specimens that were obtained without necessary permission or were illegally exported.

Policy Enforcement

All submissions describing research involving animals will be checked by journal staff and editors to ensure that the requirements above are met. Failure to meet requirements may be grounds for rejection.

We reserve the right to reject work that the editors believe has not been conducted to a high ethical standard, even if authors have obtained formal approval or if approval is not required under local regulations.

If concerns are discovered after publication, the journal staff will investigate and, should substantial concerns arise regarding the handling of animals or oversight for the research, we may issue a correction or retraction as appropriate. We also reserve the right to contact the authors’ institution, ethics committee or other appropriate body in relation to these concerns.

• Search Menu
• Browse content in Arts and Humanities
• Browse content in Archaeology
• Prehistoric Archaeology
• Browse content in Art
• History of Art
• Browse content in Classical Studies
• Classical History
• Classical Literature
• Classical Reception
• Greek and Roman Archaeology
• Digital Humanities
• Browse content in History
• Diplomatic History
• Environmental History
• Genocide and Ethnic Cleansing
• History by Period
• Legal and Constitutional History
• Regional and National History
• Social and Cultural History
• Theory, Methods, and Historiography
• World History
• Browse content in Language Teaching and Learning
• Language Teaching Theory and Methods
• Browse content in Linguistics
• Applied Linguistics
• Language Evolution
• Language Families
• Lexicography
• Browse content in Literature
• Bibliography
• Literary Studies (American)
• Literary Studies (20th Century onwards)
• Literary Studies (British and Irish)
• Literary Studies (Women's Writing)
• Literary Theory and Cultural Studies
• Shakespeare Studies and Criticism
• Browse content in Media Studies
• Browse content in Music
• Applied Music
• Medicine and Music
• Music Theory and Analysis
• Musical Structures, Styles, and Techniques
• Musicology and Music History
• Browse content in Philosophy
• Aesthetics and Philosophy of Art
• Epistemology
• History of Western Philosophy
• Metaphysics
• Moral Philosophy
• Philosophy of Science
• Philosophy of Mind
• Philosophy of Mathematics and Logic
• Practical Ethics
• Browse content in Religion
• Christianity
• Judaism and Jewish Studies
• Religion and Science
• Religion and Law
• Religion and Art, Literature, and Music
• Religious Studies
• Browse content in Society and Culture
• Ethical Issues and Debates
• Browse content in Law
• Arbitration
• Company and Commercial Law
• Comparative Law
• Competition Law
• Browse content in Constitutional and Administrative Law
• Parliamentary and Legislative Practice
• Employment and Labour Law
• Environment and Energy Law
• Financial Law
• History of Law
• Human Rights and Immigration
• Intellectual Property Law
• Browse content in International Law
• Private International Law and Conflict of Laws
• Public International Law
• IT and Communications Law
• Jurisprudence and Philosophy of Law
• Law and Society
• Legal System and Practice
• Medical and Healthcare Law
• Browse content in Medicine and Health
• Browse content in Allied Health Professions
• Dietetics and Nutrition
• Physiotherapy
• Radiography
• Anaesthetics
• Browse content in Clinical Medicine
• Acute Medicine
• Cardiovascular Medicine
• Clinical Pharmacology and Therapeutics
• Dermatology
• Endocrinology and Diabetes
• Gastroenterology
• Geriatric Medicine
• Infectious Diseases
• Medical Toxicology
• Medical Oncology
• Rheumatology
• Sleep Medicine
• Clinical Neuroscience
• Community Medical Services
• Critical Care
• Forensic Medicine
• History of Medicine
• Browse content in Medical Dentistry
• Restorative Dentistry and Orthodontics
• Medical Skills
• Medical Ethics
• Medical Statistics and Methodology
• Browse content in Neurology
• Neuropathology
• Nursing Studies
• Browse content in Obstetrics and Gynaecology
• Gynaecology
• Occupational Medicine
• Paediatrics
• Browse content in Pathology
• Clinical Cytogenetics and Molecular Genetics
• Medical Microbiology and Virology
• Patient Education and Information
• Browse content in Pharmacology
• Psychopharmacology
• Browse content in Preclinical Medicine
• Molecular Biology and Genetics
• Reproduction, Growth and Development
• Primary Care
• Professional Development in Medicine
• Browse content in Psychiatry
• Child and Adolescent Psychiatry
• Forensic Psychiatry
• Browse content in Public Health and Epidemiology
• Epidemiology
• Public Health
• Browse content in Radiology
• Clinical Radiology
• Interventional Radiology
• Radiation Oncology
• Reproductive Medicine
• Browse content in Surgery
• Cardiothoracic Surgery
• Gastro-intestinal and Colorectal Surgery
• Neurosurgery
• Plastic and Reconstructive Surgery
• Trauma and Orthopaedic Surgery
• Browse content in Science and Mathematics
• Browse content in Biological Sciences
• Aquatic Biology
• Biochemistry
• Bioinformatics and Computational Biology
• Developmental Biology
• Ecology and Conservation
• Evolutionary Biology
• Genetics and Genomics
• Microbiology
• Molecular and Cell Biology
• Plant Sciences and Forestry
• Research Methods in Life Sciences
• Structural Biology
• Systems Biology
• Zoology and Animal Sciences
• Browse content in Chemistry
• Medicinal Chemistry
• Mineralogy and Gems
• Physical Chemistry
• Browse content in Computer Science
• Artificial Intelligence
• Computer Architecture and Logic Design
• Human-Computer Interaction
• Mathematical Theory of Computation
• Browse content in Computing
• Computer Security
• Computer Networking and Communications
• Browse content in Earth Sciences and Geography
• Atmospheric Sciences
• Environmental Geography
• Geology and the Lithosphere
• Meteorology and Climatology
• Browse content in Engineering and Technology
• Agriculture and Farming
• Biological Engineering
• Civil Engineering, Surveying, and Building
• Energy Technology
• Engineering (General)
• Environmental Science, Engineering, and Technology
• Transport Technology and Trades
• Browse content in Environmental Science
• Environmental Sustainability
• Management of Land and Natural Resources (Environmental Science)
• Browse content in Materials Science
• Ceramics and Glasses
• Composite Materials
• Nanotechnology
• Browse content in Mathematics
• Applied Mathematics
• Biomathematics and Statistics
• Mathematical Education
• Mathematical Analysis
• Probability and Statistics
• Pure Mathematics
• Browse content in Neuroscience
• Cognition and Behavioural Neuroscience
• Neuroscientific Techniques
• Browse content in Physics
• Astronomy and Astrophysics
• Classical Mechanics
• Relativity and Gravitation
• Browse content in Psychology
• Clinical Psychology
• Cognitive Psychology
• Cognitive Neuroscience
• Health Psychology
• Music Psychology
• Neuropsychology
• Organizational Psychology
• Browse content in Social Sciences
• Browse content in Anthropology
• Human Evolution
• Browse content in Business and Management
• Human Resource Management
• Industrial and Employment Relations
• Industry Studies
• Information and Communication Technologies
• Organizational Theory and Behaviour
• Public and Nonprofit Management
• Browse content in Criminology and Criminal Justice
• Criminology
• Browse content in Economics
• Agricultural, Environmental, and Natural Resource Economics
• Behavioural Economics and Neuroeconomics
• Econometrics and Mathematical Economics
• Economic History
• Economic Development and Growth
• Financial Markets
• Financial Institutions and Services
• Health, Education, and Welfare
• Labour and Demographic Economics
• Law and Economics
• Public Economics
• Urban, Rural, and Regional Economics
• Browse content in Education
• Schools Studies
• Teaching of Specific Groups and Special Educational Needs
• Environment
• Browse content in Human Geography
• Economic Geography
• Browse content in Interdisciplinary Studies
• Communication Studies
• Museums, Libraries, and Information Sciences
• Browse content in Politics
• Foreign Policy
• Gender and Politics
• International Relations
• International Organization (Politics)
• Political Sociology
• Political Behaviour
• Political Economy
• Political Institutions
• Political Theory
• Public Administration
• Public Policy
• Quantitative Political Methodology
• Regional Political Studies
• Security Studies
• Browse content in Regional and Area Studies
• African Studies
• Japanese Studies
• Research and Information
• Browse content in Social Work
• Addictions and Substance Misuse
• Browse content in Sociology
• Economic Sociology
• Gender and Sexuality
• Gerontology and Ageing
• Health, Illness, and Medicine
• Migration Studies
• Race and Ethnicity
• Social Movements and Social Change
• Social Research and Statistics
• Social Stratification, Inequality, and Mobility
• Sociology of Religion
• Urban and Rural Studies
• Journals A to Z
• Books on Oxford Academic

Articles making an impact in Animal Science and Zoology

Discover impactful articles published in our animal science and zoology journal portfolio with our High-Impact Research collections, featuring the most read, most cited, and most discussed articles published in recent years, which have caught the interest of your peers.

Animal Science

Ornithology, high-impact research from behavioral ecology.

Behavioral Ecology  is broad-based and covers both empirical and theoretical approaches and published studies on the whole range of behaving organisms, including plants, invertebrates, vertebrates, and humans.

High-Impact Research from BioScience

BioScience presents timely and authoritative overviews of current research in biology, accompanied by essays and discussion sections on education, public policy, history, and the conceptual underpinnings of the biological sciences.

High-Impact Research from Current Zoology

Open access, Current Zoology  publishes review articles and research papers in the fields of ecology, evolution and behaviour.

High-Impact Research from Integrative and Comparative Biology

Integrative and Comparative Biology  publishes forward-looking reviews, synthesis, perspectives and empirical articles in integrative, comparative and organismal biology.

High-Impact Research from Animal Frontiers

Animal Frontiers publishes discussion and position papers that present international perspectives on high-impact, global issues in animal agriculture. 

High-Impact Research from Journal of Animal Science

Journal of Animal Science provides new knowledge and perspectives across a range of topics in both animal production and fundamental aspects of genetics, nutrition, physiology, and the preparation and utilization of animal products.

High-Impact Research from Translational Animal Science

Translational Animal Science encompasses a broad scope of research topics in animal science, focusing on translating basic science to innovation.

Aquatic Science

High-impact research from journal of crustacean biology.

The  Journal of Crustacean Biology  publishes articles of broad interest on the biology of crustaceans and other marine arthropods.

High-Impact Research from Journal of Molluscan Studies

Journal of Molluscan Studies  publishes research on the biology of molluscs, including the developing subjects of molecular genetics, cladistic phylogenetics and ecophysiology, as well as ecological, behavioural and systematic malacology.

High-Impact Research from Annals of the Entomological Society of America

Annals of the Entomological Society of America publishes cutting-edge research, reviews, and collections on a common topic of broad interest, across the entomological disciplines.

High-Impact Research from Arthropod Management Tests

Arthropod Management Tests publishes short reports from a single year on a routine screening test for management of arthropods that may be harmful or beneficial.

High-Impact Research from Environmental Entomology

Environmental Entomology  reports on the interaction of insects with the biological, chemical, and physical aspects of their environment.

High-Impact Research from Insect Systematics and Diversity

Insect Systematics and Diversity publishes original research on systematics, evolution, and biodiversity of insects and related arthropods. 

High-Impact Research from Journal of Economic Entomology

The Journal of Economic Entomology is the most-cited entomological journal and publishes articles on the economic significance of insects and other arthropods.

High-Impact Research from Journal of Insect Science

The  Journal of Insect Science  publishes articles based on original research, as well as Reviews, interpretive articles in a Forum section, and Short Communications in all fields of entomology.

High-Impact Research from Journal of Integrated Pest Management

An extension-focused journal publishing original articles on any aspect of integrated pest management. The  Journal of Integrated Pest Management  is targeted at a non-technical audience of farmers, pest control operators, foresters, and others.

High-Impact Research from Journal of Medical Entomology

Journal of Medical Entomology publishes reports on all phases of medical entomology and acarology, including the systematics and biology of insects, acarines, and other arthropods of public health and veterinary significance.

High-Impact Research from Journal of Mammalogy

Promoting interest in mammals throughout the world, the Journal of Mammalogy publishes research on all aspects of the biology of mammals, including behavior, conservation, ecology, genetics, morphology, physiology, and taxonomy. 

High-Impact Research from Mammalian Species

Mammalian Species  is published by the American Society of Mammalogists with 15-35 individual species accounts issued each year, summarizing the current understanding of the biology of an individual species including systematics, distribution, fossil history, genetics, anatomy, physiology, behavior, ecology, and conservation.

High-Impact Research from Ornithology

Ornithology  publishes original research from all parts of the globe that tests fundamental, scientific hypotheses through ornithological studies and advances our understanding of living or extinct bird species. 

High-Impact Research from Ornithological Applications

Ornithological Applications  publishes original research, syntheses, and assessments on the application of scientific theory, ornithological knowledge, and methods to the conservation and management of birds and to policy.

Affiliations

• Copyright © 2024
• About Oxford Academic
• Publish journals with us
• University press partners
• What we publish
• New features  
• Open access
• Institutional account management
• Rights and permissions
• Get help with access
• Accessibility
• Advertising
• Media enquiries
• Oxford University Press
• Oxford Languages
• University of Oxford

Oxford University Press is a department of the University of Oxford. It furthers the University's objective of excellence in research, scholarship, and education by publishing worldwide

• Copyright © 2024 Oxford University Press
• Cookie settings
• Cookie policy
• Privacy policy
• Legal notice

This Feature Is Available To Subscribers Only

Sign In or Create an Account

This PDF is available to Subscribers Only

For full access to this pdf, sign in to an existing account, or purchase an annual subscription.

Key Topics in Surgical Research and Methodology pp 207–228 Cite as

An Introduction to Animal Research

• James Kinross 3 &
• Lord Ara Darzi 4  

2862 Accesses

Despite advances in computer modelling and bioinformatics, animal models remain a vital component of biomedical research. The growth in this area of work is in part due to the evolution next generation of biotechnologies, which more than ever necessitate the need for in vivo experimentation. An understanding of the principals of animal research therefore remains a necessity for medical researchers as it permits scientific analysis to be interpreted in a more critical and meaningful manner. Initiating and designing an animal experiment can be a daunting process, particularly as the law and legislation governing animal research is complex and new specialist skills must be acquired. This chapter reviews the principles of animal research and provides a practical resource for those researchers seeking to create robust animal experiments that ensure minimal suffering and maximal scientific validity.

This is a preview of subscription content, log in via an institution .

Buying options

• Available as PDF
• Read on any device
• Instant download
• Own it forever
• Available as EPUB and PDF
• Compact, lightweight edition
• Dispatched in 3 to 5 business days
• Free shipping worldwide - see info

Tax calculation will be finalised at checkout

Purchases are for personal use only

Abbreviations

The American College of Laboratory Animal Medicine

Animal and Plant Health Inspection Service

Animal Welfare Act

Control of substances hazardous to health

European Coalition for Biomedical Research

The Food and Drug Agency

Health and Safety Executive

The Human Fertilisation and Embryology Authority

Institutional Animal Care and Use Committee

Individually ventilated cage

In vitro fertilisation

Laboratory animal allergy

Named animal care and welfare officer

National Institute for Clinical Excellence

Named veterinary surgeon

The Office of Laboratory Animal Welfare

Public Health Service

Personal License under the Scientific (Animal Procedures) Act 1986

Project License under the Scientific (Animal Procedures) Act 1986

Royal Society for the Prevention of Cruelty to Animals

Specified pathogen free

The United States Department of Agriculture

(2006) Animals (scientific procedures) inspectorate annual report. Home Office: Science, Research and Statistics, London

Google Scholar  

Davidson N (2006) Davidson review: implementation of EU legislation. http://wwwhm-treasurygovuk/independent_reviews/davidson_review/davidson_indexcfm

Fox JG, Cohen BJ, Loew FM (1984) Laboratory animal medicine. Academic press, New York

RDS, Understanding Animal Research in Medicine, Coalition for Medical Progress (2007) Medical advances and animal research: the contribution of animal science to the medical revolution: some case histories. Available at: http://www.pro-test.org.uk/MAAR.pdf

Foex BA (2007) The ethics of animal experimentation. Emerg Med J 24:750–751

Article   PubMed   Google Scholar  

Matfield M (2002) Animal experimentation: the continuing debate. Nat Rev Drug Discov 1:149–152

Rollin BE (2007) Animal research: a moral science. Talking Point on the use of animals in scientific research. EMBO Rep 8:521–525

Article   PubMed   CAS   Google Scholar  

Singer P (2006) In defense of animals: the second wave. Blackwell, Malden, MA

Croce P (1999) Vivisection or science? An investigation into testing drugs and safeguarding health. Zed Books, London

Schechter AN, Rettig RA (2002) Funding priorities for medical research. JAMA 288:832; author reply 832

Pound P, Ebrahim S, Sandercock P et al (2004) Where is the evidence that animal research benefits humans? BMJ 328:514–517

Perel P, Roberts I, Sena E et al (2007) Comparison of treatment effects between animal experiments and clinical trials: systematic review. BMJ 334:197

Mayor S (2007) UK regulatory body wants public consultation on human—animal hybrid research. BMJ 334:112

Anon (2007) Avoiding a chimaera quagmire. Nature 445:1

Article   Google Scholar  

Karpowicz P, Cohen CB, van der Kooy D (2004) It is ethical to transplant human stem cells into nonhuman embryos. Nat Med 10:331–335

Thompson SG (1994) Why sources of heterogeneity in meta-analysis should be investigated. BMJ 309:1351–1355

Auer JA, Goodship A, Arnoczky S et al (2007) Refining animal models in fracture research: seeking consensus in optimising both animal welfare and scientific validity for appropriate biomedical use. BMC Musculoskelet Disord 8:72

Sarker SK, Patel B (2007) Simulation and surgical training. Int J Clin Pract 61:2120–2125

Dolan K (2007) Laboratory animal law: legal control of the use of animals in research. Blackwell, New York

Book   Google Scholar  

Rice M (2007) Deadline approaches for animal experimentation directive. Eur J Cancer 43:1641

PubMed   Google Scholar  

Field K, Bailey M, Foresman LL et al (2007) Medical records for animals used in research, teaching, and testing: public statement from the American College of Laboratory Animal Medicine. ILAR J 48:37–41

Hauser R, Marczak M, Karaszewski B et al (2008) A preliminary study for identifying olfactory markers of fear in the rat. Lab Anim (NY) 37:76–80

Vermeulen JK, de Vries A, Schlingmann F, Remie R. (1997) Food Deprivation: common sense or nonsense? Animal Tech 48:45–54

Levine S, Saltzman A (2000) Feeding sugar overnight maintains metabolic homeostasis in rats and is preferable to overnight starvation. Lab Anim 34:301–306

Brown C (2007) Endotracheal intubation in the dog. Lab Anim (NY) 36:23–24

Price H (2007) Intubating rabbits. Vet Rec 160:744

Spoelstra EN, Ince C, Koeman A et al (2007) A novel and simple method for endotracheal intubation of mice. Lab Anim 41:128–135

Buscaglia JM (2007) Animal laboratory endoscopic research: a fellow's perspective. Gastrointest Endosc 65:882–883

Fritscher-Ravens A, Patel K, Ghanbari A et al (2007) Natural orifice transluminal endoscopic surgery (NOTES) in the mediastinum: long-term survival animal experiments in transesoph-ageal access, including minor surgical procedures. Endoscopy 39:870–875

Radostits OM, Mayhew IG, Houston DM (2000) Veterinary clinical examination and diagnosis, 2nd edn. W.B. Saunders Company, Philadelphia

James Thomson et al, “Embryonic stem cell lines derived from human blastocysts,” Science 282:1145–1147, November 6, 1998

The National Academies' Guidelines for Human Embryonic Stem Cell Research. 2008. http://books.nap.edu/catalog.php?record_id=12260

Download references

Author information

Authors and affiliations.

Department of Biosurgery and Surgical Technology, Imperial College, 10th floor, QEQM, St. Mary's Hospital, Praed Street, London, W2 1NY, UK

James Kinross

The Department of Biosurgery and Surgical Technology, Imperial College London, 10th Floor, Queen Elizabeth the Queen Mother (QEQM) Building, Imperial College Healthcare NHS Trust, St Mary's Hospital Campus, Praed Street, London, W2 1NY, UK

Lord Ara Darzi

You can also search for this author in PubMed   Google Scholar

Editor information

Editors and affiliations.

Dept. Biosurgery & Surgical Technology, Imperial College London St. Mary's Hospital London, 10th floor QEQM Bldg. Praed Street, London, W2 1NY, United Kingdom

Thanos Athanasiou  & Ara Darzi  & 

UCSF Global Health Sciences, 3333 California Street, Suite 285, San Francisco L, CA, 94143-0443, USA

Haile Debas

Rights and permissions

Reprints and permissions

Copyright information

© 2010 Springer-Verlag Berlin Heidelberg

About this chapter

Cite this chapter.

Kinross, J., Darzi, L.A. (2010). An Introduction to Animal Research. In: Athanasiou, T., Debas, H., Darzi, A. (eds) Key Topics in Surgical Research and Methodology. Springer, Berlin, Heidelberg. https://doi.org/10.1007/978-3-540-71915-1_17

Download citation

DOI : https://doi.org/10.1007/978-3-540-71915-1_17

Publisher Name : Springer, Berlin, Heidelberg

Print ISBN : 978-3-540-71914-4

Online ISBN : 978-3-540-71915-1

eBook Packages : Medicine Medicine (R0)

Share this chapter

Anyone you share the following link with will be able to read this content:

Sorry, a shareable link is not currently available for this article.

Provided by the Springer Nature SharedIt content-sharing initiative

• Publish with us

Policies and ethics

• Find a journal
• Track your research

Thank you for visiting nature.com. You are using a browser version with limited support for CSS. To obtain the best experience, we recommend you use a more up to date browser (or turn off compatibility mode in Internet Explorer). In the meantime, to ensure continued support, we are displaying the site without styles and JavaScript.

• View all journals

Animal biotechnology articles from across Nature Portfolio

Animal biotechnology is a branch of biotechnology in which molecular biology techniques are used to genetically engineer (i.e. modify the genome of) animals in order to improve their suitability for pharmaceutical, agricultural or industrial applications. Animal biotechnology has been used to produce genetically modified animals that synthesize therapeutic proteins, have improved growth rates or are resistant to disease.

Related Subjects

• Genetic engineering

Latest Research and Reviews

Enhanced high β-carotene yeast cell production by Rhodotorula paludigena CM33 and in vitro digestibility in aquatic animals

• Namphet Thumkasem
• Thapanut On-mee
• Theppanya Charoenrat

Comprehensive assessment of goat adipose tissue-derived mesenchymal stem cells cultured in different media

• Michelle Abraham
• Sandeep Goel

3D printed orthopedic prostheses for domestic and wild birds—case reports

• Lucas Rannier R. A. Carvalho

UFObow: A single-wavelength excitable Brainbow for simultaneous multicolor ex-vivo and in-vivo imaging of mammalian cells

UFObow is a single-wavelength excitable Brainbow technique, incorporating three newly developed blue-excitable fluorescent proteins. This method facilitates mapping of immune cells’ spatial distribution at a single-cell resolution.

• Fangfang Yang

Non-encapsulated, encapsulated, and lyophilized probiotic Limosilactobacillus reuteri SW23 influenced the growth and gut health in calves

• Manish Yadav
• Sachin Kumar
• Ashis Kumar Samanta

Gene targeting in adult organs using in vivo cleavable donor plasmids for CRISPR-Cas9 and CRISPR-Cas12a

• Riki Ishibashi
• Ritsuko Maki
• Fumiko Toyoshima

News and Comment

Beef paddies

An article in Matter uses rice grains as scaffolds for cultivated beef cells to produce an innovative beef-infused rice.

• Ariane Vartanian

How biofabrication can accelerate cultured meat’s path to market

The production of conventional meat contributes to climate change and uses up around 70% of available arable land. Cultured meat is emerging as a potential solution, but presently can be only produced at the pilot scale. Biofabrication technologies developed for biomedical applications could be leveraged to introduce automation and standardization in the production of cultured meat, accelerating its path to market.

