ORIGINAL RESEARCH published: 18 February 2020 doi: 10.3389/fvets.2020.00068 Near Real-Time Monitoring of Clinical Events Detected in Swine Herds in Northeastern Spain Ana Alba-Casals 1,2*, Eduard Allue 3 , Vicens Tarancon 3 , Jordi Baliellas 3 , Elena Novell 3 , Sebastián Napp 1,2 and Lorenzo Fraile 4 1 IRTA, Centre de Recerca en Sanitat Animal (CReSA, IRTA-UAB), Campus de la Universitat Autònoma de Barcelona, Barcelona, Spain, 2 The OIE Collaborating Centre for the Research and Control of Emerging and Re-emerging Diseases in Europe (IRTA-CReSA), Barcelona, Spain, 3 Grup de Sanejament Porcí, Lleida, Spain, 4 Departament de Ciència Animal, ETSEA, Universitat de Lleida-Agrotecnio, Lleida, Spain Novel techniques of data mining and time series analyses allow the development of new methods to analyze information relating to the health status of the swine population in near real-time. A swine health monitoring system based on the reporting of clinical events detected at farm level has been in operation in Northeastern Spain since 2012. This initiative was supported by swine stakeholders and veterinary practitioners of the Catalonia, Aragon, and Navarra regions. The system aims to evidence the occurrence of endemic diseases in near real-time by gathering data from practitioners that visited Edited by: Lesley Stringer, swine farms in these regions. Practitioners volunteered to report data on clinical events Independent Researcher, Millport, detected during their visits using a web application. The system allowed collection, Scotland transfer and storage of data on different clinical signs, analysis, and modeling of the Reviewed by: Hsin-Yi Weng, diverse clinical events detected, and provision of reproducible reports with updated Purdue University, United States results. The information enables the industry to quantify the occurrence of endemic Coralie Lupo, diseases on swine farms, better recognize their spatiotemporal distribution, determine Institut Français de Recherche pour l’Exploitation de la Mer factors that influence their presence and take more efficient prevention and control (IFREMER), France measures at region, county, and farm level. This study assesses the functionality of *Correspondence: this monitoring tool by evaluating the target population coverage, the spatiotemporal Ana Alba-Casals ana.alba@irta.cat patterns of clinical signs and presumptive diagnoses reported by practitioners over more than 6 years, and describes the information provided by this system in near real-time. Specialty section: Between January 2012 and March 2018, the system achieved a coverage of 33 of This article was submitted to the 62 existing counties in the three study regions. Twenty-five percent of the target Veterinary Epidemiology and Economics, swine population farms reported one or more clinical events to the system. During the a section of the journal study period 10,654 clinical events comprising 14,971 clinical signs from 1,693 farms Frontiers in Veterinary Science were reported. The most frequent clinical signs detected in these farms were respiratory, Received: 31 May 2019 Accepted: 27 January 2020 followed by digestive, neurological, locomotor, reproductive, and dermatological signs. Published: 18 February 2020 Respiratory disorders were mainly associated with microorganisms of the porcine Citation: respiratory disease complex. Digestive signs were mainly related to colibacilosis and Alba-Casals A, Allue E, Tarancon V, clostridiosis, neurological signs to Glässer’s disease and streptococcosis, reproductive Baliellas J, Novell E, Napp S and Fraile L (2020) Near Real-Time signs to PRRS, locomotor to streptococcosis and Glässer’s disease, and dermatological Monitoring of Clinical Events Detected signs to exudative epidermitis. in Swine Herds in Northeastern Spain. Front. Vet. Sci. 7:68. Keywords: swine health, endemic diseases, monitoring, data mining, web application, endemic-epidemic doi: 10.3389/fvets.2020.00068 multivariate time-series model Frontiers in Veterinary Science | www.frontiersin.org 1 February 2020 | Volume 7 | Article 68 Alba-Casals et al. Monitoring Clinical Events in Swine Farms INTRODUCTION functionality of this system developed to monitor the frequency of endemic diseases in the swine population at region and county The prevention and control of diseases are essential to ensure level, and discusses the advantages and limitations related to efficient and sustainable swine production. Getting updated its implementation. information on the health status of the target swine population in near real-time can facilitate the implementation of efficient measures by swine stakeholders, veterinary practitioners, and MATERIALS AND METHODS government. Innovative surveillance methods based on the To illustrate how the GSP monitoring system operates, we analyses of various types of data, which may serve as