Leveraging artificial intelligence and software engineering methods in epidemiology for the co-creation of decision-support tools based on mechanistic models.

Epidemiological modeling is a key lever for infectious disease control and prevention on farms. It makes it possible to understand the spread of pathogens, but also to compare intervention scenarios even in counterfactual situations. However, the actual capability of decision makers to use mechanistic models to support timely interventions is limited. This study demonstrates how artificial intelligence (AI) techniques can make mechanistic epidemiological models more accessible to farmers and veterinarians, and how to transform such models into user-friendly decision-support tools (DST). By leveraging knowledge representation methods, such as the textual formalization of model components through a domain-specific language (DSL), the co-design of mechanistic models and DST becomes more efficient and collaborative. This facilitates the integration of explicit expert knowledge and practical insights into the modeling process. Furthermore, the utilization of AI and software engineering enables the automation of web application generation based on existing mechanistic models. This automation simplifies the development of DST, as tool designers can focus on identifying users' needs and specifying expected features and meaningful presentations of outcomes, instead of wasting time in writing code to wrap models into web apps. To illustrate the practical application of this approach, we consider the example of Bovine Respiratory Disease (BRD), a tough challenge in fattening farms where young beef bulls often develop BRD shortly after being allocated into pens. BRD is a multi-factorial, multi-pathogen disease that is difficult to anticipate and control, often resulting in the massive use of antimicrobials to mitigate its impact on animal health, welfare, and economic losses. The DST developed from an existing mechanistic BRD model empowers users, including farmers and veterinarians, to customize scenarios based on their specific farm conditions. It enables them to anticipate the effects of various pathogens, compare the epidemiological and economic outcomes associated with different farming practices, and decide how to balance the reduction of disease impact and the reduction of antimicrobial usage (AMU). The generic method presented in this article illustrates the potential of artificial intelligence (AI) and software engineering methods to enhance the co-creation of DST based on mechanistic models in veterinary epidemiology. The corresponding pipeline is distributed as an open-source software. By leveraging these advancements, this research aims to bridge the gap between theoretical models and the practical usage of their outcomes on the field.

Modeling the effects of farming practices on bovine respiratory disease in a multi-batch cattle fattening farm.

Bovine Respiratory Disease (BRD) affects young bulls, causing animal welfare and health concerns as well as economical costs. BRD is caused by an array of viruses and bacteria and also by environmental and abiotic factors. How farming practices influence the spread of these causal pathogens remains unclear. Our goal was to assess the impact of zootechnical practices on the spread of three causal agents of BRD, namely the bovine respiratory syncytial virus (BRSV), Mannheimia haemolytica and Mycoplasma bovis. In that extent, we used an individual based stochastic mechanistic model monitoring risk factors, infectious processes, detection and treatment in a farm possibly featuring several batches simultaneously. The model was calibrated with three sets of parameters relative to each of the three pathogens using data extracted from literature. Separated batches were found to be more effective than a unique large one for reducing the spread of pathogens, especially for BRSV and M.bovis. Moreover, it was found that allocating high risk and low risk individuals into separated batches participated in reducing cumulative incidence, epidemic peaks and antimicrobial usage, especially for M. bovis. Theses findings rise interrogations on the optimal farming practices in order to limit BRD occurrence and pave the way to models featuring coinfections and collective treatments p { line-height: 115%; margin-bottom: 0.25 cm; background: transparent}a:link { color: #000080; text-decoration: underline}a.cjk:link { so-language: zxx}a.ctl:link { solanguage: zxx}.

The DECIDE project: from surveillance data to decision-support for farmers and veterinarians.

Farmers, veterinarians and other animal health managers in the livestock sector are currently missing sufficient information on prevalence and burden of contagious endemic animal diseases. They need adequate tools for risk assessment and prioritization of control measures for these diseases. The DECIDE project develops data-driven decision-support tools, which present (i) robust and early signals of disease emergence and options for diagnostic confirmation; and (ii) options for controlling the disease along with their implications in terms of disease spread, economic burden and animal welfare. DECIDE focuses on respiratory and gastro-intestinal syndromes in the three most important terrestrial livestock species (pigs, poultry, cattle) and on reduced growth and mortality in two of the most important aquaculture species (salmon and trout). For each of these, we (i) identify the stakeholder needs; (ii) determine the burden of disease and costs of control measures; (iii) develop data sharing frameworks based on federated data access and meta-information sharing; (iv) build multivariate and multi-level models for creating early warning systems; and (v) rank interventions based on multiple criteria. Together, all of this forms decision-support tools to be integrated in existing farm management systems wherever possible and to be evaluated in several pilot implementations in farms across Europe. The results of DECIDE lead to improved use of surveillance data and evidence-based decisions on disease control. Improved disease control is essential for a sustainable food chain in Europe with increased animal health and welfare and that protects human health.