• Simon Heine
• Tilman Ahlfeld
• Petra J. Kluger

Sir Ian Wilmut 1944–2023

• Alan Trounson
• Jose Cibelli

Rapid-response manufacturing of adenovirus-vectored vaccines

• Carina C. D. Joe
• Nitin Chopra
• Alexander D. Douglas

Evolutionary divergence impact on de-extinction

Lin et al. explore the impact of evolutionary divergence on de-extinction efforts that use genome editing using the extinct Christmas Island rat.

Clarifying US regulations on xenotransplantation

• Peter Marks
• Steven Solomon

Quick links

• Explore articles by subject
• Guide to authors
• Editorial policies

• Open access
• Published: 06 March 2021

Research perspectives on animal health in the era of artificial intelligence

• Pauline Ezanno   ORCID: orcid.org/0000-0002-0034-8950 1 ,
• Sébastien Picault 1 ,
• Gaël Beaunée 1 ,
• Xavier Bailly 2 ,
• Facundo Muñoz 3 ,
• Raphaël Duboz 3 , 4 ,
• Hervé Monod 5 &
• Jean-François Guégan 3 , 6 , 7  

Veterinary Research volume  52 , Article number:  40 ( 2021 ) Cite this article

14k Accesses

30 Citations

10 Altmetric

Metrics details

Leveraging artificial intelligence (AI) approaches in animal health (AH) makes it possible to address highly complex issues such as those encountered in quantitative and predictive epidemiology, animal/human precision-based medicine, or to study host × pathogen interactions. AI may contribute (i) to diagnosis and disease case detection, (ii) to more reliable predictions and reduced errors, (iii) to representing more realistically complex biological systems and rendering computing codes more readable to non-computer scientists, (iv) to speeding-up decisions and improving accuracy in risk analyses, and (v) to better targeted interventions and anticipated negative effects. In turn, challenges in AH may stimulate AI research due to specificity of AH systems, data, constraints, and analytical objectives. Based on a literature review of scientific papers at the interface between AI and AH covering the period 2009–2019, and interviews with French researchers positioned at this interface, the present study explains the main AH areas where various AI approaches are currently mobilised, how it may contribute to renew AH research issues and remove methodological or conceptual barriers. After presenting the possible obstacles and levers, we propose several recommendations to better grasp the challenge represented by the AH/AI interface. With the development of several recent concepts promoting a global and multisectoral perspective in the field of health, AI should contribute to defract the different disciplines in AH towards more transversal and integrative research.

1 Introduction

Artificial intelligence (AI) encompasses a large range of theories and technologies used to solve problems of high logical or algorithmic complexity. It crosses many disciplines, including mechanistic modelling, software engineering, data science, and statistics (Figure  1 ). Introduced in the 1950s, many AI methods have been developed or extended recently with the improvement of computer performance. Recent developments have been fuelled by the interfaces created between AI and other disciplines, such as bio-medicine, as well as massive data from different fields, particularly those associated with healthcare [ 1 , 2 ].

Interactions between animal health (AH), artificial intelligence (AI), and closely related research domains. This illustration is pinpointing only the links between AH (in blue), AI and its main subfields (in red), and other related fields of research (in black). It can be naturally complexified through the interactions between AH and other research topics (e.g., human medicine) or between core disciplines (e.g., statistics and physics).

AI addresses three challenges that also make sense in animal health (AH): (1) understanding a situation and its dynamics, e.g., epidemic spread; (2) the perception of the environment, which corresponds in AH to the detection of patterns (e.g., repeated sequence of observations), forms (e.g., of a protein) and signals (e.g., increased mortality compared to a baseline) at different scales; (3) computer-based decision making, or, more realistically, human decision support (e.g., expert systems, diagnostic support, resource allocation).

To answer these challenges, a wide range of concepts and methods are developed in AI. This includes machine learning (ML), a widely known AI method nowadays, which has been developing since the 1980s [ 3 ]. Since the 2000s, deep learning is developing with the rise of big data and the continuous increasing of computing capacities, enabling the exploration of massive amount of information that cannot be processed by conventional statistical methods. In addition, this also includes methods and algorithms for solving complex problems, automating tasks or reasoning, integrating information from heterogeneous sources, or decision support (Figure  1 ). These methods are now uprising in the human health sector, but are still rarely used to study animal health issues that they would help to revisit.

Part of the scientific challenges faced in AH can be approached from a new perspective by using some of these AI methods to analyse the ever-increasing data collected on animals, pathogens, and their environment. AH research benefits from advances in machine and deep learning methods, e.g., in predictive epidemiology, individual-based precision medicine, and to study host–pathogen interactions [ 2 , 4 ]. These methods contribute to disease diagnosis and individual case detection, to more reliable predictions and reduced errors, to speed-up decisions and improved accuracy in risk analysis, and to better targeting interventions in AH [ 5 ]. AH research also benefits from scientific advances in other domains of AI. Knowledge representation and modelling of reasoning [ 6 ] allow more realistic representations of complex socio-biological systems such as those encountered in AH. Examples include processes related to decision-making and uncertainty management [ 7 , 8 ], as well as of patient life courses like in human epidemiology [ 9 ]. This contributes to making them more readable by noncomputer experts. In addition, advances in problem solving under constrained resource allocation [ 10 ], in autonomous agents [ 11 ], multi-agent systems [ 12 ], and multi-level systems [ 13 ], as well as on automatic computer code generation [ 14 ] can be mobilised to enhance efficient and reliable epidemiological models. Interestingly, this may aid to anticipate the effect of control and management decisions at different spatial and temporal scales (animal, herd, country…).

Conducting research at the AH/AI interface also leads to identify new challenges for AI, on themes common with human health but considering different contexts and perspectives [ 15 ]. First, taking into account the particular agro- and socio-economic conditions of production systems is crucial when dealing with AH. Animal production systems depend on human activities and decisions. They can be a source of income (e.g., livestock) or labour forces and source of food in family farming. Citizens have also high expectations in terms of ethics and animal welfare [ 16 ]. Conventional measures to control animal diseases may no longer be acceptable by society (e.g., mass culling during outbreaks [ 17 ], antimicrobial usage, [ 18 ]). Alternatives must be identified and assessed, and AI can contribute. For example, individual-based veterinarian medicine is emerging, mobilising both AI methods and new AH data streams, these data differing from data in human health [ 19 ]. The integration of data from deep sequencing in AH, including emerging technologies for studying the metabolome and epigenome, is also a challenge [ 20 , 21 ]. Second, interactions between animal species, in particular between domestic animals and wildlife, lead to specific infectious disease risks (e.g., multi-host pathogens such as for African swine fever, pathogens crossing the species barrier facilitated by frequent contacts and promiscuity). The intensity of such interactions could increase due to separate or synergistic actions of environmental (e.g., landscape homogenisation, land use change for agriculture development, climate change), demographic (e.g., increasing global demand for animal production) and societal (e.g., outdoor livestock management) pressures. In addition, working on multi-species disease networks provides crucial information on the underlying molecular mechanisms favouring interspecific transmission [ 22 ]. Third, animal populations are governed by recurrent decision-making that also impacts health management (e.g., trade, control measures). Economic criteria as consequences on livestock farmers’ incomes are therefore essential indicators for evaluating AH control strategies, which can sometimes be misunderstood or may be at odds with societal expectations. These specificities make the AH/AI interface a theme of interest to stimulate new methodological work and to solve some of old and current locks faced by AH research today. With the development of new concepts in health such as One Health, Ecohealth and Planetary Health, promoting multidisciplinarity, stakeholders’ participation, data sharing, and tackling the complexity of health issues (e.g., multi-host pathogen transmission, short and long-term climatic impacts on disease patterns [ 23 ]), AI could participate in this new development by making it possible to technically solve some of the complex problems posed.

Mobilising the literature published at the AH/AI interface between 2009 and 2019 (Additional file 1 A), focusing our literature search on mainly livestock and wildlife, as well as interviews conducted with French researchers positioned at this interface (Additional file 1 B), we identified the main research areas in AH in which AI is currently involved country-wide. We explored how AI methods contribute to revisiting AH questions and may help remove methodological or conceptual barriers within the field. We also analysed how AH questions interrogate and stimulate new AI technical or scientific developments. In this paper, we first discuss issues related to data collection, organisation and access (Section  1 ), then we discuss how AI methods contribute to revisiting our understanding of animal epidemiological systems (Section  2 ), to improving case detection and diagnosis at different scales (Section  3 ), and to anticipating pathogen spread and control in a wide range of scenarios in order to improve AH management, facilitate decision-making and stimulate innovation (Section  4 ). Finally, we present the possible obstacles and levers to the development of AI in modern AH (Section  5 ), before making recommendations to best address the new challenges represented by this AH/IA interface (Section  6 ).

2 Collect, organise and make accessible quality data

A central point for research in AH remains the quality and availability of data, at the different organization levels of living systems and therefore at different spatial and temporal scales [ 24 ]. Data of interest are diverse. They can be obtained thanks to molecular analysis (e.g., genomic, metagenomics, or metabolic data), from observational data on individuals (e.g., body temperature, behaviour, milk production and composition, weight, feed intake), or from the production system (e.g., herd structure, breeding, management of sanitary issues). They can also be obtained at larger scales, beyond herds or local groups of animals (e.g., epidemiological data, demographic events, commercial movements, meteorological data, land-use occupation).

Even though the acquisition of these massive and heterogeneous data remain challenging (e.g., metabolome data), a large and diverse amount is already collected: (i) through mandatory reporting in accordance with regulations (e.g., commercial movements of cattle, epidemio-surveillance platform), (ii) by automatic devices (e.g., sensors, video surveillance systems), and (iii) on an ad hoc basis as part of research programs. This leads to a very wide diversity of data properties, and therefore of their management, access and possible uses. These data can be specifically obtained for certain animals or herds (e.g., during cohort monitoring programs) or by private companies (e.g., pig trade movements such as in France, milk collection). This can limit accessibility to academics and public research. Globalisation and large-scale animal trade may generate the need to use data obtained at worldwide scale in AH, especially for quantitative epidemiology (e.g., transcontinental spread of pathogens, animal genetics and breed management) leading to standardisation issues [ 25 ].

Consideration should be given to future systems for observing, collecting and managing these data [ 26 ], and to practices aimed at better collaboration between stakeholders. While data management has always been an important element in applied research to facilitate their use and valorisation, it is now a strategic issue both in theoretical and more applied research, coupled with a technical and algorithmic challenge [ 27 , 28 , 29 ]. Indeed, producing algorithms to manage massive data flows and stocks, by optimising calculations, is a challenge, particularly in real time. It seems also necessary to make heterogeneous data sources interoperable, requiring dedicated methodological developments [ 25 ]. In addition, much of the data is private, with ownership often heterogeneous (e.g., multiple owners, non-centralised data, closed data) and sometimes unclear (e.g., lack of knowledge of the real owner of the data between, for example, farmers, the data collector or the farmers’ union). All this tends to considerably complicate access to the data, raises questions about intellectual property, and raises questions in relation to regulations with regards to data protection, e.g., the adaptation of regulation to AH while respecting the confidentiality of the personal data mobilised. What is the relevant business model for data collection or access to existing databases? What about the openness of AH data (e.g., duality between the notion of public good and the private nature of certain data) to make it possible to experiment in real situations and compare the performance of AI algorithms? Answering these questions would facilitate the collection and sharing of ad hoc data. AI, particularly when combining a participatory framework with expert systems and multi-agent systems, helps to build realistic representations of complex socio-biological systems. Thus, it proves to be an effective tool to promote the collaboration of different stakeholders in collective and optimised decision-making, and to assess of the impact of changes in uses and practices [ 30 ].

Encouraging experimentation of AI technologies at a territorial scale becomes crucial to favour their development, validate their performance, and measure their predictive quality. In AH, simplified access to data-generating facilities would allow innovative solutions to be tested on a larger scale and would accelerate their development and evaluation. Substantial expertise exists (e.g., epidemiological data platform, large cohorts, experimental farms) that could be put to good use. In addition, AI could help to revisit sampling methods for field data collection in AH and epidemiological surveillance, by better and more dynamically targeting the data to be collected while avoiding redundant collinear, non-necessary data.

3 Contribution of AI to better understand animal epidemiological systems

Recent technological advances involving AI approaches have made it possible to obtain vast quantities of measurements and observations, as well as to store and share these data more efficiently. This has resulted in an increasing requirement for appropriate data analytical methods. AI methods emerged as the response of the computer-science community to these requirements, leveraging the exponential improvements in computational power. In parallel, statistical methods have greatly evolved in the last few decades as well, e.g., with regards to dimensionality-reduction in the spaces of variables and parameters, variable selection, and model comparison and combination. The rise in computational power has unleashed the development of Bayesian inference through simulation or approximate methods [ 31 ]. Bayesian methods have, in turn, facilitated the integration of data from diverse sources, the incorporation of prior knowledge and allowed for inference on more complex and realistic models while changing the paradigm of statistical inference [ 32 , 33 , 34 ].

3.1 Better understanding the evolution of AH and socio-ecological systems in a One Health context

Learning methods can be used to do phylogenetic reconstructions, contributing in particular to new evolutionary scenarios of pathogens and their transmission pathways. For example, phylogenetic models offer an interesting perspective for identifying environmental bacterial strains with high infectious potentiality [ 35 ], or for predicting the existence of putative host reservoirs or vectors [ 36 ]. The analysis of pathogen sharing among hosts has been used to classify the potential reservoirs of zoonotic diseases using machine learning [ 37 ]. The analysis of pathogen genomes can also be used to identify genotypes of animal pathogens that are more likely to infect humans [ 38 ].

Using phenomenological niche models that rely on data distribution more than on hypotheses about ecological processes at play, disease occurrence data or retrospective serological data coupled with environmental variables can be related to the risk of being exposed to a pathogen. Thus, they can help monitor potential spillovers and emerging risks and anticipate the epidemic pathogen spread [ 39 ]. For instance, Artificial Neural Networks (ANN) have identified the level of genetic introgression between wild and domesticated animal populations in a spatialized context [ 40 ], which may help to understand gene diffusion in host × pathogen systems involving multiple host species, and characterise specimen pools at higher risk to act as pathogen spreaders or sinks. Other AI approaches such as multi-agent models, a more mechanistic approach, have been used in an explicit spatial context for vector-borne pathogen transmissions, and proved to be sufficiently versatile to be adapted to several other particular contexts [ 12 ].

It should be noted here that several studies reveal the relatively ancient nature of AI research in AH. Such AI methods have often made it possible to identify signals (e.g., genetic introgression) or even particular patterns or properties (e.g., importance of density-dependence in the vector-borne transmission) that are less visible or hardly detectable by more conventional statistical treatments.

All these approaches contribute to better understand pathogen transmission in complex system networks as generally observed for emerging infections in tropical, developing regions of the world. On this matter, an improved knowledge is key for protecting humans against these new threats, and AI/AH interfaces development and training in cooperation with the poorest countries would facilitate synergistic effects and actions to predict and tackle new disease threats.

3.2 Reliability, reproducibility and flexibility of mechanistic models in AH

Better understanding and predicting pathogen spread often requires an explicit and integrative representation of the mechanisms involved in the dynamics of AH systems, irrespective of the scale (within-host: [ 41 ]; along a primary production chain: [ 42 ]; in a territory: [ 43 , 44 ]; over a continent: [ 45 ]).

Mathematical (equations) or computer-based (simulations) models can be used. Such mechanistic models (i.e., which represent the mechanisms involved in the infection dynamics), when sufficiently modular to represent contrasted situations, make it possible to anticipate the effects of conventional but also innovative control measures (e.g., new candidate molecules, sensors, genomic selection; [ 46 ]).

However, to assess realistic control measures, mechanistic epidemiological models require the integration of observational data and knowledge from biology, epidemiology, evolution, ecology, agronomy, sociology or economics. Their development can rapidly face challenges of reliability, transparency, reproducibility, and usage flexibility. Moreover, these models are often developed de novo, making little use of previous models from other systems. Finally, these models, even based on realistic biological hypotheses, may be considered negatively as black boxes by end users (health managers), because the underlying assumptions often became hidden in the code or equations.

The integration of multiple modelling perspectives (e.g., disciplines, points of view, spatio-temporal scales) is an important question in the modelling-simulation field. Epidemiological modelling could benefit from existing tools and methods developed in this field [ 47 , 48 , 49 ]. Although essential, good programming practices alone [ 50 ] cannot meet these challenges [ 51 ]. Scientific libraries and platforms accelerate the implementation of the complex models often needed in AH. For example, the R library SimInf [ 52 ] helps integrate observational data into mechanistic models. The BROADWICK framework [ 53 ] provides reusable software components for several scales and modelling paradigms, but still requires modellers to write large amounts of computer code.

New methods at the crossroads between software engineering and AI can enhance transparency and reproducibility in mechanistic modelling, fostering communication between software scientists, modellers and AH researchers throughout the modelling process (e.g., assumption formulation, assessment, and revision). Knowledge representation methods from symbolic AI, formalised using advanced software engineering methods such as domain-specific languages (DSL, e.g., in KENDRICK for differential equation models: [ 54 ]), makes model components accessible in a readable structured text file instead of computer code. Hence, scientists from various disciplines and field managers can be more involved in the model design and evaluation. Scenario exploration and model revision also no longer require rewriting the model code.

Other AI methods can improve model flexibility and modularity. Autonomous software agents enable to represent various levels of abstraction and organisation [ 55 ], helping modellers go more easily back and forth within small and larger scales, and ensure that all relevant mechanisms are adequately formalised at proper scales (i.e., scale-dependency of determinants and drivers in hierarchical living systems). Combining knowledge representation (through a DSL) and such a multi-level agent-based simulation architecture (e.g., in EMULSION, Figure  2 , [ 56 ]) enables to encompass several types of models (e.g., compartmental, individual-based) and scales (e.g., individual, population, territory), and it tackles simultaneously the recurring needs for transparency, reliability and flexibility in modelling contagious diseases. This approach should also facilitate in the future the production of support decision tools for veterinary and public health managers and stakeholders.

AI at the service of mechanistic epidemiological modelling (adapted from [ 51 ] ) . A . Modellers develop each epidemiological model de novo, producing specific codes not easily readable by scientists from other disciplines or by model end-users. B . Using AI approaches to combine a domain-specific language and an agent-based software architecture enhances reproducibility, transparency, and flexibility of epidemiological models. A simulation engine reads text files describing the system to automatically produce the model code. Complementary add-ons can be added if required. Models are easier to transfer to animal health managers as decision support tools.

3.3 Extracting knowledge from massive data in basic AH biology

Supervised, unsupervised and semi-supervised learning methods facilitate basic research development in biology and biomedicine, for example by using morphological analyses to study cell mobility [ 57 ]. The use of classification approaches and smart filters allows nowadays to sort massive molecular data (e.g., data from high throughput sequencing and metagenomics). Metabolic, physiological and immunological signalling pathways are explored, and metabolites are identified and quantified in complex biological mixtures, which was before a major challenge [ 58 ]. In addition, diagnostic time may be reduced by developing image analysis processing (e.g., accelerated detection of clinical patterns; [ 59 , 60 ]), often necessary to study host–pathogen interactions in animal pathology. For example, the use of optimisation methods has improved the understanding of the fragmentation of prion assemblages, contributing to a significant reduction in the time required to diagnose neurodegenerative animal diseases, thus paving the way for identifying potential therapeutic targets [ 61 ]. In livestock breeding, there is a methodological transition underway from traditional prediction strategies to more advanced machine learning approaches including artificial neural networks, deep learning and Bayesian networks which are being used to improve the reliability of genetic predictions and further the understanding of phenotypes biology. [ 62 ].

In human health, new disciplines have emerged in the second half of the 20 th century at the interface between AI and flagship disciplines, such as cell biology and immunology. Interface disciplines have developed, e.g., computational biology and immunology, which today must spread to AH. Current human immunology is based on the description of fine molecular and cellular mechanisms (e.g., the number of known interleukins has increased considerably compared to the 1970s). The desire to understand the processes underlying immune responses has led to a revolution by inviting this discipline to focus on complex systems biology and AI-based approaches [ 63 ]. However, the imbalance between the numbers of immunologists and immunology modellers is hampering the fantastic growth of this new discipline.

As an additional level of complexity, the hierarchical nature of biological systems makes that at the individual level, animals including humans must be considered as holobionts made of myriads of hosted microbial forms that form discrete ecological units (i.e., infracommunities). The potential of AI to grasp such diversity and complexity (e.g., tissue-specific microbiotes) and to scaling-up to higher levels of organization (e.g., component and compound communities of microbes, including pathogens, circulating in herd and in a given region) is certainly tremendous and should be studied with the same vigour as recent development in computational biology and immunology [ 40 ].

4 Revisiting AH case detection methods at different scales

Managing livestock health issues requires effective case detection methods, at the individual or even infra-individual (organ) scale, at the group/herd scale, or at larger scales (e.g., territories, countries). Machine learning methods allow detecting patterns and signals in massive data, e.g., in spatial data or time-series of health syndromes and disease cases, contributing to the development of smart agriculture and telemedicine (Figure  3 ). Alerts can be produced, and contribute to management advice in numerical agriculture [ 64 ] and veterinary practices [ 65 ]. AI may contribute to an earlier detection of infected cases and the rationalisation of treatments (including antimicrobials) in farm animals, by analysing data collected from connected sensors [ 66 ], by targeting individuals or groups of animals [ 59 ], or even by using mechanistic models to predict the occurrence of case detections and their treatment [ 67 ]. Also, machine learning methods enable to discriminate pathogen strains and thus to better understand their respective transmission pathways if different [ 68 ]. Finally, therapeutic strategies can be reasoned through multi-criteria optimisation, by identifying whom to treat in a herd, when, according to what protocol and for how long, in order to maximise the probability of cure while minimising both the risk of drug resistance and the volume or number of doses that are necessary (i.e., individual-based and precision medicine).

Extracting information from massive data to monitor animal health and better rationalise treatments.

Nevertheless, alert quality depends on the quality and representativeness of the datasets used by the learning algorithms. Numerous biases (e.g., hardware, software, human) can affect prediction accuracy. Moreover, alerts produced after training necessarily reflect the specificities of the system from which the data originates (e.g., area, period, rearing practices). Thus, result transposition to other epidemiological systems or to the same system subjected to environmental or regulatory changes remains risky. Furthermore, while machine learning methods (e.g., classification, image analysis, pattern recognition, data mining) provide solutions for a wide range of biomedical and bio-health research questions, it is crucial to demonstrate the performance of these methods by measuring their predictive quality and comparing them to alternative statistical methods whenever possible [ 69 ].

At the population level, case detection is based on direct (detection of syndromes) or indirect surveillance, mobilising syndrome proxies. Hence, the emergence of some animal diseases can be detected by syndromic surveillance, by detecting abnormal or rare signals in routine data (e.g., mortality, reproduction, abortion, behaviour, milk production, increased drug use; [ 70 ]). Also, serological data can be used retrospectively to identify individual characteristics related to a risk of being exposed to a pathogen, and thus orientate management efforts (e.g., in wildlife; [ 71 ]). Statistics and AI are largely complementary to address such issues. Both mobilise the wide range of available data, which are highly heterogeneous, massive and mostly sparse, to detect signals that are often weak or scarce [ 28 , 72 , 73 ]. Such signals can be proxy records (e.g., emergence of infectious diseases following environmental disturbances), health symptoms and syndromes, or even metabolic pathways in cascades which can be precursors of chronic or degenerative diseases. AI also includes methods to mobilise information available on the web. For example, semi-automatic data mining methods enabled to identify emerging signals for international surveillance of epizooties [ 74 ] or to analyse veterinary documents such as necropsy reports [ 75 , 76 ]. Methods from the field of natural language processing can compensate the scarcity of data by extracting syntactic and semantic information from textual records, triggering alerts on new emerging threats that could have been missed otherwise.

On a large to very large scale (i.e., territory, country, continent, global), data analysis of commercial animal movements between farms makes it possible to predict the associated epidemic risk [ 77 , 78 ]. These movements are difficult to predict, particularly since animal trade is based on many factors associated with human activities and decisions. Methods for recognising spatio-temporal patterns and methodological developments for the analysis of oriented and weighted dynamic relational graphs are required in this field because very few of the existing methods allow large-scale systems to be studied, whereas datasets are often very large (e.g., several tens or even hundreds of thousands of interacting operations).