indirect analyzed the data of clinical events reported voluntarily by health indicators, are under development (1–3). The ability veterinary clinicians from swine farms of the Catalonia, Aragon, to collect data in a cost-effective and timely manner from a and Navarra regions (Northeastern Spain) between January 2012 wide range of sources, the use of data mining techniques and and March 2018. time series analyses, and the possibility of generating dynamic reproducible reports, has led to the development of new ways of conducting surveillance in near real-time (4, 5). Development of a Web Application to In recent years, the Spanish swine sector has grown Report Clinical Events Detected in Swine significantly, with over 50% of the pig herds concentrated in Farms Catalonia and Aragon (regions located in the North East of The researchers and technicians of GSP, in collaboration with the country). In those areas, the number of sows in large- many swine stakeholders and veterinary practitioners, developed scale operations has increased, and an important proportion a web application (app) to collect and store data on clinical events of facilities are part of integrated industries with highly detected by veterinarians during their visits to farms. Before specialized farrowing, post-weaning, and finishing sites (6). In launching the system, all the practitioners and representatives this context of swine production, it is essential to maintain a good of the swine industry of this zone were convened to a sanitary status. recruitment meeting. Afterwards, twice a year the participants The Porcine Sanitation Group of Lleida, Spain (GSP) is were convened to a meeting for promoting their continuous a non-profit association that brings together pig owners, participation. The veterinarians that participated worked for independent breeders, and companies associated to the swine large integrated companies as well as small individual farms. sector in Northeastern Spain. The GSP aims to improve the Several meetings with representatives of the swine sector and swine health in farms and collaborates closely with the official veterinary practitioners took place to define and agree what animal health authorities carrying out actions related to disease data fields to include in the app, which could be executed by surveillance, prevention, and control. In 2012, the GSP decided desktop computer, smartphone, tablet, or other mobile device. to carry out a near-real time monitoring system in Aragon, Data from farms was supplemented with diagnostic test results if Catalonia and Navarra to gather data on clinical events detected samples had been submitted to the official laboratory. A program by practitioners. The GSP hypothesized that, by monitoring, was created to analyze the data automatically and report the targeting, and reporting clinical signs and presumptive diagnoses, health status of the swine population to veterinary practitioners it would be possible to reveal in near real-time the occurrence that participated. This app is currently accessible using a user of endemic diseases that are not notifiable. This information code and password through the link: http://www.gsplleida.net/es/ might help assess the spatiotemporal distribution of such diseases content/app-del-gsp. Figure 1 shows the app interface with the in these populations, identify subpopulations at high risk and fields to be filled out by a user detecting a clinical outbreak in a factors that influence disease presence. Practitioners and swine pig farm. stakeholders would benefit from this information to plan and take more efficient control measures. Data Source, Types and Preparation It is important to highlight that the initial intention of Data were mainly sourced from veterinarians who routinely this monitoring tool was not associated with a pre-defined visited the pig farms. If a veterinarian detected pigs with clinical control plan against a specific disease. The main aim of the signs during a visit to a farm, he/she registered the following tool was to gather data from swine herds in near real-time variables in the app for each clinical event at farm level: severity and provide accessible and regularly updated information to of the clinical event, date of the visit, official identification veterinary practitioners and swine stakeholders. The system of the farm, company to which the farm belonged, location aimed to visually track the spatiotemporal distribution and of the farm, type of animal (i.e., sows or pigs), category of spread of endemic diseases and support the decision of where age affected, body system affected, lesions observed during and when actions were necessary. Moreover, the system aimed to necropsy (if applicable), vaccines applied, and presumptive enhance the communication and cooperation within the swine diagnosis of the disease. The veterinarian identified