Learning and strategic imitation in modelling farmers' dynamic decisions on bovine viral diarrhoea vaccination.

Considering human decision-making is essential for understanding the mechanisms underlying the propagation of real-life diseases. We present an extension of a model for pathogen spread that considers farmers' dynamic decision-making regarding the adoption of a control measure in their own herd. Farmers can take into account the decisions and observed costs of their trade partners or of their geographic neighbours. The model and construction of such costs are adapted to the case of bovine viral diarrhoea, for which an individual-based stochastic model is considered. Simulation results suggest that obtaining information from geographic neighbours might lead to a better control of bovine viral diarrhoea than considering information from trade partners. In particular, using information from all geographic neighbours at each decision time seems to be more beneficial than considering only the information from one geographic neighbour or trade partner at each time. This study highlights the central role that social dynamics among farmers can take in the spread and control of bovine viral diarrhoea, providing insights into how public policy efforts could be targeted in order to increase voluntary vaccination uptake against this disease in endemic areas.

Mechanistic models of Rift Valley fever virus transmission: A systematic review.

Rift Valley fever (RVF) is a zoonotic arbovirosis which has been reported across Africa including the northernmost edge, South West Indian Ocean islands, and the Arabian Peninsula. The virus is responsible for high abortion rates and mortality in young ruminants, with economic impacts in affected countries. To date, RVF epidemiological mechanisms are not fully understood, due to the multiplicity of implicated vertebrate hosts, vectors, and ecosystems. In this context, mathematical models are useful tools to develop our understanding of complex systems, and mechanistic models are particularly suited to data-scarce settings. Here, we performed a systematic review of mechanistic models studying RVF, to explore their diversity and their contribution to the understanding of this disease epidemiology. Researching Pubmed and Scopus databases (October 2021), we eventually selected 48 papers, presenting overall 49 different models with numerical application to RVF. We categorized models as theoretical, applied, or grey, depending on whether they represented a specific geographical context or not, and whether they relied on an extensive use of data. We discussed their contributions to the understanding of RVF epidemiology, and highlighted that theoretical and applied models are used differently yet meet common objectives. Through the examination of model features, we identified research questions left unexplored across scales, such as the role of animal mobility, as well as the relative contributions of host and vector species to transmission. Importantly, we noted a substantial lack of justification when choosing a functional form for the force of infection. Overall, we showed a great diversity in RVF models, leading to important progress in our comprehension of epidemiological mechanisms. To go further, data gaps must be filled, and modelers need to improve their code accessibility.

Modelling the effects of antimicrobial metaphylaxis and pen size on bovine respiratory disease in high and low risk fattening cattle.

Bovine respiratory disease (BRD) dramatically affects young calves, especially in fattening facilities, and is difficult to understand, anticipate and control due to the multiplicity of factors involved in the onset and impact of this disease. In this study we aimed to compare the impact of farming practices on BRD severity and on antimicrobial usage. We designed a stochastic individual-based mechanistic BRD model which incorporates not only the infectious process, but also clinical signs, detection methods and treatment protocols. We investigated twelve contrasted scenarios which reflect farming practices in various fattening systems, based on pen sizes, risk level, and individual treatment vs. collective treatment (metaphylaxis) before or during fattening. We calibrated model parameters from existing observation data or literature and compared scenario outputs regarding disease dynamics, severity and mortality. The comparison of the trade-off between cumulative BRD duration and number of antimicrobial doses highlighted the added value of risk reduction at pen formation even in small pens, and acknowledges the interest of collective treatments for high-risk pens, with a better efficacy of treatments triggered during fattening based on the number of detected cases.

The African swine fever modelling challenge: Model comparison and lessons learnt.