On this topic, the specific frontier between learning methods of AI and statistics is relatively blurred, lying most on the relative prominence of the computational performance of algorithms versus mathematics, probability and rigorous statistical inference. While machine learning methods are more empirical, focused on improving their predictive performance, statistics is more concerned with the quantification and modelling of uncertainties and errors [ 79 , 80 ]. In the last decade, both communities have started to communicate and to mix together. Methods have cross-fertilised, giving birth to statistical models using synthetic variables generated by AI methods, or AI algorithms optimising statistical measures of likelihood or quality. New research areas, such as Probabilistic Machine Learning, have emerged at the interface between the two domains [ 1 , 80 , 81 ]. Meanwhile, machine learning and statistics have kept their specific interests and complementarity; machine learning methods are especially well-suited to processing non-standard data types (e.g., images, sounds), while statistics can draw inference and model processes for which only few data are available, or where the quantities of interest are extreme events.

5 Targeted interventions, model of human decisions, and support of AH decisions

5.1 choosing among alternatives.

A challenge for animal health managers is to identify the most relevant combinations of control measures according to local (e.g., farm characteristics, production objectives) and territorial (e.g., available resources, farm location, management priorities) specificities. They have to anticipate the effects of health, environmental and regulatory changes, and deliver quality health advice. The question also arises of how to promote innovation in AH, such as to anticipate the required characteristics of candidate molecules in vaccine strategies or drug delivery [ 82 , 83 ], or to assess the competitive advantage of new strategies (e.g., genomic selection of resistant animals, new vaccines) over more conventional ones. Private (e.g., farmers, farmers’ advisors) and collective managers (e.g., farmer groups, public authorities) need support decision tools to better target public incentives, identify investments to be favoured by farmers [ 46 ] and target the measures as effectively as possible: who to target (which farms, which animals)?; with which appropriate measure(s)?; when and for how long? These questions become essential to reasoning about input usage (e.g., antimicrobials, pesticides, biocides) within the framework of the agro-ecological transition.

The use of mechanistic modelling is a solution to assess, compare and prioritise ex ante a wide range of options (Figure  4 ; [ 84 ]). However, most of the available models do not explicitly integrate human decision-making, while control decisions are often made by farmers (e.g., unregulated diseases), with sometimes large-scale health and decision-making consequences (e.g., pathogen spread, dissemination of information and rumours, area of influence). Recent work aims to integrate humans and their decisions by mobilising optimal control and adaptive strategies from AI [ 7 , 85 ] or health economics methods [ 86 , 87 ]. A challenge is to propose clear and context-adapted control policies [ 88 ]. Such research is just starting in AH [ 46 ] and must be extended as part of the development of agro-ecology, facing current societal demand for product quality and respect for ecosystems and their biodiversity on one side, animal well-being and ethics on the other side, and more generally international health security.

Identifying relevant strategies to control bovine paratuberculosis at a regional scale (adapted from [ 76 ] ) . Classically, identifying relevant strategies means defining them a priori and comparing them, e.g., by modelling. Only a small number of alternatives can be considered. If all alternatives are considered as in the figure, it results in a multitude of scenarios whose analysis becomes challenging. Here, each point corresponds to the epidemiological situation after 9 years of pathogen spread over a network of 12 500 dairy cattle herds for a given strategy (asterisk: no control). Initially, 10% of the animals are infected on average in 30% of the herds. The blue dots correspond to the most favourable strategies. Mobilizing AI approaches in such a framework, especially optimization under constraints, would facilitate the identification of relevant strategies by exploring the space of possibilities in a more targeted manner.

5.2 Accounting for expectations and fears of animal health managers

Animal health managers should have access to model predictions in a time frame compatible with management needs, which is problematic in the face of unpredictable emerging events (e.g., new epidemiological systems, transmission pathways, trade patterns, control measures). Developing a library of models included in a common framework would strengthen the responsiveness of modellers in animal epidemiology. Relevant models would be developed more quickly and would gain accuracy from real-time modelling as epidemics progress [ 89 , 90 ]. However, if this makes move more quickly from concepts (knowledge and assumptions) to simulations and support decision tools, a gain in performance is still required to perform analyses at a very large scale. The automatic generation of high-performance computer code could be a relevant solution, which however remains a crucial methodological lock to be addressed in AI. In addition, it is often required to perform a very large number of calculations or to analyse very large datasets, which call for a rational use of computing resources. Software transferred to health managers sometimes require the use of private cloud resources (i.e., it does not run on simple individual computers), highlighting the trade-offs between simulation cost, service continuity (e.g., failure management) and time required to obtain simulation results [ 91 ]. These questions are currently related to computer science research, and collaborations are desirable between these researchers and those from AH.

Managers also wish to rely on accurate predictions from realistic representations of the biological systems. Before being used, model behaviour should be analysed, which raises the questions of exploring the space of uncertainties and data, and of optimization under constraints. This often requires intensive simulations, which would benefit from optimization algorithms to explore more efficiently the space of possibilities. In turn, this would allow, for example, the automatic identification of how to achieve a targeted objective (e.g., reducing the prevalence of a disease below an acceptable threshold) while being constrained in resource allocation. While this issue finds solutions in modern statistics for relatively simple systems, it represents a science front for complex systems (e.g., large scale, multi-host/multi-pathogen systems) that are becoming the norm. In addition, optimization goals specific to AH may generate ad hoc methodological needs [ 92 ]. The needs in abstraction and analysis capacity are massive and could benefit from complementarities between AI (e.g., reasoned exploration, intelligent use of computer resources, optimized calculations) and statistics to extract as much information as possible from the data: (1) explore, analyse, predict; (2) infer processes and emergent properties. Methodological developments are still required and would benefit many health issues, particularly in relation to the currently evolving concepts of reservoir-host, edge-host and species barrier [ 93 ]. Furthermore, methodological developments and dissemination of existing methods should be reinforced.

Finally, three barriers have been identified to the development of support decision tools for health managers, related to the societal issue of the acceptability of AI sensu lato, as a major factor of progress. First, ethical issues, which are obvious when it comes to human health, are just as important to consider in AH. Which AI-based tools do we want for modern animal husbandries and trades, and for which objectives? Are these tools not likely to lead to discrimination against farms according to their health status, even when this status cannot be managed by the farmer alone? Second, in AH too, there is a fear that AI-tools may replace human expertise. However, automating does not mean replacing human, his expertise and decision [ 94 ], but rather supporting his capacities for abstraction and analysis, accelerating the global process, making predictions more reliable, guiding complementary research. Nevertheless, a significant development of computer resources and equipment is not without impacting the environment in terms of carbon footprint (e.g., energy-intensive servers, recycling of sensors), which must also be accounted for. Third, the very high complexity of analysing results and acculturating end-users with knowledge issued from academic research, particularly AI, is an obstacle to the appropriation of AI-tools by their users. This may lead to the preference for simpler and more easily accessible methods. However, the latter may not always be the most relevant or reliable. Citizen science projects, also known as community participation in human epidemiology, enable AH to co-design and co-construct the AI-tools of tomorrow with their end-users [ 95 ], to better meet their expectations and needs, and to increase their confidence in the predictions of sometimes obscure research models, especially when they are hard to read (e.g., lines of code). Similarly, these AI-tools could be developed together with public decision-makers, livestock farmers, agro-food industries and sectoral trade unions. Co-construction gives time to explain the science behind the tools and makes it more transparent and useful. This citizen participation, which is nowadays supported in many countries, guarantees decisions more in line with citizens’ expectations and corresponds to a general trend towards structured decision-making. AI must contribute to this democratisation of aid in public decision-making in AH.

6 Barriers to the development of research at the AI/AH interface

Research conducted at the interface between AI and AH requires strong interactions between biological disciplines (e.g., infectiology, immunology, clinical sciences, genetics, ecology, evolution, epidemiology, animal and veterinary sciences) and more theoretical disciplines (e.g., modelling, statistics, computer science), sometimes together with sociology and economics. Conducting research at this interface requires strengthening the few teams already positioned in Western Europe, but also bringing together teams working around the concepts of One Health, Ecohealth and Planetary Health to benefit from recent achievements in infectious disease ecology and modelling, plant health and environmental health [ 96 ]. This work must be based on a wide range of methodological skills (e.g., learning methods, data mining, information systems, knowledge representation, multi-agent systems, problem solving, metamodeling, optimisation, simulation architecture, model reduction, decision models). The need for research, training and support are crucial issues at national, European and international levels. Also, a facilitated and trusted connection is required between academics, technical institutes, and private partners, who are often the holders or collectors of data of interest to solve AH research questions through AI approaches. The construction of better inter-sectoral communication and coordination must be done at supra-institutional level, as this theme seems hyper-competitive and as some current divisions still go against information and data sharing.

An acculturation of researchers to AI, its methods and potential developments, but also its limitations, must be proposed to meet the challenges of 21 st century agriculture. Indeed, there are obstacles to conducting research on this scientific front. Establishing the new collaborations required between teams conducting methodological work and teams in the fields of application remains difficult given the low number of academic staff on these issues, their very high current mobilisation and their low availability to collaborate on new subjects, as well as the difficulty of understanding and mastering these methods. There is a need for watching and training on AI methods available or under development, new softwares/packages, and their applicability. To develop key collaborations and establish a strategic positioning, an interconnection can also be made via transversal teams which appears as a preferential path. Solutions must also be found to encourage method percolation in the community and the development of scientific and engineering skills.

Finally, AI methods, such as classification, machine learning, data mining, and the innovations in AH to which these methods can lead are rarely discussed in veterinary high school education, whereas these students represent the future professionals of AH [ 96 ]. Similarly, there is a quasi-absence of sentinel networks of veterinarians, even if it is developing, although AH questions can arise on a large and collective scale. The scientific community would also benefit from further increasing its skills and experience in the valuation, transfer and protection of intellectual property on these AI methods and associated outcomes.

7 Levers to create a fruitful AI/AH interface

7.1 data sharing and protection.

No innovation at the interface between AI and AH is possible without strong support for the organisation of data storage, management, analysis, calculation, and restitution. The major risk is that demands for AI developments inflate without being supported by available human resources. In addition, an expertise in law, jurisdiction and ethics is required with regard to the acquisition, holding, use and protection of data in AH. This question must be considered at least at the inter-institutional/national level, and could benefit from a similar thinking already engaged in human health. The issue is to be able to support any change with regard to data traceability to their ownership, whether being from public or private domains.

New data are rich and must be valued as much as possible, not by each owner separately, but through data sharing and the mobilisation of multi-disciplinary skills to analyse such heterogeneous and complex data. Hence, data interoperability skills are required and must be developed. Models for making federated data sustainable over decades are required [ 97 ]. In addition, further encouraging the publication of data papers as valuable research products can help to develop the necessary culture of sharing, documentation and metadata.

Finally, to be able to launch ambitious experiments with AI methods on real data, it is necessary to (1) remove unauthorised access to data by negotiating with owners at large scale; (2) analyse and understand the related effect on methodological developments; and (3) if necessary, extend such initiatives to other areas, at national scale, or even across European countries.

7.2 Attract the necessary skills

An undeniable barrier to conduct such research comes from human resources, in particular the current insufficient capacity of supervision by permanent scientists. Collaborations are a solution to attract new skills. However, initiating collaborations at the AH/AI interface becomes very complicated because the qualified teams are already overwhelmed. Skill development at this interface must be supported, the cross-fertilisation of disciplines being essential. A watch on methods must also be carried out, accompanied by explanations for application fields, to train researchers and engineers. Financial incentives for scientist internships in specialised laboratories would increase skill capitalisation in advanced methods, while facilitating future national or international collaborations. In a context of limited resources as observed in many countries nowadays (e.g., new opened positions in national institutions) and limited experts pool (e.g., skills), facilitating post-doctoral fellows and continuing education of researchers becomes crucial. Finally, to consolidate the pool of future researchers in AH, promoting basic AI education in initial training of AH researchers, engineers and veterinarians is paramount.

More specifically concerning current research in immunology, cell biology and infectiology, the contribution of AI has been more widely considered in human health, which could feed a similar reflection in AH as locks and advances are not very specific. Before embarking on the fronts of science (e.g., emerging epigenomics and metabolomics in AH), a few persons from these biological disciplines should acculturate into AI, or even acquire autonomy in the use of methods [ 98 ], which internationally tends to be the trend [ 63 ]. This can be done through the sharing of experiences and basic training on existing methods, their advantages and limitations compared to other methods coming from statistics and mathematical modelling.

7.3 Encourage the development of AH/AI projects

Projects at the AH/AI interface, like any interdisciplinary project, must mobilise teams from both groups of disciplines and allow everyone to progress in their own discipline. However, identifying the issues shared between the most relevant disciplines requires a good acculturation of the disciplines between them, as well as an otherness aimed at better understanding each other [ 95 ], which is not yet the case at the AH/AI interface.

In terms of funding, European project calls offer interesting opportunities, but a significant imbalance persists between the ability to generate data and analyse complex issues, and the availability of human resources and skills to address such issues through AI methods or other modern methods in statistics, mathematics and computer science. The major international foundations (e.g., Bill and Melinda Gates) can also be mobilised on emerging infectious diseases at the animal/human interface (e.g., characterisation of weak signals, phenologies, emergence precursors), with a more significant methodological value. However, risk-taking is rarely allowed by funding agencies, although it is crucial to initiate interdisciplinary work. Dedicated incentive funding would support projects in their initial phase and make larger projects emerge after consolidation of the necessary disciplinary interactions.

Finally, these projects are generally based on the use of significant computing resources. Thus, research institutes and private partners should contribute in a financial or material way to the shared development of digital infrastructures, data centres, supercomputing centres on a national scale, as well as support recognised open-source software platforms on which a large part of the research conducted is based (e.g., Python, R ).

7.4 Promoting innovation and public–private partnership

Encouraging public–private partnership would promote a leverage effect on public funding and would make it possible to place AI research and development on a long-term basis in AH. Mapping the highly changing landscape of companies in the AH/AI sector, whether international structures or start-ups, would provide a better understanding of the possible interactions. Similarly, mapping academic deliverables produced at this interface would increase their visibility and highlight their potential for valorisation or transfer. Finally, considering the production of documented algorithms as scientific deliverables, along with publications, would help support this more operational research. More broadly, it would be advisable to initiate a communication and education/acculturation policy around AI and its development in AH (e.g., links with the society, farmers, agricultural unions, public services).

8 Conclusion

The use of AI methods (e.g., machine learning, expert systems, analytical technologies) converges today with the collecting of massive and complex data, and allows these fields to develop rapidly. However, it is essential not to perceive massive data and AI as the same trend, because the accumulation of data does not always lead to an improvement in knowledge. Nevertheless, the more data are numerous and representative of working concepts and hypotheses, the more important results can be obtained from AI applications. The underlying ethical, deontological and legal aspects of data ownership, storage, management, sharing and interoperability also require that a reflection be undertaken nationally and internationally in AH to better manage these data of multi-sectoral origin and their various uses. Moreover, while the effort to acquire such data is impressive, the development of AI skills within the AH community remains limited in relation to the needs. Opportunities for collaborations with AI teams are limited because these teams are already in high demand. To ensure that AH researchers are well aware of the opportunities offered by AI, but also of the limits and constraints of AI approaches, a training effort must be provided and generalized. Finally, the current boom in AI now makes it possible to integrate the knowledge and points of view of the many players in the field of animal health and welfare further upstream. However, this requires that AI and its actors accept to deal with the specificity and complexity of AH, which is not a simple library of knowledge that can be digitised to search for sequences or informative signals.

Hosny A, Parmar C, Quackenbush J, Schwartz LH, Aerts HJWL (2018) Artificial intelligence in radiology. Nat Rev Cancer 18:500–510

Article   CAS   Google Scholar  

Karczewski KJ, Snyder MP (2018) Integrative omics for health and disease. Nat Rev Genet 19:299–310

Murphy KP (2012) Machine learning: a probabilistic perspective. In: Adaptive computation and machine learning series. MIT Press, USA

Zhang W, Chien J, Yong J, Kuang R (2017) Network-based machine learning and graph theory algorithms for precision oncology. NPJ Precis Oncol 1:25

Article   Google Scholar  

Saria S, Butte A, Sheikh A (2018) Better medicine through machine learning: what’s real, and what’s artificial? PLoS Med 15:e1002721. https://doi.org/10.1371/journal.pmed.1002721

Article   PubMed   PubMed Central   Google Scholar  

Bedi G, Carrillo F, Cecchi GA, Fernández Slezak D, Sigman M, Mota NB, Ribeiro S, Javitt DC, Copelli M, Corcoran CM (2015) Automated analysis of free speech predicts psychosis onset in high-risk youths. Schizophrenia 1:15030

Maclachlan MJ, Springborn MR, Fackler PL (2017) Learning about a moving target in resource management: optimal Bayesian disease control. Am J Agri Econ 99:140–162. https://doi.org/10.1093/ajae/aaw033

Lynn LA (2019) Artificial intelligence systems for complex decision-making in acute care medicine: a review. Patient Saf Surg 13:6. https://doi.org/10.1186/s13037-019-0188-2

Pinaire J, Azé J, Bringay S, Landais P (2017) Patient healthcare trajectory. An essential monitoring tool: a systematic review. Health Inf Sci Syst 5:1

Vrakas D, Vlahavas IPL (2008) Artificial intelligence for advanced problem solving techniques. Information Science Reference, Hershey, PA, pp. 369

Shakshuki E, Reid M (2015) Multi-agent system applications in healthcare: current technology and future roadmap. Proc Comput Sci 52:252–261. https://doi.org/10.1016/j.procs.2015.05.071

Roche B, Guégan JF, Bousquet F (2008) Multi-agent systems in epidemiology: a first step for computational biology in the study of vector-borne disease transmission. BMC Bioinform 9:435. https://doi.org/10.1186/1471-2105-9-435

Picault S, Huang Y-L, Sicard V, Ezanno P (2017) Enhancing Sustainability of Complex Epidemiological Models through a Generic Multilevel Agent-based Approach. In: Proceedings of the 26 th International Joint Conference on Artificial Intelligence (IJCAI). pp. 374–380, AAAI. https://doi.org/10.24963/ijcai.2017/53p

Russell S, Norvig P (2010) Artificial intelligence a modern approach. 3 rd edn. Upper Saddle River, New Jersey, pp. 1132

Google Scholar  

Ducrot C, Bed’Hom B, Béringue V, Coulon JB, Fourichon C, Guérin JL, Krebs S, Rainard P, Schwartz-Cornil I, Torny D, Vayssier-Taussat M, Zientara S, Zundel E, Pineau T (2011) Issues and special features of animal health research. Vet Res 42:96

Clark B, Stewart GB, Panzone LA, Kyriazakis I, Frewer LJ (2016) A systematic review of public attitudes, perceptions and behaviours towards production diseases associated with farm animal welfare. J Agric Environ Ethics 29:455–478. https://doi.org/10.1007/s10806-016-9615-x

Miguel E, Grosbois V, Caron A, Pople D, Roche B, Donnelly C (2020) A systemic approach to assess the potential and risks of wildlife culling for infectious disease control. Commun Biol 3:353. https://doi.org/10.1038/s42003-020-1032-z

Hur B, Hardefeldt LY, Verspoor K, Baldwin T, Gilkerson JR (2019) Using natural language processing and VetCompass to understand antimicrobial usage patterns in Australia. Aust Vet J 97:298–300. https://doi.org/10.1111/avj.12836

Article   CAS   PubMed   Google Scholar  

Behmann J, Hendriksen K, Mueller U, Buescher W, Pluemer L (2016) Support vector machine and duration-aware conditional random field for identification of spatio-temporal activity patterns by combined indoor positioning and heart rate sensors. Geoinformatica 20:693–714. https://doi.org/10.1007/s10707-016-0260-3

Suravajhala P, Kogelman LJA, Kadarmideen HN (2016) Multi-omic data integration and analysis using systems genomics approaches: methods and applications in animal production, health and welfare. Genet Sel Evol 48:38. https://doi.org/10.1186/s12711-016-0217-x

Article   CAS   PubMed   PubMed Central   Google Scholar  

Goldansaz SA, Guo AC, Sajed T, Steele MA, Plastow GS, Wishart DS (2017) Livestock metabolomics and the livestock metabolome: a systematic review. PLoS One 12:e0177675. https://doi.org/10.1371/journal.pone.0177675

Anvar SY, Tucker A, Vinciotti V, Venema A, van Ommen GJ, van der Maarel SM, Raz V, ’t Hoen PA (2011) Interspecies translation of disease networks increases robustness and predictive accuracy. PLoS Comput Biol 7:e1002258. https://doi.org/10.1371/annotation/fc0b4192-6427-4fb3-b347-c66651adf855

Morand S, Guégan J-F, Laurans Y (2020) From One Health to Ecohealth, mapping the incomplete integration of human, animal and environmental health. Iddri, Issue Brief No. 04/20

Ezenwa VO, Prieur-Richard A-H, Roche B, Bailly X, Becquart P, Garcia-Peña GE, Hosseini PR, Keesing F, Rizzoli A, Suzán GA, Vignuzzi M, Vittecoq M, Mills JN, Guégan J-F (2015) Interdisciplinarity and infectious diseases: an Ebola case study. PLoS Pathog 11:e1004992. https://doi.org/10.1371/journal.ppat.1004992

Van Boeckel TP, Takahashi S, Liao Q, Xing W, Lai S, Hsiao V, Liu F, Zheng Y, Chang Z, Yuan C, Metcalf CJE, Yu H, Grenfell BT (2016) Hand, foot, and mouth disease in China: critical community size and spatial vaccination strategies. Sci Rep 6:25248. https://doi.org/10.1038/srep25248

Holmstrom LK, Beckham TR (2017) Technologies for capturing and analysing animal health data in near real time. Rev Sci Tech 36:525–538

Neethirajan S (2017) Recent advances in wearable sensors for animal health management. Sens Biosensing Res 12:15–29

Perez AM, Zeng D, Tseng CJ, Chen H, Whedbee Z, Paton D, Thurmond MC (2009) A web-based system for near real-time surveillance and space-time cluster analysis of foot-and-mouth disease and other animal diseases. Prev Vet Med 91:39–45. https://doi.org/10.1016/j.prevetmed.2009.05.006

Article   PubMed   Google Scholar  

Wilkinson MD, Dumontier M, Aalbersberg IJ, Appleton G, Axton M, Baak A, Blomberg N, Boiten J-W, da Silva Santos LB, Bourne PE, Bouwman J, Brookes AJ, Clark T, Crosas M, Dillo I, Dumon O, Edmunds S, Evelo CT, Finkers R, Gonzalez-Beltran A, Gray AJG, Groth P, Goble C, Grethe JS, Heringa J, ’t Hoen PAC, Hooft R, Kuhn T, Kok R, Kok J, et al (2016) The FAIR guiding principles for scientific data management and stewardship. Sci Data 3:160018. https://doi.org/10.1038/sdata.2016.18

Binot A, Duboz R, Promburom P, Phimpraphai W, Cappelle J, Lajaunie C, Goutard FL, Pinyopummintr T, Figuié M, Roger FL (2015) A framework to promote collective action within the One Health community of practice: using participatory modelling to enable interdisciplinary, cross-sectoral and multi-level integration. One Health 1:44–48. https://doi.org/10.1016/j.onehlt.2015.09.001

Robert CP (2014) Bayesian computational tools. Annu Rev Stat Appl 1:153–177. https://doi.org/10.1146/annurev-statistics-022513-115543

Dunson DB (2001) Commentary: practical advantages of Bayesian analysis of epidemiologic data. Am J Epidemiol 153:1222–1226. https://doi.org/10.1093/aje/153.12.1222

Uusitalo L (2007) Advantages and challenges of Bayesian networks in environmental modelling. Ecol Model 203:312–318. https://doi.org/10.1016/j.ecolmodel.2006.11.033