the affected sector in Northeastern Spain. This work aims to evaluate the body system according to the clinical signs observed in swine, distinguishing between respiratory, digestive, neurological, Abbreviations: GSP, Porcine Sanitation Group; app, web application; REGA, locomotor, dermatological, and reproductive system. In the event National official farm identifier; APP, Actinobacillus pleuropneumoniae infection of detecting multiple disorders in the same farm (e.g., respiratory or porcine pleuropneumonia; PRRS, porcine respiratory reproductive syndrome. and digestive), each sign could be recorded individually. Frontiers in Veterinary Science | www.frontiersin.org 2 February 2020 | Volume 7 | Article 68 Alba-Casals et al. Monitoring Clinical Events in Swine Farms FIGURE 1 | Interface of a web application to record clinical event data from swine herds in Northeastern Spain. Fields written in Spanish. English translation: Case (Caso clínico), Severity (Gravedad), Date (Fecha), Regional official farm identifier (Marca Oficial). Company reference (Referencia empresa), National official farm identifier (REGA), Owner (Titular), Municipality (Población), Type of animal (Tipo de animal), Piglets or fattening pigs (Cerdos), Sows (Madres), Farm Location of the affected animals within the farm (Ubicación del animal), Maternity (Maternidad), Nursery (Destete), Fattening (Engorde), Clinical signs (Clínica), Dermatological (Dermatológica), Digestive (Digestiva), Locomotor (Locomotora), Neurological (Nerviosa), Reproductive (Reproductiva), Respiratory (Respiratoria), Presumptive disease (Enfermedad), Geographical situation (Ámbito geográfico), Productive phase (Etapa productiva), Date of clinical onset (Fecha inicio), Date of clinical end (Fecha fin), See (Ver en), List (Listado), Map representation (En mapa), Closest clinical cases (Cerca de mí), Search (Buscar), Advanced search (Búsqueda avanzada). Moreover, the veterinarian indicated the most plausible experience and considering the rates of mortality and morbidity presumptive diagnosis based on his/her clinical experience. and the negative impact of the event on productive performance. The presumptive diagnoses comprised a closed list of endemic The second source of data was the GSP official laboratory for diseases that included: porcine pleuropneumonia (APP- due to swine diseases. If practitioners submitted clinical samples from Actinobacillus pleuropneumoniae), porcine circovirus associated a reported affected herd, the laboratory carried out diagnostic disease (due to Porcine Circovirus type 2), clostridiosis (due to testing to confirm or rule out a suspected endemic disease. Clostridium spp.), unspecific diarrhea (when the microorganism The type of test used and the results obtained at farm level involved was unknown), swine dysentery (due to Brachyspira were then recorded to the app. Data of clinical cases and hyodisenteriae), colibacillosis (due to Escherichia coli), exudative laboratory confirmation testing were integrated at farm level epidermitis (due to Staphylococcus hycus), streptococcosis (due using a relational database built by the IT services of GSP. to Streptococcus suis), Glässer’s disease (due to Haemophilus The elapsed time between the report of a clinical event and parasuis), swine influenza (due to Swine Influenza virus), ileitis its laboratory confirmation ranged between 24 h and a week (due to Lawsonia intracellularis), leptospirosis (due to Leptospira depending on whether the diagnosis was performed by PCR, spp.), mycoplasmosis (due to Mycoplasma hyopneumoniae), any serology or microbiology. swine parasitosis, pasteurellosis (due to Pasteurella multocida), The third data source was the official census of the active rectal prolapse, matrix prolapse, porcine respiratory reproductive swine farms in the regions of study (i.e., Aragon, Catalonia, and syndrome (due to PRRSV), atrophic rhinitis (due to Bordetella Navarra) (7). This census contained the following fields: a unique bronchiseptica and/or Pasteurella multocida), salmonellosis identifier of the farm, the company to which the farm belonged, (due to Salmonella spp.), and gastric ulcers. The app allowed the municipality, the county, the province, the number of adult reporting of several presumptive diagnoses during a single visit. sows/boars, the number of fattening pigs, the number of piglets in However, this situation was very unusual, since the clinician nursery, the type of production, and the UTM coordinates (x, y). usually indicated a unique presumptive diagnosis. Finally, the The data registered by the veterinarians during their visits, and by veterinarian also categorized the severity of a clinical event as the official laboratory were pre-processed and integrated with the mild, moderate or severe taking into account his/her own clinical census data in order to get a final data