Robust epidemiological knowledge and predictive modelling tools are needed to address challenging objectives, such as: understanding epidemic drivers; forecasting epidemics; and prioritising control measures. Often, multiple modelling approaches can be used during an epidemic to support effective decision making in a timely manner. Modelling challenges contribute to understanding the pros and cons of different approaches and to fostering technical dialogue between modellers. In this paper, we present the results of the first modelling challenge in animal health - the ASF Challenge - which focused on a synthetic epidemic of African swine fever (ASF) on an island. The modelling approaches proposed by five independent international teams were compared. We assessed their ability to predict temporal and spatial epidemic expansion at the interface between domestic pigs and wild boar, and to prioritise a limited number of alternative interventions. We also compared their qualitative and quantitative spatio-temporal predictions over the first two one-month projection phases of the challenge. Top-performing models in predicting the ASF epidemic differed according to the challenge phase, host species, and in predicting spatial or temporal dynamics. Ensemble models built using all team-predictions outperformed any individual model in at least one phase. The ASF Challenge demonstrated that accounting for the interface between livestock and wildlife is key to increasing our effectiveness in controlling emerging animal diseases, and contributed to improving the readiness of the scientific community to face future ASF epidemics. Finally, we discuss the lessons learnt from model comparison to guide decision making.

The African swine fever modelling challenge: Objectives, model description and synthetic data generation.

African swine fever (ASF) is an emerging disease currently spreading at the interface between wild boar and pig farms in Europe and Asia. Current disease control regulations, which involve massive culling with significant economic and animal welfare costs, need to be improved. Modelling enables relevant control measures to be explored, but conducting the exercise during an epidemic is extremely difficult. Modelling challenges enhance modellers' ability to timely advice policy makers, improve their readiness when facing emerging threats, and promote international collaborations. The ASF-Challenge, which ran between August 2020 and January 2021, was the first modelling challenge in animal health. In this paper, we describe the objectives and rules of the challenge. We then demonstrate the mechanistic multi-host model that was used to mimic as accurately as possible an ASF-like epidemic, provide a detailed explanation of the surveillance and intervention strategies that generated the synthetic data, and describe the different management strategies that were assessed by the competing modelling teams. We then outline the different technical steps of the challenge as well as its environment. Finally, we synthesize the lessons we learnt along the way to guide future modelling challenges in animal health.

Heterogeneity of Rift Valley fever virus transmission potential across livestock hosts, quantified through a model-based analysis of host viral load and vector infection.

Quantifying the variation of pathogens' life history traits in multiple host systems is crucial to understand their transmission dynamics. It is particularly important for arthropod-borne viruses (arboviruses), which are prone to infecting several species of vertebrate hosts. Here, we focus on how host-pathogen interactions determine the ability of host species to transmit a virus to susceptible vectors upon a potentially infectious contact. Rift Valley fever (RVF) is a viral, vector-borne, zoonotic disease, chosen as a case study. The relative contributions of livestock species to RVFV transmission has not been previously quantified. To estimate their potential to transmit the virus over the course of their infection, we 1) fitted a within-host model to viral RNA and infectious virus measures, obtained daily from infected lambs, calves, and young goats, 2) estimated the relationship between vertebrate host infectious titers and probability to infect mosquitoes, and 3) estimated the net infectiousness of each host species over the duration of their infectious periods, taking into account different survival outcomes for lambs. Our results indicate that the efficiency of viral replication, along with the lifespan of infectious particles, could be sources of heterogeneity between hosts. Given available data on RVFV competent vectors, we found that, for similar infectious titers, infection rates in the Aedes genus were on average higher than in the Culex genus. Consequently, for Aedes-mediated infections, we estimated the net infectiousness of lambs to be 2.93 (median) and 3.65 times higher than that of calves and goats, respectively. In lambs, we estimated the overall infectiousness to be 1.93 times higher in individuals which eventually died from the infection than in those recovering. Beyond infectiousness, the relative contributions of host species to transmission depend on local ecological factors, including relative abundances and vector host-feeding preferences. Quantifying these contributions will ultimately help design efficient, targeted, surveillance and vaccination strategies.

Modelling transmission of Mycobacterium avium subspecies paratuberculosis between Irish dairy cattle herds.

Bovine paratuberculosis is an endemic disease caused by Mycobacterium avium subspecies paratuberculosis (Map). Map is mainly transmitted between herds through movement of infected but undetected animals. Our objective was to investigate the effect of observed herd characteristics on Map spread on a national scale in Ireland. Herd characteristics included herd size, number of breeding bulls introduced, number of animals purchased and sold, and number of herds the focal herd purchases from and sells to. We used these characteristics to classify herds in accordance with their probability of becoming infected and of spreading infection to other herds. A stochastic individual-based model was used to represent herd demography and Map infection dynamics of each dairy cattle herd in Ireland. Data on herd size and composition, as well as birth, death, and culling events were used to characterize herd demography. Herds were connected with each other through observed animal trade movements. Data consisted of 13 353 herds, with 4 494 768 dairy female animals, and 72 991 breeding bulls. We showed that the probability of an infected animal being introduced into the herd increases both with an increasing number of animals that enter a herd via trade and number of herds from which animals are sourced. Herds that both buy and sell a lot of animals pose the highest infection risk to other herds and could therefore play an important role in Map spread between herds.