Fokoué E (2019) On the ubiquity of the Bayesian paradigm in statistical machine learning and data science. Math Appl 8:189–209. https://doi.org/10.13164/ma.2019.12

Bailly X (2017) Hidden Markov phylogenetic models offer an interesting perspective to identify “high risk lineages” of environmental pathogens. Infect Genet Evol 55:45–47. https://doi.org/10.1016/j.meegid.2017.08.007

Babayan SA, Orton RJ, Streicker DG (2018) Predicting reservoir hosts and arthropod vectors from evolutionary signatures in RNA virus genomes. Science 362:577–580. https://doi.org/10.1126/science.aap9072

Wardeh M, Sharkey KJ, Baylis M (2020) Integration of shared-pathogen networks and machine learning reveals the key aspects of zoonoses and predicts mammalian reservoirs. Proc Biol Sci 287:20192882. https://doi.org/10.1098/rspb.2019.2882

Li J, Zhang S, Li B, Hu Y, Kang X-P, Wu X-Y, Huang M-T, Li Y-C, Zhao Z-P, Qin C-F, Jiang T (2020) Machine learning methods for predicting human-adaptive influenza A viruses based on viral nucleotide compositions. Mol Biol Evol 37:1224–1236. https://doi.org/10.1093/molbev/msz276

Peters DPC, McVey DS, Elias EH, Pelzel-McCluskey AM, Derner JD, Burruss ND, Schrader TS, Yao J, Pauszek SJ, Lombard J, Rodriguez LL (2020) Big data-model integration and AI for vector-borne disease prediction. Ecosphere 11:e03157. https://doi.org/10.1002/ecs2.3157

Lek S, Guégan J-F (2000) Artificial neuronal networks. In: Application to ecology and evolution. Springer, Berlin. https://doi.org/10.1016/j.it.2016.11.006

Go N, Touzeau S, Islam Z, Belloc C, Doeschl-Wilson A (2019) How to prevent viremia rebound? Evidence from a PRRSv data-supported model of immune response. BMC Syst Biol 13:15

Ferrer Savall J, Bidot C, Leblanc-Maridor M, Belloc C, Touzeau S (2016) Modelling Salmonella transmission among pigs from farm to slaughterhouse: interplay between management variability and epidemiological uncertainty. Intern J Food Microbiol 229:33–43. https://doi.org/10.1016/j.ijfoodmicro.2016.03.020

Widgren S, Engblom S, Bauer P, Frössling J, Emanuelson U, Lindberg A (2016) Data-driven network modelling of disease transmission using complete population movement data: spread of VTEC O157 in Swedish cattle. Vet Res 47:81

Qi L, Beaunée G, Arnoux S, Dutta BL, Joly A, Vergu E, Ezanno P (2019) Neighbourhood contacts and trade movements drive the regional spread of bovine viral diarrhoea virus (BVDV). Vet Res 50:30. https://doi.org/10.1186/s13567-019-0647-x

Buhnerkempe MG, Tildesley MJ, Lindström T, Grear DA, Portacci K, Miller RS, Lombard JE, Werkman M, Keeling MJ, Wennergren U, Webb CT (2014) The impact of movements and animal density on continental scale cattle disease outbreaks in the United States. PLoS One 9:e91724. https://doi.org/10.1371/journal.pone.0091724

Ezanno P, Andraud M, Beaunée G, Hoch T, Krebs S, Rault A, Touzeau S, Vergu E, Widgren S (2020) How mechanistic modelling supports decision 1 making for the control of enzootic infectious diseases. Epidemics 32:100398

Garira W (2018) A primer on multiscale modelling of infectious disease systems. Infect Dis Model 3:176–191. https://doi.org/10.1016/j.idm.2018.09.005

Traoré M, Zacharewicz G, Duboz R, Zeigler B (2018) Modeling and simulation framework for value-based healthcare systems. Simulation 95:481–497. https://doi.org/10.1177/0037549718776765

Childs LM, El Moustaid F, Gajewski Z, Kadelka S, Nikin-Beers R, Smith JW Jr, Walker M, Johnson LR (2019) Multi-scale models and data for infectious diseases: a systematic review. PeerJ Preprints 7:e27485v1. https://doi.org/10.7287/peerj.preprints.27485v1

Sandve GK, Nekrutenko A, Taylor J, Hovig E (2013) Ten simple rules for reproducible computational research. PLoS Comput Biol 9:e1003285. https://doi.org/10.1371/journal.pcbi.1003285

Leek JT, Peng RD (2015) Opinion: reproducible research can still be wrong: adopting a prevention approach. Proc Natl Acad Sci USA 112:1645–1646. https://doi.org/10.1073/pnas.1421412111

Widgren S, Bauer P, Eriksson R, Engblom S (2016) SimInf: an R package for data-driven stochastic disease spread simulations. ArXiv160501421 Q-Bio Stat. http://arxiv.org/abs/1605.01421

O’Hare A, Lycett SJ, Doherty TM, Salvador LC, Kao RR (2016) Broadwick: a framework for computational epidemiology. BMC Bioinform 17:65. https://doi.org/10.1186/s12859-016-0903-2

Bui TMA, Stinckwich S, Ziane M, Roche B, Ho TV (2015) KENDRICK: a domain specific language and platform for mathematical epidemiological modelling. In: proc. IEEE RIVF International Conference on Computing and Communication Technologies, Research, Innovation, and Vision for the Future. pp. 132–7. https://doi.org/10.1109/RIVF.2015.7049888

Mathieu P, Morvan G, Picault S (2018) Multi-level agent-based simulations: four design patterns. Simul Model Pract Theory 83:51–64. https://doi.org/10.1016/j.simpat.2017.12.015

Picault S, Huang Y-L, Sicard V, Arnoux S, Beaunée G, Ezanno P (2019) EMULSION: transparent and flexible multiscale stochastic models in human, animal and plant epidemiology. PLoS Comput Biol 15:e1007342. https://doi.org/10.1371/journal.pcbi.1007342

Sebag AS, Plancade S, Raulet-Tomkiewicz C, Barouki R, Vert J-P, Walter T (2015) Inferring an ontology of single cell motions from high-throughput microscopy data. In: Proc. IEEE International Symposium on Biomedical Imaging, Apr. 2015, New-York, USA, pp. 160–163. https://doi.org/10.1109/ISBI.2015.7163840

Tardivel P, Canlet C, Lefort G, Tremblay-Franco M, Debrauwer L, Concordet D, Servien R (2017) ASICS: an automatic method for identification and quantification of metabolites in complex 1D 1H NMR spectra. Metabolomics 13:109

Dórea FC, Muckle CA, Kelton D, McClure JT, McEwen BJ, McNab WB, Sanchez J, Revie CW (2013) Exploratory analysis of methods for automated classification of laboratory test orders into syndromic groups in veterinary medicine. PLoS One 8:e57334. https://doi.org/10.1371/journal.pone.0057334

Gandia P, Jaudet C, Chatelut E, Concordet D (2017) Population pharmacokinetics of tracers: a new tool for medical imaging? Clin Pharmacokinet 56:101–106

Chyba M, Coron J-M, Mileyko Y, Rezaei H (2016) Optimization of prion assemblies fragmentation. In: Proc. IEEE Conference on Decision and Control (CDC), Las Vegas, USA, 6

Nayeri S, Sargolzaei M, Tulpan D (2019) A review of traditional and machine learning methods applied to animal breeding. Anim Health Res Rev 20:31–46. https://doi.org/10.1017/S1466252319000148

Bassaganya-Riera J, Hontecillas R (2016) Introduction to computational immunology. In: Bassaganya-Riera J (ed) Computational immunology: models and tools. pp. 1–8

Liakos KG, Busato P, Moshou D, Pearson S, Bochtis D (2018) Machine learning in agriculture: a review. Sensors 18:2674. https://doi.org/10.3390/s18082674

Jones-Diette JS, Dean RS, Cobb M, Brennan ML (2019) Validation of text-mining and content analysis techniques using data collected from veterinary practice management software systems in the UK. Prev Vet Med 167:61–67. https://doi.org/10.1016/j.prevetmed.2019.02.015

Morota G, Ventura RV, Silva FF, Koyama M, Fernando SC (2018) Big data analytics and precision animal agriculture symposium: machine learning and data mining advance predictive big data analysis in precision animal agriculture. J Anim Sci 96:1540–1550. https://doi.org/10.1093/jas/sky014

Picault S, Ezanno P, Assié S (2019) Combining early hyperthermia detection with metaphylaxis for reducing antibiotics usage in newly received beef bulls at fattening operations: a simulation-based approach. In: Society of veterinary epidemiology and preventive medicine (SVEPM), pp. 13. Utrecht, The Netherland, 27-30/3/2019

Esener N, Green MJ, Emes RD, Jowett B, Davies PL, Bradley AJ, Dottorini T (2018) Discrimination of contagious and environmental strains of Streptococcus uberis in dairy herds by means of mass spectrometry and machine-learning. Sci Rep 8:17517. https://doi.org/10.1038/s41598-018-35867-6

Hepworth PJ, Nefedov AV, Muchnik IB, Morgan KL (2012) Broiler chickens can benefit from machine learning: support vector machine analysis of observational epidemiological data. J R Soc Interface 9:1934–1942. https://doi.org/10.1098/rsif.2011.0852

Marceau A, Madouasse A, Lehébel A, van Schaik G, Veldhuis A, Van der Stede Y, Fourichon C (2014) Can routinely recorded reproductive events be used as indicators of disease emergence in dairy cattle? An evaluation of 5 indicators during the emergence of bluetongue virus in France in 2007 and 2008. J Dairy Sci 97:6135–6150. https://doi.org/10.3168/jds.2013-7346

Fountain-Jones NM, Machado G, Carver S, Packer C, Recamonde-Mendoza M, Craft ME (2019) How to make more from exposure data? An integrated machine learning pipeline to predict pathogen exposure. J Anim Ecol 88:1447–1461. https://doi.org/10.1111/1365-2656.13076

Charras-Garrido M, Azizi L, Forbes F, Doyle S, Peyrard N, Abrial D (2013) On the difficulty to delimit disease risk hot spots. Int J Appl Earth Obs 22:99–105. https://doi.org/10.1016/j.jag.2012.04.005

Forbes F, Charras-Garrido M, Azizi L, Doyle S, Abrial D (2013) Spatial risk mapping for rare disease with hidden Markov fields and variational EM. Annals Appl Stat 7:1192–1216

Arsevska E, Valentin S, Rabatel J, de Goër de Hervé J, Falala S, Lancelot R, Roche M (2018) Web monitoring of emerging animal infectious diseases integrated in the French Animal Health Epidemic Intelligence System. PLoS One 13:0199960. https://doi.org/10.1371/journal.pone.0199960

Küker S, Faverjon C, Furrer L, Berezowski J, Posthaus H, Rinaldi F, Vial F (2018) The value of necropsy reports for animal health surveillance. BMC Vet Res 14:191. https://doi.org/10.1186/s12917-018-1505-1

Bollig N, Clarke L, Elsmo E, Craven M (2020) Machine learning for syndromic surveillance using veterinary necropsy reports. PLoS One 15:e0228105. https://doi.org/10.1371/journal.pone.0228105

Hoscheit P, Geeraert S, Beaunée G, Monod H, Gilligan CAG, Filipe J, Vergu E, Moslonka-Lefebvre M (2016) Dynamical network models for cattle trade: towards economy-based epidemic risk assessment. J Complex Netw 5:604–624. https://doi.org/10.1093/comnet/cnw026

Moslonka-Lefebvre M, Gilligan CA, Monod H, Belloc C, Ezanno P, Filipe JAN, Vergu E (2016) Market analyses of livestock trade networks to inform the prevention of joint economic and epidemiological risks. J R Soc Interface 13:20151099. https://doi.org/10.1098/rsif.2015.1099

Efron B (2020) Prediction, estimation, and attribution. J Am Stat Ass 115:636–655. https://doi.org/10.1080/01621459.2020.1762613

Ghahramani Z (2012) Probabilistic modelling, machine learning, and the information revolution. MIT Computer Science and Artificial Intelligence Lab, http://mlg.eng.cam.ac.uk/zoubin/talks/mit12csail.pdf , Accessed 17 Oct 2019

Hastie T, Tibshirani R, Friedman JH (2009) The elements of statistical learning: data mining, inference, and prediction. 2 nd edn. Springer Series in Statistics. Springer

Goodswen SJ, Kennedy PJ, Ellis JT (2017) On the application of reverse vaccinology to parasitic diseases: a perspective on feature selection and ranking of vaccine candidates. Int J Parasitol 47:779–790. https://doi.org/10.1016/j.ijpara.2017.08.004

Schneider G (2019) Mind and machine in drug design. Nat Mach Intell 1:128–130. https://doi.org/10.1038/s42256-019-0030-7

Beaunée G, Vergu E, Joly A, Ezanno P (2017) Controlling bovine paratuberculosis at a regional scale: towards a decision modeling tool. J Theor Biol 435:157–183. https://doi.org/10.1016/j.jtbi.2017.09.012

Viet A-F, Krebs S, Rat-Aspert O, Jeanpierre L, Belloc C, Ezanno P (2018) A modelling framework based on MDP to coordinate farmers’ disease control decisions at a regional scale. PLoS One 13:e0197612. https://doi.org/10.1371/journal.pone.0197612

Wang T, Hennessy DA (2015) Strategic interactions among private and public efforts when preventing and stamping out a highly infectious animal disease. Am J Agri Econ 97:435–451. https://doi.org/10.1093/ajae/aau119

Tago D, Hammitt JK, Thomas A, Raboisson D (2016) The impact of farmers’ strategic behavior on the spread of animal infectious diseases. PLoS One 11:e0157450. https://doi.org/10.1371/journal.pone.0157450

Probert WJM, Lakkur S, Fonnesbeck CJ, Shea K, Runge MC, Tildesley MJ, Ferrari MJ (2019) Context matters: using reinforcement learning to develop human-readable, state-dependent outbreak response policies. Phil Trans R Soc B 374:20180277. https://doi.org/10.1098/rstb.2018.0277

Liang R, Lu Y, Qu X, Su Q, Li C, Xia S, Liu Y, Zhang Q, Cao X, Chen Q, Niu B (2020) Prediction for global African swine fever outbreaks based on a combination of random forest algorithms and meteorological data. Transbound Emerg Dis 67:935–946. https://doi.org/10.1111/tbed.13424

Salje H, Tran Kiem C, Lefrancq N, Courtejoie N, Bosetti P, Paireau J, Andronico A, Hozé N, Richet J, Dubost C-L, Le Strat Y, Lessler J, Levy Bruhl D, Fontanet A, Opatowski L, Boelle P-Y, Cauchemez S (2020) Estimating the burden of SARS-CoV-2 in France. Science 369:208–211

Parlavantzas N, Pham LM, Morin C, Arnoux S, Beaunée G, Qi L, Gontier P, Ezanno P (2019) A service-based framework for building and executing epidemic simulation applications in the cloud. Concurr Comp Pract Exper 32:e5554. https://doi.org/10.1002/cpe.5554

Shah N, Malensek M, Shah H, Pallickara S, Pallickara SL (2019) Scalable network analytics for characterization of outbreak influence in voluminous epidemiology datasets. Concurr Comp Pract Exper 31:e4998. https://doi.org/10.1002/cpe.4998

Han BA, Majumdar S, Calmon FP, Glicksberg BS, Horesh R, Kumar A, Perer A, von Marschall EB, Wei D, Mojsilović A, Varshney KR (2019) Confronting data sparsity to identify potential sources of Zika virus spillover infection among primates. Epidemics 27:59–65. https://doi.org/10.1016/j.epidem.2019.01.005

Reddy S, Fox J, Purohit MP (2019) Artificial intelligence-enabled healthcare delivery. J R Soc Med 112:22–28

Duboz R, Echaubard P, Promburom P, Kilvington M, Ross H, Allen W, Ward J, Deffuant G, de Garine-Wichatitsky M, Binot A (2018) Systems thinking in practice: participatory modelling as a foundation for integrated approaches to health. Front Vet Sci 5:303. https://doi.org/10.3389/fvets.2018.00303

Van der Waal K, Morrison RB, Neuhauser C, Vilalta C, Perez AM (2017) Translating big data into smart data for veterinary epidemiology. Front Vet Sci 4:110. https://doi.org/10.3389/fvets.2017.00110

Reichman OJ, Jones MB, Schildhauer MP (2011) Challenges and opportunities of open data in ecology. Science 331:703–705. https://doi.org/10.1126/science.1197962

Schultze JL (2015) Teaching ‘big data’ analysis to young immunologists. Nat Immunol 16:902–905

Download references

Acknowledgements

This work has benefited from interactions with many French researchers (Additional file 1 B) interested in the AI/AH interface, for which we thank them here. We also thank Stéphane Abrioux, Didier Concordet and Human Rezaie for participating to the discussions.

PE is supported by the French Research Agency (project CADENCE: ANR-16-CE32-0007). XB is involved in the project “MOnitoring Outbreak events for Disease surveillance in a data science context” supported by the EU Framework Programme for Research and Innovation H2020 (H2020-SC1-BHC-2018–2019, Grant 874850). JFG is supported by both an “Investissement d’Avenir” managed by the French Research Agency (LABEX CEBA: ANR-10-LABX-25-01) and a US NSF-NIH Ecology of infectious diseases award (NSF#1911457), and is also supported by IRD, INRAE, and Université of Montpellier. The funding bodies had no role in the study design, data analysis and interpretation, and manuscript writing.

Author information

Authors and affiliations.

INRAE, Oniris, BIOEPAR, Nantes, France

Pauline Ezanno, Sébastien Picault & Gaël Beaunée

INRAE, EpiA, Theix, France

Xavier Bailly

ASTRE, Univ Montpellier, CIRAD, INRAE, Montpellier, France

Facundo Muñoz, Raphaël Duboz & Jean-François Guégan

Sorbonne Université, IRD, UMMISCO, Bondy, France

Raphaël Duboz

Université Paris-Saclay, INRAE, Jouy-en-Josas, MaIAGE, France

Hervé Monod

MIVEGEC, IRD, CNRS, Univ Montpellier, Montpellier, France

Jean-François Guégan

Comité National Français Sur Les Changements Globaux, Paris, France

You can also search for this author in PubMed   Google Scholar

Contributions

PE carried out the literature review and analysed the interviews. PE and JFG conducted the interviews, drafted and wrote the manuscript. SP, GB, FM, RD, HM provided complementary views and references in their respective disciplines. All authors read and approved the final manuscript.

Corresponding author

Correspondence to Pauline Ezanno .

Ethics declarations

Competing interests.

The authors declare that they have no competing interests.

Additional information

Publisher's note.

Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.

Supplementary Information

Additional file 1..

Systematic literature review, interviews, previous publication in French.

Rights and permissions

Open Access This article is licensed under a Creative Commons Attribution 4.0 International License, which permits use, sharing, adaptation, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons licence, and indicate if changes were made. The images or other third party material in this article are included in the article's Creative Commons licence, unless indicated otherwise in a credit line to the material. If material is not included in the article's Creative Commons licence and your intended use is not permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder. To view a copy of this licence, visit http://creativecommons.org/licenses/by/4.0/ . The Creative Commons Public Domain Dedication waiver ( http://creativecommons.org/publicdomain/zero/1.0/ ) applies to the data made available in this article, unless otherwise stated in a credit line to the data.

Reprints and permissions

About this article

Cite this article.

Ezanno, P., Picault, S., Beaunée, G. et al. Research perspectives on animal health in the era of artificial intelligence. Vet Res 52 , 40 (2021). https://doi.org/10.1186/s13567-021-00902-4

Download citation

Received : 16 June 2020

Accepted : 20 January 2021

Published : 06 March 2021

DOI : https://doi.org/10.1186/s13567-021-00902-4

Share this article

Anyone you share the following link with will be able to read this content:

Sorry, a shareable link is not currently available for this article.

Provided by the Springer Nature SharedIt content-sharing initiative

• Animal disease
• Artificial intelligence
• Decision support tool

Veterinary Research

ISSN: 1297-9716

• Submission enquiries: [email protected]
• General enquiries: [email protected]

• +1 408-816-2779
• [email protected]

About the Journal

Aims and scope.

Journal of Animal Science and Research (JASR) recognises that to properly study and manage the animals that are useful to humans and those in the wild we need to understand the science behind them. That implies learning Biology, Reproduction, Physiology, Nutrition (feeds, feeding), Pathology (including disease emergence and zoonoses) and behaviour. Moreover, JASR wishes to contribute to bridging the gap between scientific research and the application of sound science in the management and conservation of domestic mammals, non-domestic mammals, avian, reptiles, and fish and aquatic species as well as other wild species.

The journal will focus on studies that involve carefully selected reports that provide a link between basic research and application in animal management. The editors support the dissemination of original solutions to stimulating questions in animal science and animal management. JASR is dedicated to publishing material of research in animal science, such as Biology of Reproduction, Gametes and Embryo Conservation, Immunology, Disease and Disease Prevention and Treatment, Animal Welfare, well-being and stress, Breeding, Genetics, and Genomics and Sustainability of animal production. Such papers will be interesting to an international audience of professionals in academia, breeders, veterinarians, and biologists. Presumably, the readership of JASR will be engaged in the management, use, and protection of animals in the wild, on the farm or as pets.

The journal pursues contributions which prioritise distinction, originality, reduction to practice, accuracy, focus and appeal to the readership. Most likely veterinarians, agronomists, and various scientists devoted to animal fields in the broadest sense will engage as authors of JASR articles.

Call for submissions

JASR publishes articles of original, interdisciplinary, and applied research. The Editor actively encourages scientists and professionals dealing with research in animal management for production, veterinary care in research or conservation programs, basic scientists engaged in animal physiology, endocrinology and systems biologists. Emphasis is given to papers that heighten capabilities for sustainable livestock production with environmental and animal welfare focus. Reports directed to and based on research on a comprehensive range of subjects as well as experimental papers may be submitted. Moreover, authors may contribute short reviews (by invitation from the Editor) and letters to the editor.

Fields covered and integrated by The Journal of Animal Science and Research (JASR)

• Animal Reproduction and Assisted Technologies
• Animal Immunology and Immunopathology
• Veterinary Anesthesiology
• Veterinary Medicines
• Veterinary Microbiology
• Veterinary medicinal care and practices
• Animal Pathology and parasitology
• Animal Genetics and Breeding
• Veterinary Surgery and Radiology
• Anatomy and Histology
• Veterinary Physiology and Biochemistry
• Animal endocrinology and metabolism
• Veterinary Gynecology and Obstetrics
• Animal Metabolic systems and Nutrition
• Animal Dermatology
• Animal Health
• Meat science
• Animal Vaccines and Immunization
• Animal Viruses
• Animal Behaviour
• Animal Welfare
• Equine and Bovine Research
• Animal Ethics and Legal Issues
• Animal-Husbandry
• Small animal Critical Care
• Non-Mammalian Medicine
• Animals and Entomology
• Animals in Clinical Trials
• Veterinary nursing
• Veterinary education
• Veterinary epidemiology
• Veterinary parasitology
• Veterinary public health
• Epizooties, Zoonosis
• Dairy Diseases
• Dairy Products
• Livestock Production and management
• Poultry science
• Aquaculture
• Fisheries science and fish nutrition
• Zoo animal management
• Zoo animal clinical studies
• Wild animal disease
• Wildlife Studies
• Herpetology
• Avian Science

Journal Indexing

SCI FORSCHEN JOURNALS

All Sci Forschen Journals are Open Access

New Journals

• Company Information |
• Open Access |
• Peer Review |
• Submit Manuscript |

• Research article
• Open access
• Published: 05 February 2021

Increasing adoption rates at animal shelters: a two-phase approach to predict length of stay and optimal shelter allocation

• Janae Bradley 1 &
• Suchithra Rajendran   ORCID: orcid.org/0000-0002-0817-6292 2 , 3  

BMC Veterinary Research volume  17 , Article number:  70 ( 2021 ) Cite this article

39k Accesses

18 Citations

74 Altmetric

Metrics details

Among the 6–8 million animals that enter the rescue shelters every year, nearly 3–4 million (i.e., 50% of the incoming animals) are euthanized, and 10–25% of them are put to death specifically because of shelter overcrowding each year. The overall goal of this study is to increase the adoption rates at animal shelters. This involves predicting the length of stay of each animal at shelters considering key features such as animal type (dog, cat, etc.), age, gender, breed, animal size, and shelter location.

Logistic regression, artificial neural network, gradient boosting, and the random forest algorithms were used to develop models to predict the length of stay. The performance of these models was determined using three performance metrics: precision, recall, and F1 score. The results demonstrated that the gradient boosting algorithm performed the best overall, with the highest precision, recall, and F1 score. Upon further observation of the results, it was found that age for dogs (puppy, super senior), multicolor, and large and small size were important predictor variables.

The findings from this study can be utilized to predict and minimize the animal length of stay in a shelter and euthanization. Future studies involve determining which shelter location will most likely lead to the adoption of that animal. The proposed two-phased tool can be used by rescue shelters to achieve the best compromise solution by making a tradeoff between the adoption speed and relocation cost.

As the problem of overpopulation of domestic animals continues to rise, animal shelters across the nation are faced with the challenge of finding solutions to increase the adoption rates. In the United States, about 6–8 million dogs and cats enter animal shelters every year, and 3–4 million of those animals are euthanized [ 1 ]. In other words, about 50% of the total canines and felines that enter animal shelters are put to death annually. Moreover, 10–25% of the total euthanized population in the United States is explicitly euthanized because of shelter overcrowding each year [ 2 ]. Though animal shelters provide incentives such as reduced adoption fees and sterilizing animals before adoption, only a quarter of total animals living in the shelter are adopted.

Animal adoption from shelters and rescues

There are various places to adopt an animal, and each potential owner must complete the adoption process and paperwork to take their new animal home [ 3 ]. Public and private animal shelters include animal control, city and county animal shelters, and police and health departments. Staff and volunteers run these facilities. Animals may also be adopted from a rescue organization, where pets are fostered in a home or a private boarding facility. These organizations are usually run by volunteers, and animals are viewed during local adoption events that are held at different locations, such as a pet store [ 3 ].

There could be several reasons for the euthanization of animals in a shelter, such as overcrowding, medical issues (ex. sick, disabled), or behavioral issues (ex. too aggressive). The causes for the overpopulation of animals include failure to spay or neuter animals leading to reckless breeding habits and abandonment or surrender of offspring, animal abandonment from owners who are no longer able to take care of or do not want the animal, and individuals still buying from pet stores [ 4 ]. With the finite room capacity for animals that are abandoned or surrendered, overpopulation becomes a key challenge [ 5 ]. Though medical and behavioral issues are harder to solve, the overpopulation of healthy adoptable animals in shelters is a problem that can be addressed through machine learning and predictive analytics.

Literature review

In this section, we describe the research conducted on animal shelters evaluating euthanasia and factors associated with animal adoption. The articles provide insights into factors that influence the length of stay and what characteristics influence adoption.

Studies have been conducted investigating the positive influence of pre-adoption neutering of animals on the probability of pet adoption [ 2 ]. The author investigated the impact of the cooperation of veterinary medical schools in increasing pet adoption by offering free sterilization. Results demonstrated that the collaboration between veterinary hospitals and local animal shelters decreased the euthanization of adoptable pets.

Hennessy et al. [ 6 ] conducted a study to determine the relationship between the behavior and cortisol levels of dogs in animal shelters and examined its effect in predicting behavioral issues after adoption. Shore et al. [ 7 ] analyzed the reasons for returning adopted animals by owners and obtained insights for these failed adoptions to attain more successful future approvals. The researchers found that prior failed adoption had led to longer-lasting future acceptances. They hypothesized that the failed adoptions might lead owners to discover their dog preferences by assessing their living situation and the type of animal that would meet that requirement.

Morris et al. [ 8 ] evaluated the trends in income and outcome data for shelters from 1989 to 2010 in a large U.S. metropolitan area. The results showed a decrease in euthanasia, adoption, and intake for dogs. For cats, a reduction in intake was observed until 1998, a decrease in euthanasia was observed until 2000, and the adoption of cats remained the same. Fantuzzi et al. [ 9 ] explored the factors that are significant for the adoption of cats in the animal shelter. The study investigated the effects of toy allocation, cage location, and cat characteristics (such as age, gender, color, and activity level). Results demonstrated that the more active cats that possessed toys and were viewed at eye level were more likely to impress the potential adopter and be adopted. Brown et al. [ 10 ] conducted a study evaluating the influence of age, breed, color, and coat pattern on the length of stay for cats in a no-kill shelter. The authors concluded that while color did not influence the length of stay for kittens, whereas gender, coat patterning, and breed were significant predictors for both cats and kittens.

Machine learning

Machine learning is one possible tool that can be used to identify risk factors for animal adoption and predict the length of stay for animals in shelters. Machine learning is the ability to program computers to learn and improve all by itself using training experience [ 11 ]. The goal of machine learning is to develop a system to analyze big data, quickly deliver accurate and repeatable results, and to adapt to new data independently. A system can be trained to make accurate predictions by learning from examples of desired input-output data. More specifically, machine learning algorithms are utilized to detect classification and prediction patterns from large data and to develop models to predict future outcomes [ 12 ]. These patterns show the relationship between the attribute variables (input) and target variables (output) [ 13 ].

Widely used data mining tasks include supervised learning, unsupervised learning, and reinforcement learning [ 14 ]. Unsupervised learning involves the use of unlabeled datasets to train a system for finding hidden patterns within the data [ 15 ]. Clustering is an example of unsupervised learning. Reinforcement learning is where a system is trained through direct interaction with the environment by trial and error [ 15 ]. Supervised learning encompasses classification and prediction using labeled datasets [ 15 ]. These classification and regression algorithms are used to classify the output variable with a discrete label or predict the outcome as a continuous or numerical value. Traditional algorithms such as neural networks, decision trees, and logistic regression typically use supervised learning. Figure  1 provides a pictorial of the steps for developing and testing a predictive model.

Pictorial Representation of Developing a Predictive Model

Contributions to the literature

Although prior studies have investigated the impact of several factors, such as age and gender, on the length of stay, they focus on a single shelter, rather than multiple organizations, as in this study. The goal of this study is to investigate the length of stay of animals at shelters and the factors influencing the rate of animal adoption. The overall goal is to increase adoption rates of pets in animal shelters by utilizing several factors to predict the length of stay. Machine learning algorithms are used to predict the length of stay of each animal based on numerous factors (such as breed, size, and color). We address several objectives in this study that are listed below.

Identify risk factors associated with adoption rate and length of stay

Utilize the identified risk factors from collected data to develop predictive models

Compare statistical models to determine the best model for length of stay prediction

Exploratory Data results

From Fig.  2 , it is evident that the return of dogs is the highest outcome type at 43.3%, while Fig.  3 shows that the adoption of cats is the highest outcome type at 46.1%. Both figures illustrate that the euthanization of both cats and dogs is still prevalent (~ 20%). The results from Table 1 demonstrate that the longest time spent in the shelter is at 355 days by a male cat that is adopted and a female dog that is euthanized. Observing the results, adoption has the lowest variance among all animal types compared to the other outcome types. Adopted male cats have the lowest variance for days spent in the shelter, followed by female dogs. Female cats that are returned have the highest variance for days spent in the shelter.

Distribution of Outcome Types for Dogs

Distribution of Outcome Types for Cats

Figure  4 shows a comparison of cats and dogs for the three different outcome types. It is observed from the data that there are more dogs returned than cats. From Fig.  5 , it is observed that the number of days a dog stays in the shelter decreases as the age increases. This is not expected, as it is predicted that the number of days in a shelter would be lower for younger dogs and puppies. This observation could be due to having more data points for younger dogs.

Comparison of Outcome Types for Cats and Dogs

Age vs. Days in Shelter for Cats and Dogs

Machine learning results

Examining Table 2 , it is clear that the most proficient predictive model is developed by the gradient boosting algorithm for this dataset, followed by the random forest algorithm. The logistic regression algorithm appears to perform the worst with low precision, recall, and F1 score performance metrics for all categories of length of stay. For the prediction of low length of stay in a shelter, the random forest algorithm is the best performing model in comparison to the others at around 64–70% performance for precision, recall, and F1 score. The ANN algorithm is found to be the best when evaluating the precision and F1 score for medium length of stay, while the random forest algorithm is better for assessing recall. However, the performance of these models in predicting the medium length of stay for the given dataset is low for all three-performance metrics. The gradient boosting algorithm performs the best when predicting the high length of stay. Finally, the gradient boosting and random forest algorithms perform well when predicting the very high length of stay at around 70–80%.

Results from Table 2 also demonstrate that the model developed from the gradient boosting algorithm has a higher performance when predicting the high length of stay that leads to adoption, and when the outcome is euthanization. Evaluating the average of all three-performance metrics for all algorithms, the gradient boosting is the most proficient model at almost 60%, while logistic regression appears to be the worst. Table 2 also provides the computational time for each machine learning algorithm. For the given dataset, logistic regression runs the fastest at 9.41 s, followed by gradient boosting, artificial neural network, and finally, random forest running the longest. The gap in the performance measure ( pm ) is calculated by \( \frac{p{m}_{best}-p{m}_{worst}}{p{m}_{best}} \) , and is nearly 34, 39, and 32% for precision, recall, and F1 score, respectively.

Table 3 provides information on the top features or factors from each machine learning algorithm. Observing the table, we find that age (senior, super senior, and puppy), size (large and small), and color (multicolor) has a significant impact or influence on the length of stay. Specifically, we observe that older-aged animals (senior and/or super senior) appear as a significant factor for every algorithm. For the artificial neural network, older age is the #2 and #3 predictor, and super senior is the #2 predictor for the gradient boosting algorithm. Large and small-sized animals are also observed to be important features, as both are shown as the #1 predictor in the gradient boosting and ANN algorithms. The results also demonstrate that gender, animal type, other colors besides multicolor, middle age, and medium-sized animals did not significantly impact the length of stay.

Results from our study provided information on what factors are significant in influencing length of stay. Brown et al. [ 10 ] conducted research that found that age, breed designation, coat color, and coat pattern influenced the length of stay for cats in animal shelters. Similar to these studies, observations from our study also suggest that age and color have a significant impact or influence on the length of stay.

Determining which algorithm will develop the best model for the given set of data is critical to predict the length of stay and minimize the chances of euthanization. The goal of predictive analytics is to develop a model that best approximates the true mapping function for the relationship between the input and output variables. To approximate this function, parametric or non-parametric algorithms can be used. Parametric algorithms simplify the unknown function to a known form. Non-parametric algorithms do not make assumptions about the structure of the mapping function, allowing free learning of any functional form. In this study, we utilize both parametric (logistic regression and artificial neural network) and non-parametric (random forest and gradient boosting) algorithms on the given data. Observing the results from Table 2 , the gradient boosting and random forest (non-parametric algorithms) perform the best on the dataset. It is observed from the results that using a non-parametric approach leads to a better approximation of the true mapping function for the given records. These results also support prior studies on parametric versus non-parametric methods. Neely et al. [ 16 ] detailed the theoretical superiority of non-parametric algorithms for detecting pharmacokinetic and pharmacodynamic subgroups in a study population. The author suggests this superiority comes from the lack of assumptions made about the distribution of parameter values in a dataset. Bissantz et al. [ 17 ] discussed a resampling algorithm that evaluates the deviations between parametric and non-parametric methods to be noise or systematic by comparing parametric models to a non-parametric “supermodel”. Results demonstrate the non-parametric model to be significantly better. The use of algorithms that do not approximate the true function of the relationship between input and output provides better performance results for this application as well.

Current literature also supports the use of ensemble methods to increase prediction accuracy and performance. Dietterich [ 18 ] discussed the ongoing research into developing good ensemble methods as well as the discovery that ensemble algorithms are often more accurate than individual algorithms that are used to create them. Pandey, and S, T [ 19 ]. conducted a study to compare the accuracy of ensemble methodology on predicting student academic performance as research has demonstrated better results for composite models over a single model. This study applied ensemble techniques on learning algorithms (AdaBoost, Random Forest, Rotation Forest, and Bagging). For our study with the given records, the results support this claim. Both the gradient boosting and random forest algorithms are ensemble algorithms and performed the best on the animal shelter data.

Results from Table 2 demonstrate the best performance of the gradient boosting and random forest algorithm when the length of stay was classified as very high or the animal was euthanized. This is beneficial as the models can predict long stays where the outcome is euthanasia. This can lead to shelters identifying at-risk animals and implementing methods and solutions to ensure their adoption. These potential methods are the second phase of this research study, which will involve relocating animals to shelters where they will more likely be adopted. This phase is discussed in the future directions section.

Studies have been conducted evaluating euthanasia-related stress on workers (e.g., [ 1 ]). In other words, overpopulation not only leads to euthanasia but can, in turn, cause mental and emotional problems for the workers. For instance, Reeve et al. [ 20 ] evaluated the strain related to euthanasia among animal workers. Results demonstrated that euthanasia related strain was prevalent, and an increase in substance abuse, job stress, work causing family conflict, complaints, and low job satisfaction was observed. Predicting the length of stay for animals will aid in them being more likely to be adopted and will lead to fewer animals being euthanized, adding value not only to animals finding a home but also less stress on the workers.

The approach developed in this paper could be beneficial not only to reduce euthanasia but also to reduce overcrowding in shelters operated in countries where euthanasia of healthy animals is illegal, and all animals must be housed in shelters until adoption (or natural death). It is essential to develop an information system for a collaborative animal shelter network in which the entities can coordinate with each other, exchanging information about the animal inventory. Another benefit of this study is that it investigates applying machine learning to the animal care domain. Previous studies have looked into what factors influence the length of stay; however, this study utilizes these factors in addition to classification algorithms to predict how long an animal will stay in the shelter. Moreover, the use of a prescriptive analytics approach is discussed in this paper, where the predictions made by the machine learning algorithms will be used along with a goal programming model to decide in what shelter is an animal most likely to be adopted.

Limitations of this study include lack of behavioral data, limited sample size, and the use of simple algorithms. The first limitation, lack of behavioral data of the animal during intake and outcome, would be beneficial to develop a more comprehensive model. Though behavioral problems are harder to solve, having data would provide insight into how long these animals with behavioral issues are staying in shelters and what the outcome is. Studies have shown that behavioral problems play a significant role in preventing bonding between owners and their animals and one of the most common reasons cited for animal surrender [ 21 , 22 ]. These behavioral problems can include poor manners, too much energy, aggression, and destruction of the household. Dogs surrendered to shelters because of behavioral issues have also been shown to be less likely to be adopted or rehomed, and the ones that are adopted are more likely to be returned [ 21 ]. Studies have also been conducted to evaluate the effect of the length of time on the behavior of dogs in rescue shelters [ 23 , 24 , 25 ]. Most of them concluded that environmental factors led to changes in the behavior of dogs and that a prolonged period in a shelter may lead to unattractive behavior of dogs to potential owners. Acquiring information on behavioral problems gives more information for the algorithm to learn when developing the predictive model. This allows more in-depth predictions to be made on how long an animal will stay in a shelter, which could also aid in adoption. This approach can be used to shorten the length of stay, which makes sure that healthy animals are not developing behavioral problems in the shelters. It is not only crucial for the animal to be adopted, but also that the adoption is a good fit between owner and pet. Shortening the length of stay would also lessen the chance that the animal will be returned by the adopter because of behavior. Having this information will also allow shelters to find other shelters close by where animals with behavioral issues are more likely to be adopted. To overcome this limitation of the lack of data on behavioral problems, behavioral issues will be used as a factor and will be specifically asked for when acquiring data from shelters.

Another limitation includes collecting more data from animal shelters across the United States, allowing for more representative data to be collected and inputted into these algorithms. However, this presents a challenge due to most shelters being underfunded and low on staff. Though we reached out to shelters, most replied that they lacked the resources and staff to provide the information needed. Future work would include applying for funding to provide a stipend to staff for their assistance in gathering the data from respective shelters. With more data, the algorithm has more information to learn on, which could improve the performance metrics of the predictive models developed. There may also be other factors that show to be significant as more data is collected.

Finally, the last limitation is the use of simpler algorithms. This study considers basic ML algorithms. Nevertheless, in recent years, there has been development in the ML field of more complex networks. For instance, Zhong et al. [ 26 ] proposed a novel reinforcement learning method to select neural blocks and develop deep learning networks. Results demonstrated high efficiency in comparison to most of the previous deep network search approaches. Though only four algorithms were considered, future work would investigate deep learning networks, as well as bagging algorithms. Using more complex algorithms could ensure that if intricate patterns in the data are present, the algorithm can learn them.

Future direction

Phase 2: goal programming approach for making relocation decisions.

Using the information gathered in this study, we can predict the type of animals that are being adopted the most in each region and during each season of the year. To accomplish this, we utilize a two-phase approach. The first phase was leveraging the machine learning algorithms to predict the length of stay of each animal based on numerous factors (such as breed, size, and color). Phase-2 involves determining the best shelter to transport adoptable animals to increase the adoption rates, based on several conflicting criteria. This criterion includes predicted length of stay from phase-1, the distance between where the animal is currently housed and the potential animal shelters, transportation costs, and transportation time. Therefore, our goal is to increase adoption rates of pets in animal shelters by utilizing several factors to predict the length of stay, as well as determine the optimal animal shelter location where the animal will have the least amount stay in a shelter and most likely be adopted.

After predicting the length of stay of an incoming animal that is currently housed in the shelter l ′ using the machine learning algorithms, the next phase is to evaluate the potential relocation options for that animal. This strategic decision is specifically essential if the length of stay of the animal at its current location is high/very high. Nevertheless, while making this relocation decision, it is also necessary to consider the cost of transporting the animal between the shelters. For instance, if a dog is brought into a shelter in Houston, Texas, and is estimated to have a high/very high length of stay. Suppose if the dog is predicted to have a low length of stay at New York City and a medium length of stay at Oklahoma City, then a tradeoff has to be made between the relocation cost and the adoption speed. The objectives, length of stay, and relocation costs are conflicting and have to be minimized. Phase-2 attempts to yield a compromise solution that establishes a trade-off between these two criteria.

Goal programming (GP) is a widely used approach to solve problems involving multiple conflicting criteria. Under this method, each objective function is assigned as a goal, and a target value is specified for the individual criterion [ 27 ]. These target numbers can be fulfilled by the model with certain deviations, while the objective of the GP model is to minimize these deviations. Pertaining to this study, the desired values for the length of stay and relocation cost is pre-specified in the model and can be fulfilled with deviations. The GP model attempts to minimize these deviations. Thus, this technique attempts to produce a solution that is as close as possible to the targets, and the model solutions are referred to as the “most preferred solution” by prior studies (e.g., [ 28 , 29 ]).

As mentioned earlier, the primary task to be completed using this phase-2 goal programming approach is the relocation decisions considering the adoption speed and the cost of transporting the animal from the current location.

Model notations

Goal programming model formulation, goal constraints.

Objective 1: Minimize the overall length of stay of the animal under consideration (Eq. 1 ).

Goal constraint for objective 1: The corresponding goal constraint of objective 2 is given using Equation [ 30 ].

Objective 2: Minimize the overall relocation cost for transporting the animal under consideration (Eq. 3 ).

Goal constraint for objective 2: The corresponding goal constraint of objective 2 is given using Equation [ 18 ].

Hard constraints

Equation [ 9 ] ensures that the animal can be assigned to only one shelter.

The animal can be accommodated in shelter l only if there are a shelter capacity and type for that particular animal size category, and this is guaranteed using constraint [ 31 ]. It is important to note that both y and s are input parameters , whereas l is the set of shelters.

Equation [ 21 ] sets an upper limit on the length of stay category if the shelter l is assigned as the destination location. This prevents relocating animals to a shelter that might potentially have a high or very high length of stay.

Similarly, Equation [ 32 ] sets an upper limit on the relocation cost, if the shelter l is assigned as the destination location. This prevents relocating animals to a very far location. The current shelter location, l ′ , that is hosting the animal is an input parameter.

Objective function

Since the current problem focuses on minimizing the expected length of stay and relocation cost, the objective function of the goal programming approach is to reduce the sum of the weighted positive deviations given in Equations ([ 18 , 30 ], as shown in Equation [ 6 ].

where w g is the weight assigned for each goal g .

It is necessary to scale the deviation (since the objectives have different magnitudes as well as units) to avoid a biased solution.

If the scaling factors are represented by f g for goal g , then the scaled objective function is given in Equation [ 14 ].

Using this goal programming approach, the potential relocation options are evaluated considering the length of stay from phase-1. This phase-2 goal programming approach is useful, especially if the length of stay of the animal at its current location is high/very high, and a trade-off has to be made between relocation cost and length of stay. Phase-2 acts as a recommendation tool for assisting administrators with relocation decisions.

Nearly 3–4 million animals are euthanized out of the 6–8 million animals that enter shelters annually. The overall objective of this study is to increase the adoption rates of animals entering shelters by using key factors found in the literature to predict the length of stay. The second phase determines the best shelter location to transport animals using the goal programming approach to make relocation decisions. To accomplish this objective, first, the data is acquired from online sources as well as from numerous shelters across the United States. Once the data is acquired and cleaned, predictive models are developed using logistic regression, artificial neural network, gradient boosting, and random forest. The performance of these models is determined using three performance metrics: precision, recall, and F1 score.

The results demonstrate that the gradient boosting algorithm performed the best overall, with the highest precision, recall, and F1 score. Followed closely in second is the random forest algorithm, then the artificial neural network, and then finally, the logistic regression algorithm is the worst performer. We also observed from the data that the gradient boosting performed better when predicting the high or very high length of stay. Further observing the results, it is found that age for dogs (e.g., puppy, super senior), multicolor, and large and small size are important predictor variables.

The findings from this study can be utilized to predict how long an animal will stay in a shelter, as well as minimize their length of stay and chance of euthanization by determining which shelter location will most likely lead to the adoption of that animal. For future studies, we will implement phase 2, which will determine the best shelter location to transport animals using the goal programming approach to make relocation decisions.

Data description

A literature review is conducted to determine the factors that might potentially influence the length of stay for animals in shelters. These factors include gender, breed, age, and several other variables that are listed in Table 4 . These features will be treated as input variables for the machine learning algorithms. Overall, there are eight input or predictor variables and one output variable, which is the length of stay.

Animal shelter intake and outcome data are publicly made available by several state/city governments on their website (e.g., [ 33 , 34 ]), specifically in several southern and south-western states. These online sources provide datasets for animal shelters from Kentucky (150,843 data rows), California (334,016), Texas (155,115), and Indiana (4132). Since there is no nationwide database for animal shelters, information is also collected through individual animal shelters that conduct euthanization of animals. We contacted over 100 animal shelters across the United States and inquired for data on the factors mentioned in Table 4 . We received responses from 20 of the animal shelters that were contacted. Most responses received stated there was not enough staff or resources to be able to provide this information. From the responses that were received back, only four shelters were able to provide any information. Of those four, only two of the datasets contained the factors and information needed, which are Colorado (8488 data rows) and Arizona (4, 667 data rows).

The data that is collected from the database and animal shelters included information such as animal type, intake and outcome date, gender, color, breed, and intake and outcome status (behavior of animal entering the shelter and behavior of animal at outcome type). These records also included information on several types of animals, such as dogs, cats, birds, rabbits, and lizards. For this study, the focus is on dogs and cats. After filtering through these records, we found that only California, Kentucky, Colorado, Arizona, and Indiana had all of the factors needed for the study. Upon downloading data from the database and receiving data from the animal shelters, the acquired data underwent data integration, data transformation, and data cleaning (as detailed in Fig.  1 ). After data pre-processing, there are over 113,000 animal records.

Data cleaning methods

Next, data cleaning methods are utilized to detect discrepancies in the data, such as missing values, erroneous data, and inconsistencies. Data cleaning is an essential step for obtaining unbiased results [ 35 , 36 ]. In other words, identifying and cleaning erroneous data must be performed before inputting the data into the algorithm as it can significantly impact the output results.