Dynamic resource allocation for controlling pathogen spread on a large metapopulation network.

To control the spread of an infectious disease over a large network, the optimal allocation by a social planner of a limited resource is a fundamental and difficult problem. We address this problem for a livestock disease that propagates on an animal trade network according to an epidemiological-demographic model based on animal demographics and trade data. We assume that the resource is dynamically allocated following a certain score, up to the limit of resource availability. We adapt a greedy approach to the metapopulation framework, obtaining new scores that minimize approximations of two different objective functions, for two control measures: vaccination and treatment. Through intensive simulations, we compare the greedy scores with several heuristics. Although topology-based scores can limit the spread of the disease, information on herd health status seems crucial to eradicating the disease. In particular, greedy scores are among the most effective in reducing disease prevalence, even though they do not always perform the best. However, some scores may be preferred in real life because they are easier to calculate or because they use a smaller amount of resources. The developed approach could be adapted to other epidemiological models or to other control measures in the metapopulation setting.

To Vaccinate or Not: Impact of Bovine Viral Diarrhoea in French Cow-Calf Herds.

Bovine viral diarrhoea (BVD) remains an issue despite control programs implemented worldwide. Virus introduction can occur through contacts with neighbouring herds. Vaccination can locally protect exposed herds. However, virus spread depends on herd characteristics, which may impair vaccination efficiency. Using a within-herd epidemiological model, we compared three French cow-calf farming systems named by their main breed: Charolaise, Limousine, and Blonde d'Aquitaine. We assessed vaccination strategies of breeding females assuming two possible protections: against infection or against vertical transmission. Four commercial vaccines were considered: Bovilis®, Bovela®, Rispoval®, and Mucosiffa®. We tested various virus introduction frequency in a naïve herd. We calculated BVD economic impact and vaccination reward. In Charolaise, BVD economic impact was 113€ per cow over 5 years after virus introduction. Irrespective of the vaccine and for a high enough risk of introduction, the yearly expected reward was 0.80€ per invested euro per cow. Vaccination should not be stopped before herd exposure has been decreased. In contrast, the reward was almost nil in Blonde d'Aquitaine and Limousine. This highlights the importance of accounting for herd specificities to assess BVD impact and vaccination efficiency. To guide farmers' vaccination decisions against BVD, we transformed this model into a French decision support tool.

Accounting for farmers' control decisions in a model of pathogen spread through animal trade.

Accounting for individual decisions in mechanistic epidemiological models remains a challenge, especially for unregulated endemic animal diseases for which control is not compulsory. We propose a new integrative model by combining two sub-models. The first one for the dynamics of a livestock epidemic on a metapopulation network, grounded on demographic and animal trade data. The second one for farmers' behavior regarding the adoption of a control measure against the disease spread in their herd. The measure is specified as a protective vaccine with given economic implications, and the model is numerically studied through intensive simulations and sensitivity analyses. While each tested parameter of the model has an impact on the overall model behavior, the most important factor in farmers' decisions is their frequency, as this factor explained almost 30% of the variation in decision-related outputs of the model. Indeed, updating frequently local health information impacts positively vaccination, and limits strongly the propagation of the pathogen. Our study is relevant for the understanding of the interplay between decision-related human behavior and livestock epidemic dynamics. The model can be used for other structures of epidemic models or different interventions, by adapting its components.

Research perspectives on animal health in the era of artificial intelligence.

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.

Dispersal in heterogeneous environments drives population dynamics and control of tsetse flies.

Spatio-temporally heterogeneous environments may lead to unexpected population dynamics. Knowledge is needed on local properties favouring population resilience at large scale. For pathogen vectors, such as tsetse flies transmitting human and animal African trypanosomosis, this is crucial to target management strategies. We developed a mechanistic spatio-temporal model of the age-structured population dynamics of tsetse flies, parametrized with field and laboratory data. It accounts for density- and temperature-dependence. The studied environment is heterogeneous, fragmented and dispersal is suitability-driven. We confirmed that temperature and adult mortality have a strong impact on tsetse populations. When homogeneously increasing adult mortality, control was less effective and induced faster population recovery in the coldest and temperature-stable locations, creating refuges. To optimally select locations to control, we assessed the potential impact of treating them and their contribution to the whole population. This heterogeneous control induced a similar population decrease, with more dispersed individuals. Control efficacy was no longer related to temperature. Dispersal was responsible for refuges at the interface between controlled and uncontrolled zones, where resurgence after control was very high. The early identification of refuges, which could jeopardize control efforts, is crucial. We recommend baseline data collection to characterize the ecosystem before implementing any measures.