The following is a list of commonly used data cleaning techniques in the literature [ 11 ]:

Substitution with Median: Missing or incorrect data are replaced with the median value for that predictor variable.

Substitution with a Unique Value: Erroneous data are replaced with a value that does not fall within the range that the input variables can accept (e.g., a negative number)

Discard Variable and Substitute with a Median: When an input variable has a significant number of missing values, these values are removed from the dataset, and the features that remain with missing or erroneous values are replaced with the median.

Discard Variable and Substitute with a Unique Value: Input variables with a significant number of missing values are removed from the dataset, and the features that remain with missing or erroneous values are coded as − 1.

Remove Incomplete Rows Entirely: Incomplete Rows are removed from the dataset.

Data preprocessing

Some animal breeds are listed in multiple formats and are changed to maintain uniformity. An example of this is a Russian Blue cat, which is formatted in several ways such as “Russian”, “Russian Blue”, and “RUSSIAN BLUE”. Animals with multiple breeds such as “Shih Tzu/mix” or “Shih Tzu/Yorkshire Terr” are classified as the first breed listed. Other uncommon breeds are classified as “other” for simplicity. Finally, all animal breeds are summarized into three categories (small, medium, or large) using the American Kennel Clubs’ breed size classification [ 37 ]. Part of the data cleansing process also includes categorizing multiple colors found throughout the sample size into five distinct color categories (brown, black, blue, white, and multicolor). We classified age into five categories for dogs and cats (puppy or kitten, adolescent, adult, senior, super senior). The puppy or kitten category includes data points 0–1 year, adolescence includes data points 2–3 years old, adulthood includes animals 4–7 years of age, and senior animals are 8–10 years of age. Any animal that is older than ten years are categorized as a super senior, based on the recommendations provided in Wapiti Labs [ 38 ].

As mentioned previously, the output variable is the length of stay and is classified as low, medium, high, and very high/euthanization. The length of stay is calculated by taking the difference between the intake date and outcome date. To remove erroneous data entries and special cases, the number of days in the animal shelter is also capped at a year. The “low” category represents animals that are returned (in which case, they are assigned the days in the shelter as 0) or spent less than 8 days before getting adopted. It is important to keep these animals at the shelter so that the owner may find them or they are transferred to their new homes. Animals that stayed in a shelter for 9–42 days and are adopted are categorized as “medium” length of stay. The “high” category is given to animals that stayed in the shelter for 43–365 days. Finally, animals that are euthanized are categorized as “very high”.

After integrating all data points from each animal shelter, the sample size includes 119,691 records. After the evaluation of these data points, 5436 samples are found to have miscellaneous (such as a negative length of stay) or missing values. After applying data cleaning techniques, the final cleaned dataset includes 114,256 data points, with 50,466 cat- and 63,790 dog-records.

Machine learning algorithms to predict the length of stay

The preprocessed records are then separated into training and testing datasets based on the type of classification algorithm used. Studies have demonstrated the need for testing and comparing machine learning algorithms, as the performance of the models depends on the application. While an algorithm may develop a predictive model that performs well in one application, it may not be the best performing model for another. A comparison between the statistical models is conducted to determine the overall best performing model. In this section, we provide a description as well as the advantages of each classification algorithm that is utilized in this study.

Logistic regression

Logistic regression (LR) is a machine learning algorithm that is used to understand the probability of the occurrence of an event [ 39 ]. It is typically used when the model output variable is binary or categorical (see Fig.  6 ), unlike linear regression, where the dependent variable is numeric [ 40 ]. Logistic regression involves the use of a logistic function, referred to as a “sigmoid function” that takes a real-valued number and maps it into a value between 0 and 1 [ 41 ]. The probability that the length of stay of the animal at a specific location will be low, medium, high, or very high, is computed using the input features discussed in Table 4 .

Pictorial Representation of the Logistic Regression Algorithm

The linear predictor function to predict the probability that the animal in record i has a low, medium, high, and very high length of stay categories is given by Equations ( 11 ) –[ 3 ], respectively.

Where β v , l is a set of multinomial logistic regression coefficients for variable v of the length of stay category l , and x v , i is the input feature v corresponding to data observation i .

Artificial neural network

Artificial Neural Network (ANN) algorithms were inspired by the brain’s neuron, which transmits signals to other nerve cells [ 40 , 42 ]. ANN’s were designed to replicate the way humans learn and were developed to imitate the operational sequence in which the body sends signals in the nervous system [ 43 ]. In an ANN, there exists a network structure with directional links connecting multiple nodes or “artificial neurons”. These neurons are information-processing units, and the ties that connect them represent the relationship between each of the connected neurons. Each ANN consists of three layers - the input layer, hidden layer, and the output layer [ 32 , 44 ]. The input layer is where each of the input variables is fed into the artificial neuron. The neuron will first calculate the sum of multiple inputs from the independent variables. Each of the connecting links (synapses) from these inputs has a characterized weight or strength that has a negative or positive value [ 32 ]. When new data is received, the synaptic weight changes, and learning will occur. The hidden layer learns the relationship between the input and output variables, and a threshold value determines whether the artificial neuron will fire or pass the learned information to the output layer, as shown in Fig.  7 . Finally, the output layer is where labels are given to the output value, and backpropagation is used to correct any errors.

Pictorial Representation of the Artificial Neural Networks

Random Forest

The Random Forest (RF) algorithm is a type of ensemble methodology that combines the results of multiple decision trees to create a new predictive model that is less likely to misclassify new data [ 30 , 45 ]. Decision Trees have a root node at the top of the tree that consists of the attribute that best classifies the training data. The attribute with the highest information gain (given in Eq. 16 ) is used to determine the best attribute at each level/node. The root node will be split into more subnodes, which are categorized as a decision node or leaf node. A decision node can be divided into further subnodes, while a leaf node cannot be split further and will provide the final classification or discrete label. RF algorithm uses mtree and ntry as the two main parameters in developing the multiple parallel decision trees. Mtree specifies how many trees to train in parallel, while ntry defines the number of independent variables or attributes to choose to split each node [ 30 ].. The majority voting from all parallel trees gives the final prediction, as given in Fig.  8 .

Pictorial Representation of the Random Forest Algorithm

Gradient boosting

Boosting is another type of ensemble method that combines the results from multiple predictive algorithms to develop a new model. While the RF approach is built solely on decision trees, boosting algorithms can use various algorithms such as decision trees, logistic regression, and neural networks. The primary goal of boosting algorithms is to convert weak learners into stronger ones by leveraging weighted averages to identify “weak classifiers” [ 31 ]. Samples are assigned an initial uniformed weight, and when incorrectly labeled by the algorithm, a penalty of an increase in weight is given [ 46 ]. On the other hand, samples that are correctly classified by the algorithm will decrease in weight. This process of re-weighing is done until a weighted vote of weak classifiers is combined into a robust classifier that determines the final labels or classification [ 46 ]. For our study, gradient boosting (GB) will be used on decision trees for the given dataset, as illustrated in Fig.  9 .

Pictorial Representation of Boosting Algorithm

Machine learning model parameters

The clean animal shelter data is split into two datasets: training and testing data. These records are randomly placed in the two groups to train the algorithms and to test the model developed by the algorithm. 80% of the data is used to train the algorithm, while the other 20% is used to test the predictive model. To avoid overfitting, a tenfold cross-validation procedure is used on the training data. There are no parameters associated with the machine learning of logistic regression algorithms. However, a grid search method is used to tune the parameters of the random forest, gradient boosting, and artificial neural network algorithms. This allows the best parameter in a specific set to be chosen by running an in-depth search by the user during the training period.

The number of trees in the random forest and gradient boosting algorithms is changed from 100 to 1000 in increments of 100. A learning rate of 0.01, 0.05, and 0.10 is used based on the recommendations of previous studies [ 47 ]. The minimum observations for the trees’ terminal node are set to vary from 2 to 10 in increments of one, while the splitting of trees varies from 2 to 10 in increments of two. A feed-forward method is used to develop the predictive model using the artificial neural network algorithm. The feed-forward algorithm consists of three layers (input, hidden, output) as well as backpropagation learning. The independent and dependent variables represent the input and output layers. Since the input and output layers are already known, an optimal point is reached for the number of nodes when between 1 and the number of predictors. This means that for our study, the nodes of the hidden layer vary from 1 to 8. The learning rate values used to train the ANN are 0.01, 0.05, and 0.10.

To find the optimal setting for each machine learning algorithm, a thorough search of their corresponding parameter space is performed.

Performance measures

In this study, we use three performance measures to evaluate the ability of machine learning algorithms in developing the best predictive model for the intended application. The measures considered are precision, F1 score, and sensitivity/recall to determine the best model given the inputted data samples. Table 5 provides a confusion matrix to define the terms used for all possible outcomes.

Precision evaluates the number of correct, true positive predictions by the algorithm while still considering the incorrectly predicted positive when it should have been negative (Eq. 17 ). By having high precision, this means that there is a low rate of false positives or type I error. Sensitivity or recall evaluates the number of true positives that are correctly predicted by the algorithm while considering the incorrectly predicted negative when it should have been positive (Eq. 18 ). Recall is a good tool to use when the focus is on minimizing false negatives (type II error). F1 score (shown in Eq. 19 ) evaluates both type I and type II errors and assesses the ability of the model to resist false positives and false negatives. This performance metric evaluates the robustness (low number of missed classifications), as well as the number of data points that are classified correctly by the model.

Availability of data and materials

Most of the datasets used and/or analyzed during the current study were publicly available online as open source data. The data were available in the website details given below:

https://data.bloomington.in.gov/dataset

https://data.louisvilleky.gov/dataset

https://data.sonomacounty.ca.gov/Government

We also obtained data from Sun Cities 4 Paws Rescue, Inc., and the Rifle Animal Shelter. No administrative permission was required to access the raw data from these shelters.

Abbreviations

Logistic Regression

Artificial Neural Network

Gradient Boosting

Goal Programming

Coefficient of Variation

Anderson KA, Brandt JC, Lord LK, Miles EA. Euthanasia in animal shelters: Management's perspective on staff reactions and support programs. Anthrozoös. 2013;26(4):569–78. https://doi.org/10.2752/175303713X13795775536057 .

Article   Google Scholar  

Clevenger J, Kass PH. Determinants of adoption and euthanasia of shelter dogs spayed or neutered in the University of California veterinary student surgery program compared to other shelter dogs. J Veterinary Med Educs. 2003;30(4):372–8.

Animal Humane Society. (n.d.). Retrieved November 2019, from https://www.animalhumanesociety.org/ .

Home. (2016, July 15). Retrieved November 2019, from http://www.americanhumane.org/ .

Rogelberg SG, DiGiacomo N, Reeve CL, Spitzmüller C, Clark OL, Teeter L, et al. What shelters can do about euthanasia-related stress: an examination of recommendations from those on the front line. J Appl Anim Welf Sci. 2007;10(4):331–47. https://doi.org/10.1080/10888700701353865 .

Article   CAS   PubMed   Google Scholar  

Hennessy MB, Voith VL, Mazzei SJ, Buttram J, Miller DD, Linden F. Behavior and cortisol levels of dogs in a public animal shelter, and an exploration of the ability of these measures to predict problem behavior after adoption. Appl Anim Behav Sci. 2001;73(3):217–33.

Shore ER. Returning a recently adopted companion animal: Adopters' reasons for and reactions to the failed adoption experience. J Appl Anim Welf Sci. 2005;8(3):187–98.

Article   CAS   Google Scholar  

Morris KN, Gies DL. Trends in intake and outcome Data for animal shelters in a large U.S. metropolitan area, 1989 to 2010. J Appl Anim Welf Sci. 2014;17(1):59–72. https://doi.org/10.1080/10888705.2014.856250 .

Fantuzzi JM, Miller KA, Weiss E. Factors relevant to adoption of cats in an animal shelter. J Appl Anim Welf Sci. 2010;13(2):174–9.

Brown WP, Morgan KT. Age, breed designation, coat color, and coat pattern influenced the length of stay of cats at a no-kill shelter. J Appl Anim Welf Sci. 2015;18(2):169–80.

Srinivas, S., & Rajendran, S. (2017). A Data-driven approach for multiobjective loan portfolio optimization using machine-learning algorithms and mathematical programming. In big Data analytics using multiple criteria decision-making models (pp. 175-210): CRC press.

Waller MA, Fawcett SE. Data science, predictive analytics, and big Data: a revolution that will transform supply chain design and management. J Bus Logist. 2013;34(2):77–84.

Kantardzic M. DATA MINING: concepts, models, methods, and algorithms. 2nd ed: IEEE: Wiley; 2019.

Jordan MI, Mitchell TM. Machine learning: trends, perspectives, and prospects. Science. 2015;349(6245):255–60.

Kavakiotis I, Tsave O, Salifoglou A, Maglaveras N, Vlahavas I, Chouvarda I. Machine learning and Data mining methods in diabetes research. Computational Structural Biotechnol J. 2017;15:104–16. https://doi.org/10.1016/j.csbj.2016.12.005 .

Neely MN, van Guilder MG, Yamada WM, Schumitzky A, Jelliffe RW. Accurate detection of outliers and subpopulations with Pmetrics, a nonparametric and parametric pharmacometric modeling and simulation package for R. Ther Drug Monit. 2012;34(4):467–76. https://doi.org/10.1097/FTD.0b013e31825c4ba6 .

Article   PubMed   PubMed Central   Google Scholar  

Bissantz N, Munk A, Scholz A. Parametric versus non-parametric modelling? Statistical evidence based on P-value curves. Mon Not R Astron Soc. 2003;340(4):1190–8. https://doi.org/10.1046/j.1365-8711.2003.06377.x .

Dietterich TG. Ensemble methods in machine learning. Berlin: Heidelberg; 2000.

Book   Google Scholar  

Pandey M, S, T. A comparative study of ensemble methods for students' performance modeling. Int J Computer ApplS. 2014;103:26–32. https://doi.org/10.5120/18095-9151 .

Reeve CL, Rogelberg SG, Spitzmüller C, Digiacomo N. The caring-killing paradox: euthanasia-related strain among animal-shelter Workers1. J Appl Soc Psychol. 2005;35(1):119–43. https://doi.org/10.1111/j.1559-1816.2005.tb02096.x .

Gates MC, Zito S, Thomas J, Dale A. Post-adoption problem Behaviours in adolescent and adult dogs rehomed through a New Zealand animal shelter. Animals : an open access journal from MDPI. 2018;8(6):93. https://doi.org/10.3390/ani8060093 .

Weiss E, Gramann S, Drain N, Dolan E, Slater M. Modification of the feline-Ality™ assessment and the ability to predict adopted Cats' behaviors in their new homes. Animals : an open access journal from MDPI. 2015;5(1):71–88. https://doi.org/10.3390/ani5010071 .

Normando S, Stefanini C, Meers L, Adamelli S, Coultis D, Bono G. Some factors influencing adoption of sheltered dogs. Anthrozoös. 2006;19(3):211–24.

Protopopova A, Mehrkam LR, Boggess MM, Wynne CDL. In-kennel behavior predicts length of stay in shelter dogs. PLoS One. 2014;9(12):e114319.

Wells DL, Graham L, Hepper PG. The influence of length of time in a rescue shelter on the behaviour of Kennelled dogs. Anim Welf. 2002;11(3):317–25.

CAS   Google Scholar  

Zhong G, Jiao W, Gao W, Huang K. Automatic design of deep networks with neural blocks. Cogn Comput. 2020;12(1):1–12.

Rajendran S, Ravindran AR. Multi-criteria approach for platelet inventory management in hospitals. Int J Operational ResS. 2020;38(1):49–69.

Bastian ND, McMurry P, Fulton LV, Griffin PM, Cui S, Hanson T, Srinivas S. The AMEDD uses goal programming to optimize workforce planning decisions. Interfaces. 2015;45(4):305–24.

Rajendran S, Ansaripour A, Kris Srinivasan M, Chandra MJ. Stochastic goal programming approach to determine the side effects to be labeled on pharmaceutical drugs. IISE Transactions on Healthcare Systems Engineering. 2019;9(1):83–94.

Cutler DR, Edwards TC Jr, Beard KH, Cutler A, Hess KT, Gibson J, Lawler JJ. Random forests for classification in ECOLOGY. Ecology. 2007;88(11):2783–92.

Friedman J, Hastie T, Tibshirani R. Additive logistic regression: a statistical view of boosting (with discussion and a rejoinder by the authors). Ann Stat. 2000;28(2):337–407.

Ge Z, Song Z, Ding SX, Huang B. Data mining and analytics in the process industry: the role of machine learning. IEEE Access. 2017;5:20590–616.

Open Data: City of Austin Texas: Open Data: City of Austin Texas. (n.d.). Retrieved March 2019, from https://data.austintexas.gov//Health-and-Community-Services/Austin-Animal-Center-Outcomes/9t4d-g238 .

County of Sonoma: Open Data: Open Data. (n.d.). Retrieved March 2019, from https://data.sonomacounty.ca.gov/Government/Animal-Shelter-Intake-and-Outcome/924a-vesw .

Kambli A, Sinha AA, Srinivas S. Improving campus dining operations using capacity and queue management: a simulation-based case study. J Hosp Tour Manag. 2020;43:62–70.

Rajendran S, Zack J. Insights on strategic air taxi network infrastructure locations using an iterative constrained clustering approach. Transport Res Part E: Logistics and Transportation Review. 2019;128:470–505.

American Kennel Club. (n.d.). Retrieved November 2019, from http://www.akc.org/ .

Elk Antler Supplements & Chews: Wapiti Labs, Inc. (n.d.). Retrieved November 2019, from https://www.wapitilabsinc.com/ .

Bursac Z, Gauss CH, Williams DK, Hosmer DW. Purposeful selection of variables in logistic regression. Source Code for Biol Med. 2008;3(1):17.

Delen D, Walker G, Kadam A. Predicting breast cancer survivability: a comparison of three data mining methods. Artif Intell Med. 2005;34(2):113–27.

Kim A, Song Y, Kim M, Lee K, Cheon JH. Logistic regression model training based on the approximate homomorphic encryption. BMC Med Genet. 2018;11(4):83.

Google Scholar  

Srinivas S, Ravindran AR. Optimizing outpatient appointment system using machine learning algorithms and scheduling rules: a prescriptive analytics framework. Expert Syst Appl. 2018;102:245–61. https://doi.org/10.1016/j.eswa.2018.02.022 .

LeCun Y, Bengio Y, Hinton G. Deep learning. Nature. 2015;521:436.

Shih H, Rajendran S. Comparison of time series methods and machine learning algorithms for forecasting Taiwan blood Services Foundation’s blood supply. Journal of healthcare engineering. 2019;2019.

Srinivas S, Salah H. Consultation length and no-show prediction for improving appointment scheduling efficiency at a cardiology clinic: a data analytics approach. Int J Med Inform. 2020;145:104290.

Rokach L. Ensemble-based classifiers. Artif Intell Rev. 2010;33(1):1–39.

Srinivas S. A machine learning-based approach for predicting patient punctuality in ambulatory care centers. Int J Environ Res Public Health. 2020;17(10):3703.

Download references

Acknowledgments

We would like to thank the Sun Cities 4 Paws Rescue, Inc., and the Rifle Animal Shelter for providing the length of stay reports in order to complete this study.

This research was not funded by any agency/grant.

Author information

Authors and affiliations.

Department of Bioengineering, University of Missouri Columbia, Columbia, MO, 65211, USA

Janae Bradley

Department of Industrial and Manufacturing Systems Engineering, University of Missouri Columbia, Columbia, MO, 65211, USA

Suchithra Rajendran

Department of Marketing, University of Missouri Columbia, Columbia, MO, 65211, USA

You can also search for this author in PubMed   Google Scholar

Contributions

JB performed data mining, data cleaning and analyses of the animal shelter data and machine learning algorithms. JB was also a major contributor in writing the manuscript. SR performed data mining, cleaning, and analyses of the machine learning algorithms, as well as the goal programming. All authors have read and approved the final manuscript.

Corresponding author

Correspondence to Suchithra Rajendran .

Ethics declarations

Ethics approval and consent to participate.

Most of the datasets used in this study are open source and are publicly available. The remaining data was collected from animal shelters with their consent to use the data for research purposes.

Consent for publication

Not applicable.

Competing interests

The authors declare that they have no competing interests.

Additional information

Publisher’s note.

Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.

Rights and permissions

Open Access This article is licensed under a Creative Commons Attribution 4.0 International License, which permits use, sharing, adaptation, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons licence, and indicate if changes were made. The images or other third party material in this article are included in the article's Creative Commons licence, unless indicated otherwise in a credit line to the material. If material is not included in the article's Creative Commons licence and your intended use is not permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder. To view a copy of this licence, visit http://creativecommons.org/licenses/by/4.0/ . The Creative Commons Public Domain Dedication waiver ( http://creativecommons.org/publicdomain/zero/1.0/ ) applies to the data made available in this article, unless otherwise stated in a credit line to the data.

Reprints and permissions

About this article

Cite this article.

Bradley, J., Rajendran, S. Increasing adoption rates at animal shelters: a two-phase approach to predict length of stay and optimal shelter allocation. BMC Vet Res 17 , 70 (2021). https://doi.org/10.1186/s12917-020-02728-2

Download citation

Received : 07 January 2020

Accepted : 22 December 2020

Published : 05 February 2021

DOI : https://doi.org/10.1186/s12917-020-02728-2

Share this article

Anyone you share the following link with will be able to read this content:

Sorry, a shareable link is not currently available for this article.

Provided by the Springer Nature SharedIt content-sharing initiative

• Animal shelter
• High euthanization rates
• Machine learning algorithms
• Prediction models
• Goal programming approach
• Decision support tool

BMC Veterinary Research

ISSN: 1746-6148

• General enquiries: [email protected]

An official website of the United States government

The .gov means it’s official. Federal government websites often end in .gov or .mil. Before sharing sensitive information, make sure you’re on a federal government site.

The site is secure. The https:// ensures that you are connecting to the official website and that any information you provide is encrypted and transmitted securely.

• Publications
• Account settings

Preview improvements coming to the PMC website in October 2024. Learn More or Try it out now .

• Advanced Search
• Journal List
• J Prev Med Hyg
• v.63(2 Suppl 3); 2022 Jun

Ethical considerations regarding animal experimentation

Aysha karim kiani.

1 Allama Iqbal Open University, Islamabad, Pakistan

2 MAGI EUREGIO, Bolzano, Italy

DEREK PHEBY

3 Society and Health, Buckinghamshire New University, High Wycombe, UK

GARY HENEHAN

4 School of Food Science and Environmental Health, Technological University of Dublin, Dublin, Ireland

RICHARD BROWN

5 Department of Psychology and Neuroscience, Dalhousie University, Halifax, Nova Scotia, Canada

PAUL SIEVING

6 Department of Ophthalmology, Center for Ocular Regenerative Therapy, School of Medicine, University of California at Davis, Sacramento, CA, USA

PETER SYKORA

7 Department of Philosophy and Applied Philosophy, University of St. Cyril and Methodius, Trnava, Slovakia

ROBERT MARKS

8 Department of Biotechnology Engineering, Ben-Gurion University of the Negev, Beer-Sheva, Israel

BENEDETTO FALSINI

9 Institute of Ophthalmology, Università Cattolica del Sacro Cuore, Fondazione Policlinico Universitario A. Gemelli-IRCCS, Rome, Italy

NATALE CAPODICASA

10 MAGI BALKANS, Tirana, Albania

STANISLAV MIERTUS

11 Department of Biotechnology, University of SS. Cyril and Methodius, Trnava, Slovakia

12 International Centre for Applied Research and Sustainable Technology, Bratislava, Slovakia

LORENZO LORUSSO

13 UOC Neurology and Stroke Unit, ASST Lecco, Merate, Italy

DANIELE DONDOSSOLA

14 Center for Preclincal Research and General and Liver Transplant Surgery Unit, Fondazione IRCCS Ca‘ Granda Ospedale Maggiore Policlinico, Milan, Italy

15 Department of Pathophysiology and Transplantation, Università degli Studi di Milano, Milan, Italy

GIANLUCA MARTINO TARTAGLIA

16 Department of Biomedical, Surgical and Dental Sciences, Università degli Studi di Milano, Milan, Italy

17 UOC Maxillo-Facial Surgery and Dentistry, Fondazione IRCCS Ca Granda, Ospedale Maggiore Policlinico, Milan, Italy

MAHMUT CERKEZ ERGOREN

18 Department of Medical Genetics, Faculty of Medicine, Near East University, Nicosia, Cyprus

MUNIS DUNDAR

19 Department of Medical Genetics, Erciyes University Medical Faculty, Kayseri, Turkey

SANDRO MICHELINI

20 Vascular Diagnostics and Rehabilitation Service, Marino Hospital, ASL Roma 6, Marino, Italy

DANIELE MALACARNE

21 MAGI’S LAB, Rovereto (TN), Italy

GABRIELE BONETTI

Astrit dautaj, kevin donato, maria chiara medori, tommaso beccari.

22 Department of Pharmaceutical Sciences, University of Perugia, Perugia, Italy

MICHELE SAMAJA

23 MAGI GROUP, San Felice del Benaco (BS), Italy

STEPHEN THADDEUS CONNELLY

24 San Francisco Veterans Affairs Health Care System, University of California, San Francisco, CA, USA

DONALD MARTIN

25 Univ. Grenoble Alpes, CNRS, Grenoble INP, TIMC-IMAG, SyNaBi, Grenoble, France

ASSUNTA MORRESI

26 Department of Chemistry, Biology and Biotechnology, University of Perugia, Perugia, Italy

ARIOLA BACU

27 Department of Biotechnology, University of Tirana, Tirana, Albania

KAREN L. HERBST

28 Total Lipedema Care, Beverly Hills California and Tucson Arizona, USA

MYKHAYLO KAPUSTIN

29 Federation of the Jewish Communities of Slovakia

LIBORIO STUPPIA

30 Department of Psychological, Health and Territorial Sciences, School of Medicine and Health Sciences, University "G. d'Annunzio", Chieti, Italy

LUDOVICA LUMER

31 Department of Anatomy and Developmental Biology, University College London, London, UK

GIAMPIETRO FARRONATO

Matteo bertelli.

32 MAGISNAT, Peachtree Corners (GA), USA

Animal experimentation is widely used around the world for the identification of the root causes of various diseases in humans and animals and for exploring treatment options. Among the several animal species, rats, mice and purpose-bred birds comprise almost 90% of the animals that are used for research purpose. However, growing awareness of the sentience of animals and their experience of pain and suffering has led to strong opposition to animal research among many scientists and the general public. In addition, the usefulness of extrapolating animal data to humans has been questioned. This has led to Ethical Committees’ adoption of the ‘four Rs’ principles (Reduction, Refinement, Replacement and Responsibility) as a guide when making decisions regarding animal experimentation. Some of the essential considerations for humane animal experimentation are presented in this review along with the requirement for investigator training. Due to the ethical issues surrounding the use of animals in experimentation, their use is declining in those research areas where alternative in vitro or in silico methods are available. However, so far it has not been possible to dispense with experimental animals completely and further research is needed to provide a road map to robust alternatives before their use can be fully discontinued.

How to cite this article: Kiani AK, Pheby D, Henehan G, Brown R, Sieving P, Sykora P, Marks R, Falsini B, Capodicasa N, Miertus S, Lorusso L, Dondossola D, Tartaglia GM, Ergoren MC, Dundar M, Michelini S, Malacarne D, Bonetti G, Dautaj A, Donato K, Medori MC, Beccari T, Samaja M, Connelly ST, Martin D, Morresi A, Bacu A, Herbst KL, Kapustin M, Stuppia L, Lumer L, Farronato G, Bertelli M. Ethical considerations regarding animal experimentation. J Prev Med Hyg 2022;63(suppl.3):E255-E266. https://doi.org/10.15167/2421-4248/jpmh2022.63.2S3.2768

Introduction

Animal model-based research has been performed for a very long time. Ever since the 5 th century B.C., reports of experiments involving animals have been documented, but an increase in the frequency of their utilization has been observed since the 19 th century [ 1 ]. Most institutions for medical research around the world use non-human animals as experimental subjects [ 2 ]. Such animals might be used for research experimentations to gain a better understanding of human diseases or for exploring potential treatment options [ 2 ]. Even those animals that are evolutionarily quite distant from humans, such as Drosophila melanogaster , Zebrafish ( Danio rerio ) and Caenorhabditis elegans , share physiological and genetic similarities with human beings [ 2 ]; therefore animal experimentation can be of great help for the advancement of medical science [ 2 ].

For animal experimentation, the major assumption is that the animal research will be of benefit to humans. There are many reasons that highlight the significance of animal use in biomedical research. One of the major reasons is that animals and humans share the same biological processes. In addition, vertebrates have many anatomical similarities (all vertebrates have lungs, a heart, kidneys, liver and other organs) [ 3 ]. Therefore, these similarities make certain animals more suitable for experiments and for providing basic training to young researchers and students in different fields of biological and biomedical sciences [ 3 ]. Certain animals are susceptible to various health problems that are similar to human diseases such as diabetes, cancer and heart disease [ 4 ]. Furthermore, there are genetically modified animals that are used to obtain pathological phenotypes [ 5 ]. A significant benefit of animal experimentation is that test species can be chosen that have a much shorter life cycle than humans. Therefore, animal models can be studied throughout their life span and for several successive generations, an essential element for the understanding of disease progression along with its interaction with the whole organism throughout its lifetime [ 6 ].

Animal models often play a critical role in helping researchers who are exploring the efficacy and safety of potential medical treatments and drugs. They help to identify any dangerous or undesired side effects, such as birth defects, infertility, toxicity, liver damage or any potential carcinogenic effects [ 7 ]. Currently, U.S. Federal law, for example, requires that non-human animal research is used to demonstrate the efficacy and safety of any new treatment options before proceeding to trials on humans [ 8 ]. Of course, it is not only humans benefit from this research and testing, since many of the drugs and treatments that are developed for humans are routinely used in veterinary clinics, which help animals live longer and healthier lives [ 4 ].

COVID-19 AND THE NEED FOR ANIMAL MODELS

When COVID-19 struck, there was a desperate need for research on the disease, its effects on the brain and body and on the development of new treatments for patients with the disease. Early in the disease it was noticed that those with the disease suffered a loss of smell and taste, as well as neurological and psychiatric symptoms, some of which lasted long after the patients had “survived” the disease [ 9-15 ]. As soon as the pandemic started, there was a search for appropriate animal models in which to study this unknown disease [ 16 , 17 ]. While genetically modified mice and rats are the basic animal models for neurological and immunological research [ 18 , 19 ] the need to understand COVID-19 led to a range of animal models; from fruit flies [ 20 ] and Zebrafish [ 21 ] to large mammals [ 22 , 23 ] and primates [ 24 , 25 ]. And it was just not one animal model that was needed, but many, because different aspects of the disease are best studied in different animal models [ 16 , 25 , 26 ]. There is also a need to study the transmission pathways of the zoonosis: where does it come from, what are the animal hosts and how is it transferred to humans [ 27 ]?

There has been a need for animal models for understanding the pathophysiology of COVID-19 [ 28 ], for studying the mechanisms of transmission of the disease [ 16 ], for studying its neurobiology [ 29 , 30 ] and for developing new vaccines [ 31 ]. The sudden onset of the COVID-19 pandemic has highlighted the fact that animal research is necessary, and that the curtailment of such research has serious consequences for the health of both humans and animals, both wild and domestic [ 32 ] As highlighted by Adhikary et al. [ 22 ] and Genzel et al. [ 33 ] the coronavirus has made clear the necessity for animal research and the danger in surviving future such pandemics if animal research is not fully supported. Genzel et al. [ 33 ], in particular, take issue with the proposal for a European ban on animal testing. Finally, there is a danger in bypassing animal research in developing new vaccines for diseases such as COVID-19 [ 34 ]. The purpose of this paper is to show that, while animal research is necessary for the health of both humans and animals, there is a need to carry out such experimentation in a controlled and humane manner. The use of alternatives to animal research such as cultured human cells and computer modeling may be a useful adjunct to animal studies but will require that such methods are more readily accessible to researchers and are not a replacement for animal experimentation.

Pros and cons of animal experimentation

Arguments against animal experimentation.

A fundamental question surrounding this debate is to ask whether it is appropriate to use animals for medical research. Is our acceptance that animals have a morally lower value or standard of life just a case of speciesism [ 35 ]? Nowadays, most people agree that animals have a moral status and that needlessly hurting or abusing pets or other animals is unacceptable. This represents something of a change from the historical point of view where animals did not have any moral status and the treatment of animals was mostly subservient to maintaining the health and dignity of humans [ 36 ].

Animal rights advocates strongly argue that the moral status of non-human animals is similar to that of humans, and that animals are entitled to equality of treatment. In this view, animals should be treated with the same level of respect as humans, and no one should have the right to force them into any service or to kill them or use them for their own goals. One aspect of this argument claims that moral status depends upon the capacity to suffer or enjoy life [ 37 ].

In terms of suffering and the capacity of enjoying life, many animals are not very different from human beings, as they can feel pain and experience pleasure [ 38 ]. Hence, they should be given the same moral status as humans and deserve equivalent treatment. Supporters of this argument point out that according animals a lower moral status than humans is a type of prejudice known as “speciesism” [ 38 ]. Among humans, it is widely accepted that being a part of a specific race or of a specific gender does not provide the right to ascribe a lower moral status to the outsiders. Many advocates of animal rights deploy the same argument, that being human does not give us sufficient grounds declare animals as being morally less significant [ 36 ].

ARGUMENTS IN FAVOR OF ANIMAL EXPERIMENTATION

Those who support animal experimentation have frequently made the argument that animals cannot be elevated to be seen as morally equal to humans [ 39 ]. Their main argument is that the use of the terms “moral status” or “morality” is debatable. They emphasize that we must not make the error of defining a quality or capacity associated with an animal by using the same adjectives used for humans [ 39 ]. Since, for the most part, animals do not possess humans’ cognitive capabilities and lack full autonomy (animals do not appear to rationally pursue specific goals in life), it is argued that therefore, they cannot be included in the moral community [ 39 ]. It follows from this line of argument that, if animals do not possess the same rights as human beings, their use in research experimentation can be considered appropriate [ 40 ]. The European and the American legislation support this kind of approach as much as their welfare is respected.

Another aspect of this argument is that the benefits to human beings of animal experimentation compensate for the harm caused to animals by these experiments.

In other words, animal harm is morally insignificant compared to the potential benefits to humans. Essentially, supporters of animal experimentation claim that human beings have a higher moral status than animals and that animals lack certain fundamental rights accorded to humans. The potential violations of animal rights during animal research are, in this way, justified by the greater benefits to mankind [ 40 , 41 ]. A way to evaluate when the experiments are morally justified was published in 1986 by Bateson, which developed the Bateson’s Cube [ 42 ]. The Cube has three axes: suffering, certainty of benefit and quality of research. If the research is high-quality, beneficial, and not inflicting suffering, it will be acceptable. At the contrary, painful, low-quality research with lower likelihood of success will not be acceptable [ 42 , 43 ].

Impact of experimentations on animals

Ability to feel pain and distress.

Like humans, animal have certain physical as well as psychological characteristics that make their use for experimentation controversial [ 44 ].

In the last few decades, many studies have increased knowledge of animal awareness and sentience: they indicate that animals have greater potential to experience damage than previously appreciated and that current rights and protections need to be reconsidered [ 45 ]. In recent times, scientists as well as ethicists have broadly acknowledged that animals can also experience distress and pain [ 46 ]. Potential sources of such harm arising from their use in research include disease, basic physiological needs deprivation and invasive procedures [ 46 ]. Moreover, social deprivation and lack of the ability to carry out their natural behaviors are other causes of animal harm [ 46 ]. Several studies have shown that, even in response to very gentle handling and management, animals can show marked alterations in their physiological and hormonal stress markers [ 47 ].

In spite of the fact that suffering and pain are personalized experiences, several multi-disciplinary studies have provided clear evidence of animals experiencing pain and distress. In particular, some animal species have the ability to express pain similarly to human due to common psychological, neuroanatomical and genetic characteristics [ 48 ]. Similarly, animals share a resemblance to humans in their developmental, genetic and environmental risk factors for psychopathology. For instance, in many species, it has been shown that fear operates within a less organized subcortical neural circuit than pain [ 49 , 50 ]. Various types of depression and anxiety disorders like posttraumatic stress disorder have also been reported in mammals [ 51 ].

PSYCHOLOGICAL CAPABILITIES OF ANIMALS

Some researchers have suggested that besides their ability to experience physical and psychological pain and distress, some animals also exhibit empathy, self-awareness and language-like capabilities. They also demonstrate tools-linked cognizance, pleasure-seeking and advanced problem-solving skills [ 52 ]. Moreover, mammals and birds exhibit playful behavior, an indicator of the capacity to experience pleasure. Other taxa such as reptiles, cephalopods and fishes have also been observed to display playful behavior, therefore the current legislation prescribes the use of environmental enrichers [ 53 ]. The presence of self-awareness ability, as assessed by mirror self-recognition, has been reported in magpies, chimpanzees and other apes, and certain cetaceans [ 54 ]. Recently, another study has revealed that crows have the ability to create and use tools that involve episodic-like memory formation and its retrieval. From these findings, it may be suggested that crows as well as related species show evidence of flexible learning strategies, causal reasoning, prospection and imagination that are similar to behavior observed in great apes [ 55 ]. In the context of resolving the ethical dilemmas about animal experimentation, these observations serve to highlight the challenges involved [ 56 , 57 ].

Ethics, principles and legislation in animal experimentation

Ethics in animal experimentation.

Legislation around animal research is based on the idea of the moral acceptability of the proposed experiments under specific conditions [ 58 ]. The significance of research ethics that ensures proper treatment of experimental animals [ 58 ]. To avoid undue suffering of animals, it is important to follow ethical considerations during animal studies [ 1 ]. It is important to provide best human care to these animals from the ethical and scientific point of view [ 1 ]. Poor animal care can lead to experimental outcomes [ 1 ]. Thus, if experimental animals mistreated, the scientific knowledge and conclusions obtained from experiments may be compromised and may be difficult to replicate, a hallmark of scientific research [ 1 ]. At present, most ethical guidelines work on the assumption that animal experimentation is justified because of the significant potential benefits to human beings. These guidelines are often permissive of animal experimentation regardless of the damage to the animal as long as human benefits are achieved [ 59 ].

PRINCIPLE OF THE 4 RS

Although animal experimentation has resulted in many discoveries and helped in the understanding numerous aspects of biological science, its use in various sectors is strictly controlled. In practice, the proposed set of animal experiments is usually considered by a multidisciplinary Ethics Committee before work can commence [ 60 ]. This committee will review the research protocol and make a judgment as to its sustainability. National and international laws govern the utilization of animal experimentation during research and these laws are mostly based on the universal doctrine presented by Russell and Burch (1959) known as principle of the 3 Rs. The 3Rs referred to are Reduction, Refinement and Replacement, and are applied to protocols surrounding the use of animals in research. Some researchers have proposed another “R”, of responsibility for the experimental animal as well as for the social and scientific status of the animal experiments [ 61 ]. Thus, animal ethics committees commonly review research projects with reference to the 4 Rs principles [ 62 ].

The first “R”, Reduction means that the experimental design is examined to ensure that researchers have reduced the number of experimental animals in a research project to the minimum required for reliable data [ 59 ]. Methods used for this purpose include improved experimental design, extensive literature search to avoid duplication of experiments [ 35 ], use of advanced imaging techniques, sharing resources and data, and appropriate statistical data analysis that reduce the number of animals needed for statistically significant results [ 2 , 63 ].

The second “R”, Refinement involves improvements in procedure that minimize the harmful effects of the proposed experiments on the animals involved, such as reducing pain, distress and suffering in a manner that leads to a general improvement in animal welfare. This might include for example improved living conditions for research animals, proper training of people handling animals, application of anesthesia and analgesia when required and the need for euthanasia of the animals at the end of the experiment to curtail their suffering [ 63 ].

The third “R”, Replacement refers to approaches that replace or avoid the use of experimental animals altogether. These approaches involve use of in silico methods/computerized techniques/software and in vitro methods like cell and tissue culture testing, as well as relative replacement methods by use of invertebrates like nematode worms, fruit flies and microorganisms in place of vertebrates and higher animals [ 1 ]. Examples of proper application of these first “3R2 principles are the use of alternative sources of blood, the exploitation of commercially used animals for scientific research, a proper training without use of animals and the use of specimen from previous experiments for further researches [ 64-67 ].

The fourth “R”, Responsibility refers to concerns around promoting animal welfare by improvements in experimental animals’ social life, development of advanced scientific methods for objectively determining sentience, consciousness, experience of pain and intelligence in the animal kingdom, as well as effective involvement in the professionalization of the public discussion on animal ethics [ 68 ].

OTHER ASPECTS OF ANIMAL RESEARCH ETHICS

Other research ethics considerations include having a clear rationale and reasoning for the use of animals in a research project. Researchers must have reasonable expectation of generating useful data from the proposed experiment. Moreover, the research study should be designed in such a way that it should involve the lowest possible sample size of experimental animals while producing statistically significant results [ 35 ].

All individual researchers that handle experimental animals should be properly trained for handling the particular species involved in the research study. The animal’s pain, suffering and discomfort should be minimized [ 69 ]. Animals should be given proper anesthesia when required and surgical procedures should not be repeated on same animal whenever possible [ 69 ]. The procedure of humane handling and care of experimental animals should be explicitly detailed in the research study protocol. Moreover, whenever required, aseptic techniques should be properly followed [ 70 ]. During the research, anesthetization and surgical procedures on experimental animals should only be performed by professionally skilled individuals [ 69 ].

The Animal Research: Reporting of In Vivo Experiments (ARRIVE) guidelines that are issued by the National Center for the Replacement, Refinement, and Reduction of Animals in Research (NC3Rs) are designed to improve the documentation surrounding research involving experimental animals [ 70 ]. The checklist provided includes the information required in the various sections of the manuscript i.e. study design, ethical statements, experimental procedures, experimental animals and their housing and husbandry, and more [ 70 ].

It is critical to follow the highest ethical standards while performing animal experiments. Indeed, most of the journals refuse to publish any research data that lack proper ethical considerations [ 35 ].

INVESTIGATORS’ ETHICS

Since animals have sensitivity level similar to the human beings in terms of pain, anguish, survival instinct and memory, it is the responsibility of the investigator to closely monitor the animals that are used and identify any sign of distress [ 71 ]. No justification can rationalize the absence of anesthesia or analgesia in animals that undergo invasive surgery during the research [ 72 ]. Investigators are also responsible for giving high-quality care to the experimental animals, including the supply of a nutritious diet, easy water access, prevention of and relief from any pain, disease and injury, and appropriate housing facilities for the animal species [ 73 ]. A research experiment is not permitted if the damage caused to the animal exceeds the value of knowledge gained by that experiment. No scientific advancement based on the destruction and sufferings of another living being could be justified. Besides ensuring the welfare of animals involved, investigators must also follow the applicable legislation [ 74 , 75 ].

To promote the comfort of experimental animals in England, an animal protection society named: ‘The Society for the Preservation of Cruelty to Animals’ (now the Royal Society for the Prevention of Cruelty to Animals) was established (1824) that aims to prevent cruelty to animal [ 76 ].

ANIMAL WELFARE LAWS

Legislation for animal protection during research has long been established. In 1876 the British Parliament sanctioned the ‘Cruelty to Animals Act’ for animal protection. Russell and Burch (1959) presented the ‘3 Rs’ principles: Replacement, Reduction and Refinement, for use of animals during research [ 61 ]. Almost seven years later, the U.S.A also adopted regulations for the protection of experimental animals by enacting the Laboratory Animal Welfare Act of 1966 [ 60 ]. In Brazil, the Arouca Law (Law No. 11,794/08) regulates the animal use in scientific research experiments [ 76 ].

These laws define the breeding conditions, and regulate the use of animals for scientific research and teaching purposes. Such legal provisions control the use of anesthesia, analgesia or sedation in experiments that could cause distress or pain to experimental animals [ 59 , 76 ]. These laws also stress the need for euthanasia when an experiment is finished, or even during the experiment if there is any intense suffering for the experimental animal [ 76 ].

Several national and international organizations have been established to develop alternative techniques so that animal experimentation can be avoided, such as the UK-based National Centre for the Replacement, Refinement and Reduction of Animals in Research (NC3Rs) ( www.caat.jhsph.edu ), the European Centre for the Validation of Alternative Methods (ECVAM) [ 77 ], the Universities Federation for Animal Welfare (UFAW) ( www.ufaw.org.uk ), The Interagency Coordinating Committee on the Validation of Alternative Methods (ICCVAM) [ 78 ], and The Center for Alternatives to Animal Testing (CAAT) ( www.caat.jhsph.edu ). The Brazilian ‘Arouca Law’ also constitutes a milestone, as it has created the ‘National Council for the Control of Animal Experimentation’ (CONCEA) that deals with the legal and ethical issues related to the use of experimental animals during scientific research [ 76 ].

Although national as well as international laws and guidelines have provided basic protections for experimental animals, the current regulations have some significant discrepancies. In the U.S., the Animal Welfare Act excludes rats, mice and purpose-bred birds, even though these species comprise almost 90% of the animals that are used for research purpose [ 79 ]. On the other hand, certain cats and dogs are getting special attention along with extra protection. While the U.S. Animal Welfare Act ignores birds, mice and rats, the U.S. guidelines that control research performed using federal funding ensure protections for all vertebrates [ 79 , 80 ].

Living conditions of animals

Choice of the animal model.

Based on all the above laws and regulations and in line with the deliberations of ethical committees, every researcher must follow certain rules when dealing with animal models.

Before starting any experimental work, thorough research should be carried out during the study design phase so that the unnecessary use of experimental animals is avoided. Nevertheless, certain research studies may have compelling reasons for the use of animal models, such as the investigation of human diseases and toxicity tests. Moreover, animals are also widely used in the training of health professionals as well as in training doctors in surgical skills [ 1 , 81 ].

Researcher should be well aware of the specific traits of the animal species they intend to use in the experiment, such as its developmental stages, physiology, nutritional needs, reproductive characteristics and specific behaviors. Animal models should be selected on the basis of the study design and the biological relevance of the animal [ 1 ].

Typically, in early research, non-mammalian models are used to get rapid insights into research problems such as the identification of gene function or the recognition of novel therapeutic options. Thus, in biomedical and biological research, among the most commonly used model organisms are the Zebrafish, the fruit fly Drosophila melanogaster and the nematode Caenorhabditis elegans . The main advantage of these non-mammalian animal models is their prolific reproducibility along with their much shorter generation time. They can be easily grown in any laboratory setting, are less expensive than the murine animal models and are somewhat more powerful than the tissue and cell culture approaches [ 82 ].

Caenorhabditis elegans is a small-sized nematode with a short life cycle and that exists in large populations and is relatively inexpensive to cultivate. Scientists have gathered extensive knowledge of the genomics and genetics of Caenorhabditis elegans ; but Caenorhabditis elegans models, while very useful in some respects, are unable to represent all signaling pathways found in humans. Furthermore, due to its short life cycle, scientists are unable to investigate long term effects of test compounds or to analyze primary versus secondary effects [ 6 ].

Similarly, the fruit fly Drosophila melanogaster has played a key role in numerous biomedical discoveries. It is small in size, has a short life cycle and large population size, is relatively inexpensive to breed, and extensive genomics and genetics information is available [ 6 ]. However, its respiratory, cardiovascular and nervous systems differ considerably from human beings. In addition, its immune system is less developed when compared to vertebrates, which is why effectiveness of a drug in Drosophila melanogaster may not be easily extrapolated to humans [ 83 ].