Minimizing the number of origins in batches of weaned calves to reduce their risks of developing bovine respiratory diseases.

Bovine respiratory diseases (BRD) are a major concern for the beef cattle industry, as beef calves overwhelmingly develop BRD symptoms during the first weeks after their arrival at fattening units. These cases occur after weaned calves from various cow-calf producers are grouped into batches to be sold to fatteners. Cross-contaminations between calves from different origins (potentially carrying different pathogens), together with increased stress because of the process of batch creation, can increase their risks of developing BRD symptoms. This study investigated whether reducing the number of different origins per batch is a strategy to reduce the risk of BRD cases. We developed an algorithm aimed at creating batches with as few origins as possible, while respecting constraints on the number and breed of the calves. We tested this algorithm on a dataset of 137,726 weaned calves grouped into 9701 batches by a French organization. We also computed an index assessing the risks of developing BRD because of the batch composition by considering four pathogens involved in the BRD system. While increasing the heterogeneity of batches in calf bodyweight, which is not expected to strongly impact the performance, our algorithm successfully decreased the average number of origins in the same batch and their risk index. Both this algorithm and the risk index can be used as part of decision tool to assess and possibly minimize BRD risk at batch creation, but they are generic enough to assess health risk for other production animals, and optimize the homogeneity of selected characteristics.

It's risky to wander in September: Modelling the epidemic potential of Rift Valley fever in a Sahelian setting.

Estimating the epidemic potential of vector-borne diseases, along with the relative contribution of underlying mechanisms, is crucial for animal and human health worldwide. In West African Sahel, several outbreaks of Rift Valley fever (RVF) have occurred over the last decades, but uncertainty remains about the conditions necessary to trigger these outbreaks. We use the basic reproduction number (R0) as a measure of RVF epidemic potential in northern Senegal, and map its value in two distinct ecosystems, namely the Ferlo and the Senegal River delta and valley. We consider three consecutive rainy seasons (July-November 2014, 2015 and 2016) and account for several vector and animal species. We parametrize our model with estimates of Aedes vexans arabiensis, Culex poicilipes, Culex tritaeniorhynchus, cattle, sheep and goat abundances. The impact of RVF virus introduction is assessed every week over northern Senegal. We highlight September as the period of highest epidemic potential in northern Senegal, resulting from distinct dynamics in the two study areas. Spatially, in the seasonal environment of the Ferlo, we observe that high-risk locations vary between years. We show that decreased vector densities do not greatly reduce R0 and that cattle immunity has a greater impact on reducing transmission than small ruminant immunity. The host preferences of vectors and the temperature-dependent time interval between their blood meals are crucial parameters needing further biological investigations.

EMULSION: Transparent and flexible multiscale stochastic models in human, animal and plant epidemiology.

Stochastic mechanistic epidemiological models largely contribute to better understand pathogen emergence and spread, and assess control strategies at various scales (from within-host to transnational scale). However, developing realistic models which involve multi-disciplinary knowledge integration faces three major challenges in predictive epidemiology: lack of readability once translated into simulation code, low reproducibility and reusability, and long development time compared to outbreak time scale. We introduce here EMULSION, an artificial intelligence-based software intended to address those issues and help modellers focus on model design rather than programming. EMULSION defines a domain-specific language to make all components of an epidemiological model (structure, processes, parameters…) explicit as a structured text file. This file is readable by scientists from other fields (epidemiologists, biologists, economists), who can contribute to validate or revise assumptions at any stage of model development. It is then automatically processed by EMULSION generic simulation engine, preventing any discrepancy between model description and implementation. The modelling language and simulation architecture both rely on the combination of advanced artificial intelligence methods (knowledge representation and multi-level agent-based simulation), allowing several modelling paradigms (from compartment- to individual-based models) at several scales (up to metapopulation). The flexibility of EMULSION and its capability to support iterative modelling are illustrated here through examples of progressive complexity, including late revisions of core model assumptions. EMULSION is also currently used to model the spread of several diseases in real pathosystems. EMULSION provides a command-line tool for checking models, producing model diagrams, running simulations, and plotting outputs. Written in Python 3, EMULSION runs on Linux, MacOS, and Windows. It is released under Apache-2.0 license. A comprehensive documentation with installation instructions, a tutorial and many examples are available from: https://sourcesup.renater.fr/www/emulsion-public.