The Zebrafish ( Danio rerio ) is a small freshwater teleost, with transparent embryos, providing easy access for the observation of organogenesis and its manipulation. Therefore, Zebrafish embryos are considered good animal models for different human diseases like tuberculosis and fetal alcohol syndrome and are useful as neurodevelopmental research models. However, Zebrafish has very few mutant strains available, and its genome has numerous duplicate genes making it impossible to create knockout strains, since disrupting one copy of the gene will not disrupt the second copy of that gene. This feature limits the use of Zebrafish as animal models to study human diseases. Additionally they are rather expensive, have long life cycle, and genomics and genetics studies are still in progress [ 82 , 84 ].

Thus, experimentation on these three animals might not be equivalent to experimentation on mammals. Mammalian animal model are most similar to human beings, so targeted gene replacement is possible. Traditionally, mammals like monkey and mice have been the preferred animal models for biomedical research because of their evolutionary closeness to humans. Rodents, particularly mice and rats, are the most frequently used animal models for scientific research. Rats are the most suitable animal model for the study of obesity, shock, peritonitis, sepsis, cancer, intestinal operations, spleen, gastric ulcers, mononuclear phagocytic system, organ transplantations and wound healing. Mice are more suitable for studying burns, megacolon, shock, cancer, obesity, and sepsis as mentioned previously [ 85 ].

Similarly, pigs are mostly used for stomach, liver and transplantation studies, while rabbits are suitable for the study of immunology, inflammation, vascular biology, shock, colitis and transplantations. Thus, the choice of experimental animal mainly depends upon the field of scientific research under consideration [ 1 ].

HOUSING AND ENVIRONMENTAL ENRICHMENT

Researchers should be aware of the environment and conditions in which laboratory animals are kept during research, and they also need to be familiar with the metabolism of the animals kept in vivarium, since their metabolism can easily be altered by different factors such as pain, stress, confinement, lack of sunlight, etc. Housing conditions alter animal behavior, and this can in turn affect experimental results. By contrast, handling procedures that feature environmental enrichment and enhancement help to decrease stress and positively affect the welfare of the animals and the reliability of research data [ 74 , 75 ].

In animals, distress- and agony-causing factors should be controlled or eliminated to overcome any interference with data collection as well as with interpretation of the results, since impaired animal welfare leads to more animal usage during experiment, decreased reliability and increased discrepancies in results along with the unnecessary consumption of animal lives [ 86 ].

To reduce the variation or discrepancies in experimental data caused by various environmental factors, experimental animals must be kept in an appropriate and safe place. In addition, it is necessary to keep all variables like humidity, airflow and temperature at levels suitable for those species, as any abrupt variation in these factors could cause stress, reduced resistance and increased susceptibility to infections [ 74 ].

The space allotted to experimental animals should permit them free movement, proper sleep and where feasible allow for interaction with other animals of the same species. Mice and rats are quite sociable animals and must, therefore, be housed in groups for the expression of their normal behavior. Usually, laboratory cages are not appropriate for the behavioral needs of the animals. Therefore, environmental enrichment is an important feature for the expression of their natural behavior that will subsequently affect their defense mechanisms and physiology [ 87 ].

The features of environmental enrichment must satisfy the animals’ sense of curiosity, offer them fun activities, and also permit them to fulfill their behavioral and physiological needs. These needs include exploring, hiding, building nests and gnawing. For this purpose, different things can be used in their environment, such as PVC tubes, cardboard, igloos, paper towel, cotton, disposable masks and paper strips [ 87 ].

The environment used for housing of animals must be continuously controlled by appropriate disinfection, hygiene protocols, sterilization and sanitation processes. These steps lead to a reduction in the occurrence of various infectious agents that often found in vivarium, such as Sendai virus, cestoda and Mycoplasma pulmonis [ 88 ].

Euthanasia is a term derived from Greek, and it means a death without any suffering. According to the Brazilian Arouca Law (Article 14, Chapter IV, Paragraphs 1 and 2), an animal should undergo euthanasia, in strict compliance with the requirements of each species, when the experiment ends or during any phase of the experiment, wherever this procedure is recommended and/or whenever serious suffering occurs. If the animal does not undergo euthanasia after the intervention it may leave the vivarium and be assigned to suitable people or to the animal protection bodies, duly legalized [ 1 ].

Euthanasia procedures must result in instant loss of consciousness which leads to respiratory or cardiac arrest as well as to complete brain function impairment. Another important aspect of this procedure is calm handling of the animal while taking it out of its enclosure, to reduce its distress, suffering, anxiety and fear. In every research project, the study design should include the details of the appropriate endpoints of these experimental animals, and also the methods that will be adopted. It is important to determine the appropriate method of euthanasia for the animal being used. Another important point is that, after completing the euthanasia procedure, the animal’s death should be absolutely confirmed before discarding their bodies [ 87 , 89 ].

Relevance of animal experimentations and possible alternatives

Relevance of animal experiments and their adverse effects on human health.

One important concern is whether human diseases, when inflicted on experimental animals, adequately mimic the progressions of the disease and the treatment responses observed in humans. Several research articles have made comparisons between human and animal data, and indicated that the results of animals’ research could not always be reliably replicated in clinical research among humans. The latest systematic reviews about the treatment of different clinical conditions including neurology, vascular diseases and others, have established that the results of animal studies cannot properly predict human outcomes [ 59 , 90 ].

At present, the reliability of animal experiments for extrapolation to human health is questionable. Harmful effects may occur in humans because of misleading results from research conducted on animals. For instance, during the late fifties, a sedative drug, thalidomide, was prescribed for pregnant women, but some of the women using that drug gave birth to babies lacking limbs or with foreshortened limbs, a condition called phocomelia. When thalidomide had been tested on almost all animal models such as rats, mice, rabbits, dogs, cats, hamsters, armadillos, ferrets, swine, guinea pig, etc., this teratogenic effect was observed only occasionally [ 91 ]. Similarly, in 2006, the compound TGN 1412 was designed as an immunomodulatory drug, but when it was injected into six human volunteer, serious adverse reactions were observed resulting from a deadly cytokine storm that in turn led to disastrous systemic organ failure. TGN 1412 had been tested successfully in rats, mice, rabbits, and non-human primates [ 92 ]. Moreover, Bailey (2008) reported 90 HIV vaccines that had successful trial results in animals but which failed in human beings [ 93 ]. Moreover, in Parkinson disease, many therapeutic options that have shown promising results in rats and non-human primate models have proved harmful in humans. Hence, to analyze the relevance of animal research to human health, the efficacy of animal experimentation should be examined systematically [ 94 , 95 ]. At the same time, the development of hyperoxaluria and renal failure (up to dialysis) after ileal-jejunal bypass was unexpected because this procedure was not preliminarily evaluated on an animal model [ 96 ].

Several factors play a role in the extrapolation of animal-derived data to humans, such as environmental conditions and physiological parameters related to stress, age of the experimental animals, etc. These factors could switch on or off genes in the animal models that are specific to species and/or strains. All these observations challenge the reliability and suitability of animal experimentation as well as its objectives with respect to human health [ 76 , 92 ].

ALTERNATIVE TO ANIMAL EXPERIMENTATION/DEVELOPMENT OF NEW PRODUCTS AND TECHNIQUES TO AVOID ANIMAL SACRIFICE IN RESEARCH

Certainly, in vivo animal experimentation has significantly contributed to the development of biological and biomedical research. However it has the limitations of strict ethical issues and high production cost. Some scientists consider animal testing an ineffective and immoral practice and therefore prefer alternative techniques to be used instead of animal experimentation. These alternative methods involve in vitro experiments and ex vivo models like cell and tissue cultures, use of plants and vegetables, non-invasive human clinical studies, use of corpses for studies, use of microorganisms or other simpler organism like shrimps and water flea larvae, physicochemical techniques, educational software, computer simulations, mathematical models and nanotechnology [ 97 ]. These methods and techniques are cost-effective and could efficiently replace animal models. They could therefore, contribute to animal welfare and to the development of new therapies that can identify the therapeutics and related complications at an early stage [ 1 ].

The National Research Council (UK) suggested a shift from the animal models toward computational models, as well as high-content and high-throughput in vitro methods. Their reports highlighted that these alternative methods could produce predictive data more affordably, accurately and quickly than the traditional in vivo or experimental animal methods [ 98 ].

Increasingly, scientists and the review boards have to assess whether addressing a research question using the applied techniques of advanced genetics, molecular, computational and cell biology, and biochemistry could be used to replace animal experiments [ 59 ]. It must be remembered that each alternative method must be first validated and then registered in dedicated databases.

An additional relevant concern is how precisely animal data can mirror relevant epigenetic changes and human genetic variability. Langley and his colleagues have highlighted some of the examples of existing and some emerging non-animal based research methods in the advanced fields of neurology, orthodontics, infectious diseases, immunology, endocrine, pulmonology, obstetrics, metabolism and cardiology [ 99 ].

IN SILICO SIMULATIONS AND INFORMATICS

Several computer models have been built to study cardiovascular risk and atherosclerotic plaque build-up, to model human metabolism, to evaluate drug toxicity and to address other questions that were previously approached by testing in animals [ 100 ].

Computer simulations can potentially decrease the number of experiments required for a research project, however simulations cannot completely replace laboratory experiments. Unfortunately, not all the principles regulating biological systems are known, and computer simulation provide only an estimation of possible effects due to the limitations of computer models in comparison with complex human tissues. However, simulation and bio-informatics are now considered essential in all fields of science for their efficiency in using the existing knowledge for further experimental designs [ 76 ].

At present, biological macromolecules are regularly simulated at various levels of detail, to predict their response and behavior under certain physical conditions, chemical exposures and stimulations. Computational and bioinformatic simulations have significantly reduced the number of animals sacrificed during drug discovery by short listing potential candidate molecules for a drug. Likewise, computer simulations have decreased the number of animal experiments required in other areas of biological science by efficiently using the existing knowledge. Moreover, the development of high definition 3D computer models for anatomy with enhanced level of detail, it may make it possible to reduce or eliminate the need for animal dissection during teaching [ 101 , 102 ].

3D CELL-CULTURE MODELS AND ORGANS-ON-CHIPS

In the current scenario of rapid advancement in the life sciences, certain tissue models can be built using 3D cell culture technology. Indeed, there are some organs on micro-scale chip models used for mimicking the human body environment. 3D models of multiple organ systems such as heart, liver, skin, muscle, testis, brain, gut, bone marrow, lungs and kidney, in addition to individual organs, have been created in microfluidic channels, re-creating the physiological chemical and physical microenvironments of the body [ 103 ]. These emerging techniques, such as the biomedical/biological microelectromechanical system (Bio-MEMS) or lab-on-a-chip (LOC) and micro total analysis systems (lTAS) will, in the future, be a useful substitute for animal experimentation in commercial laboratories in the biotechnology, environmental safety, chemistry and pharmaceutical industries. For 3D cell culture modeling, cells are grown in 3D spheroids or aggregates with the help of a scaffold or matrix, or sometimes using a scaffold-free method. The 3D cell culture modeling conditions can be altered to add proteins and other factors that are found in a tumor microenvironment, for example, or in particular tissues. These matrices contain extracellular matrix components such as proteins, glycoconjugates and glycosaminoglycans that allow for cell communication, cell to cell contact and the activation of signaling pathways in such a way that the morphological and functional differentiation of these cells can accurately mimic their environment in vivo . This methodology, in time, will bridge the gap between in vivo and in vitro drug screening, decreasing the utilization of animal models during research [ 104 ].

ALTERNATIVES TO MICROBIAL CULTURE MEDIA AND SERUM-FREE ANIMAL CELL CULTURES

There are moves to reduce the use of animal derived products in many areas of biotechnology. Microbial culture media peptones are mostly made by the proteolysis of farmed animal meat. However, nowadays, various suppliers provide peptones extracted from yeast and plants. Although the costs of these plant-extracted peptones are the same as those of animal peptones, plant peptones are more environmentally favorable since less plant material and water are required for them to grow, compared with the food grain and fodder needed for cattle that are slaughtered for animal peptone production [ 105 ].

Human cell culture is often carried out in a medium that contains fetal calf serum, the production of which involves animal (cow) sacrifice or suffering. In fact, living pregnant cows are used and their fetuses removed to harvest the serum from the fetal blood. Fetal calf serum is used because it is a natural medium rich in all the required nutrients and significantly increases the chances of successful cell growth in culture. Scientists are striving to identify the factors and nutrients required for the growth of various types of cells, with a view to eliminating the use of calf serum. At present, most cell lines could be cultured in a chemically-synthesized medium without using animal products. Furthermore, data from chemically-synthesized media experiments may have better reproducibility than those using animal serum media, since the composition of animal serum does change from batch to batch on the basis of animals’ gender, age, health and genetic background [ 76 ].

ALTERNATIVES TO ANIMAL-DERIVED ANTIBODIES

Animal friendly affinity reagents may act as an alternative to antibodies produced, thereby removing the need for animal immunization. Typically, these antibodies are obtained in vitro by yeast, phage or ribosome display. In a recent review, a comparative analysis between animal friendly affinity reagents and animal derived-antibodies showed that the affinity reagents have superior quality, are relatively less time consuming, have more reproducibility and are more reliable and are cost-effective [ 106 , 107 ].

Conclusions

Animal experimentation led to great advancement in biological and biomedical sciences and contributed to the discovery of many drugs and treatment options. However, such experimentation may cause harm, pain and distress to the animals involved. Therefore, to perform animal experimentations, certain ethical rules and laws must be strictly followed and there should be proper justification for using animals in research projects. Furthermore, during animal experimentation the 4 Rs principles of reduction, refinement, replacement and responsibility must be followed by the researchers. Moreover, before beginning a research project, experiments should be thoroughly planned and well-designed, and should avoid unnecessary use of animals. The reliability and reproducibility of animal experiments should also be considered. Whenever possible, alternative methods to animal experimentation should be adopted, such as in vitro experimentation, cadaveric studies, and computer simulations.

While much progress has been made on reducing animal experimentation there is a need for greater awareness of alternatives to animal experiments among scientists and easier access to advanced modeling technologies. Greater research is needed to define a roadmap that will lead to the elimination of all unnecessary animal experimentation and provide a framework for adoption of reliable alternative methodologies in biomedical research.

Acknowledgements

This research was funded by the Provincia Autonoma di Bolzano in the framework of LP 15/2020 (dgp 3174/2021).

Conflicts of interest statement

Authors declare no conflict of interest.

Author's contributions

MB: study conception, editing and critical revision of the manuscript; AKK, DP, GH, RB, Paul S, Peter S, RM, BF, NC, SM, LL, DD, GMT, MCE, MD, SM, Daniele M, GB, AD, KD, MCM, TB, MS, STC, Donald M, AM, AB, KLH, MK, LS, LL, GF: literature search, editing and critical revision of the manuscript. All authors have read and approved the final manuscript.

Contributor Information

INTERNATIONAL BIOETHICS STUDY GROUP : Derek Pheby , Gary Henehan , Richard Brown , Paul Sieving , Peter Sykora , Robert Marks , Benedetto Falsini , Natale Capodicasa , Stanislav Miertus , Lorenzo Lorusso , Gianluca Martino Tartaglia , Mahmut Cerkez Ergoren , Munis Dundar , Sandro Michelini , Daniele Malacarne , Tommaso Beccari , Michele Samaja , Matteo Bertelli , Donald Martin , Assunta Morresi , Ariola Bacu , Karen L. Herbst , Mykhaylo Kapustin , Liborio Stuppia , Ludovica Lumer , and Giampietro Farronato

Ethical care for research animals

WHY ANIMAL RESEARCH?

The use of animals in some forms of biomedical research remains essential to the discovery of the causes, diagnoses, and treatment of disease and suffering in humans and in animals., stanford shares the public's concern for laboratory research animals..

Many people have questions about animal testing ethics and the animal testing debate. We take our responsibility for the ethical treatment of animals in medical research very seriously. At Stanford, we emphasize that the humane care of laboratory animals is essential, both ethically and scientifically.  Poor animal care is not good science. If animals are not well-treated, the science and knowledge they produce is not trustworthy and cannot be replicated, an important hallmark of the scientific method .

There are several reasons why the use of animals is critical for biomedical research: 

••  Animals are biologically very similar to humans. In fact, mice share more than 98% DNA with us!

••  Animals are susceptible to many of the same health problems as humans – cancer, diabetes, heart disease, etc.

••  With a shorter life cycle than humans, animal models can be studied throughout their whole life span and across several generations, a critical element in understanding how a disease processes and how it interacts with a whole, living biological system.

The ethics of animal experimentation

Nothing so far has been discovered that can be a substitute for the complex functions of a living, breathing, whole-organ system with pulmonary and circulatory structures like those in humans. Until such a discovery, animals must continue to play a critical role in helping researchers test potential new drugs and medical treatments for effectiveness and safety, and in identifying any undesired or dangerous side effects, such as infertility, birth defects, liver damage, toxicity, or cancer-causing potential.

U.S. federal laws require that non-human animal research occur to show the safety and efficacy of new treatments before any human research will be allowed to be conducted.  Not only do we humans benefit from this research and testing, but hundreds of drugs and treatments developed for human use are now routinely used in veterinary clinics as well, helping animals live longer, healthier lives.

It is important to stress that 95% of all animals necessary for biomedical research in the United States are rodents – rats and mice especially bred for laboratory use – and that animals are only one part of the larger process of biomedical research.

Our researchers are strong supporters of animal welfare and view their work with animals in biomedical research as a privilege.

Stanford researchers are obligated to ensure the well-being of all animals in their care..

Stanford researchers are obligated to ensure the well-being of animals in their care, in strict adherence to the highest standards, and in accordance with federal and state laws, regulatory guidelines, and humane principles. They are also obligated to continuously update their animal-care practices based on the newest information and findings in the fields of laboratory animal care and husbandry.  

Researchers requesting use of animal models at Stanford must have their research proposals reviewed by a federally mandated committee that includes two independent community members.  It is only with this committee’s approval that research can begin. We at Stanford are dedicated to refining, reducing, and replacing animals in research whenever possible, and to using alternative methods (cell and tissue cultures, computer simulations, etc.) instead of or before animal studies are ever conducted.

Organizations and Resources

There are many outreach and advocacy organizations in the field of biomedical research.

• Learn more about outreach and advocacy organizations

Stanford Discoveries

What are the benefits of using animals in research? Stanford researchers have made many important human and animal life-saving discoveries through their work. 

• Learn more about research discoveries at Stanford

EDITORIAL article

This article is part of the research topic.

Mechanical Ventilation in Anesthesia and Critical Care Animal Patients, Volume II

Editorial: Mechanical Ventilation in Anesthesia and Critical Care Animal Patients, Volume II Provisionally Accepted

• 1 University of São Paulo, Brazil

The final, formatted version of the article will be published soon.

Mechanical ventilation, a cornerstone of modern veterinary medicine, has evolved significantly. It has become an indispensable tool in ensuring the safety and stability of animals during surgery. It plays a crucial role in intensive care units, where it aids in the recovery of critically ill patients. The intricacies and nuances of mechanical ventilation are often overlooked, yet they hold the key to the successful management of anesthesia and the survival of animals in critical conditions.In this editorial, we investigate mechanical ventilation during animal anesthesia or intensive care, exploring its significance, advancements, and challenges that veterinarians must weigh. We aim to shed light on this technology's pivotal role in enhancing the quality of care provided to animals and the constant pursuit of refinement to minimize potential risks and optimize outcomes. As the field of veterinary medicine continues to advance, we must maintain a comprehensive understanding of the intricacies of mechanical ventilation. Through this exploration, we aim to underscore the importance of this life-saving technique and inspire a continued commitment to excellence in animal anesthesia and intensive care. This Research Topic presents four new papers that illuminate these issues in mechanical ventilation in horses and dogs. Horses undergoing general inhalation anesthesia often present complications related to the decubitus position in which they are lying on the operating table. Such complications are related to difficulties in gas exchange due to a decrease in the ventilation/perfusion ratio, pulmonary atelectasis, and a drop in blood pressure. Lung atelectasis in horses is produced mainly due to dorsal or lateral decubitus. In dorsal decubitus, the lungs receive compression from the diaphragm produced by the compression of the abdominal viscera 1 . In lateral decubitus, the upper lung compresses the mediastinum and, consequently, the lower lung. Due to the loss of functional areas of the lungs, there is a drop in gas exchange, causing a reduction in the partial pressure of arterial oxygen and an increase in the partial pressure of arterial carbon dioxide, impairing cellular processes. Alveolar recruitment maneuvers (ARMs) reverse atelectasis, and positive endexpiratory pressure 2 keeps the alveoli open. However, they are not free from side effects, including barotrauma, volutrauma, and atelectrauma, and monitoring is essential. The evaluation can be done through imaging tests such as computed tomography (CT) in humans and small animals. However, it is only possible in horses through electrical impedance tomography 3; 4 , respiratory mechanics, or arterial oxygenation through blood gas analysis. Therefore, Sacks et al. present a study comparing the ventilation distribution measured by electrical impedance tomography (EIT) in foals under diazepam sedation, postural changes, and continuous positive airway pressure (CPAP). Specific spirometry data and F-shunt calculation were also assessed to support the interpretation of EIT variables. They verified that in healthy foals, diazepam administration did not alter the distribution of ventilation or minute ventilation, and the lateral recumbency results in the collapse of dependent lung areas. The CPAP use in dorsal recumbency foals increases pulmonary pressures and improves ventilation in dependent regions, suggesting improvement of ventilation-perfusion mismatch. These findings will help anesthesiologists and intensivists understand what happens in these animals and how to improve ventilation in sedated and lateral recumbent foals. In adult horses, Brandly et al. studied the flow-controlled expiration technique (FLEX) during anesthesia to reduce PEEP requirement in dorsally recumbent. They observed that FLEX ventilation was associated with a lower PEEP requirement due to a more homogenous lung ventilation distribution during expiration. This lower PEEP requirement led to more stable and improved cardiovascular conditions in horses ventilated with FLEX. This study makes an essential contribution to the anesthesia and ventilation of horses as it presents a strategy to treat intraoperative hypoxemia and protect the lungs using lower PEEP to maintain alveolar recruitment. In dogs, alveolar recruitment is needed to reverse pulmonary atelectasis. Likewise, ARMs can cause lung damage and can be monitored by CT 5; 6 , a gold standard method, in addition to EIT, ventilatory mechanics, and blood gas analysis. The lung protection strategy should also employ low tidal volumes and PEEP. Sanchez et al. studied dogs submitted to a stepwise ARM and monitored lung volume distribution by CT. They verified that the CT showed maximum pulmonary aeration distribution by PEEP titration, which occurred at PEEP 20 cmH2O and maintained the lungs normoaerated and without hyperaeration. However, based on the best static compliance and driving pressure associated with the absence of hemodynamic changes, the best PEEP value to keep the alveoli open after ARMs was PEEP from 10 and 5 descending for this study condition. In another randomized clinical trial, Rodrigues et al. studied intraoperative protective mechanical ventilation in dogs based on 8 mL.kg -1 tidal volume, recruitment maneuvers, and PEEP. Their results showed the possibility of using volumes smaller than 10 mL. kg in dogs to protect the lungs against injuries caused by excessive volumes during mechanical ventilation. In surgeries lasting up to 1 hour, there is no need for ARMs if PEEP is maintained from the beginning at 5cmH2O. The four studies in this edition demonstrated that performing recruitment maneuvers and subsequent administration of PEEP to keep the alveoli open is an essential technique for reversing hypoxemia in horses and dogs during anesthesia or ICU. These studies have also demonstrated the importance of monitoring these to avoid lung injuries and hemodynamic dysfunctions.

Keywords: Mechanical ventilalion, Dogs, Horses, Alveolar Recruitment Maneuver(ARM), Foal

Received: 18 Mar 2024; Accepted: 16 Apr 2024.

Copyright: © 2024 Ambrósio and Fantoni. This is an open-access article distributed under the terms of the Creative Commons Attribution License (CC BY) . The use, distribution or reproduction in other forums is permitted, provided the original author(s) or licensor are credited and that the original publication in this journal is cited, in accordance with accepted academic practice. No use, distribution or reproduction is permitted which does not comply with these terms.

* Correspondence: Prof. Aline M. Ambrósio, University of São Paulo, São Paulo, Brazil

People also looked at