set that could be analyzed. Frontiers in Veterinary Science | www.frontiersin.org 3 February 2020 | Volume 7 | Article 68 Alba-Casals et al. Monitoring Clinical Events in Swine Farms Coverage Assessment the incidence reported over time can be additively decomposed The initial aim of the GSP monitoring system was to gather into two components: an endemic component (or baseline rate data on clinical events occurring in any swine farm of Catalonia, of cases with a stable temporal trend) and an epidemic (or Aragon and Navarra. The swine population of these three regions autoregressive component). The endemic component includes totals 6,741 active pig farms, 79% of which were located in several terms to represent the reference number of cases as the Catalonia, 18% in Aragon and 3% in Navarra. Around 90 swine intercept, the trend and the possible seasonal variation over time. practitioners routinely visited these farms, the majority of which Added to these parameters, these endemic-epidemic multivariate were located in 12 of the 62 counties of the regions. Half of the time series models also allow the inclusion of a neighbor-driven farms belonged to 20 integrated swine companies. component and random effects to explain their influence on the Over the study period, the implementation of the monitoring clinical events. system was partial, as not all the veterinarians used the app to A basic formulation of the endemic-epidemic multivariate report clinical events when visiting the swine farms. Initially, time series models can be expressed as: to evaluate the coverage achieved by the GSP system, it was assessed from which counties the swine practitioners reported X clinical events. This set of counties corresponded to the accessible µit = ei υt + λYi,t−1 + ϕ ωji Yj,t−1 (1) j6=i population. Then, the coverage was also analyzed by type of production. The comparison allowed identification of those log(υt ) = α (υ) + βt t + γ sin(̟ t) + δ cos(̟ t) (2) swine farms not participating in the monitoring system and inference of results solely to the participating population. where the mean incidence of clinical events in each county i at week t (µit ) depends on two components. Spatiotemporal Analyses and Modeling of (1) An endemic component (υt ) multiplied by an offset that Clinical Signs and Presumptive Diagnoses corresponds to the accessible population fraction located Several descriptive analyses were conducted to summarize the in each county (ei ). Here, υt is incorporated as log-linear frequencies of clinical events with different clinical signs and predictor that includes an overall trend βt and sine-cosine presumptive diagnoses, and visualize if any spatiotemporal terms to represent an annual seasonal variation with a wave pattern emerged from the data. frequency ω = 2π/52. An initial exploration was carried out to describe the trend (2) An epidemic component split into two parts: an of clinical events monitored by week. The clinical signs and the autoregressive part that reproduces the incidence respective presumptive diagnoses were grouped and summarized within county i (λYi,t−1 ), and neighborhood effects that in tables, maps, and bar plots. The severity of clinical signs, the represent P the transmission from other adjacent counties j types of production and the age categories of the affected animals (ϕ j6=i ωji Yj,t−1 ). These epidemic parameters λ = exp(α (λ) ) were also analyzed to characterize the subpopulations affected. and ϕ = exp(α (ϕ) ) are assumed homogeneous across The analysis was complemented by the diagnostic testing results geographical units and constant over time. from the laboratory. Next, to evidence possible patterns over time and space, the The multivariate count time series defined at different number of events with different clinical signs and presumptive spatiotemporal units can be fitted to a Poisson model, diagnoses were explored at low spatiotemporal granularity or a negative binomial model if we need to account for (i.e., by county and week). The counts of clinical signs and overdispersion. In the case of a negative binomial model, the presumptive diagnoses reported weekly were represented with conditional mean (µit ) remains the same, but the conditional multiple surveillance time series. These series showed the pattern variance increases to µit (1+ µit ψi ) with additional unknown of each clinical sign for each one of the 33 counties included in overdispersion parameter ψi > 0. the population of study between January 2012 and March 2018. These models are very flexible and allow the inclusion of Moreover, the cumulative counts of clinical events were mapped covariates, estimated transmission weights, and random effects monthly and yearly at county level. to eventually account for unobserved heterogeneity of the units. In this study, to model the spatiotemporal patterns of clinical Spatiotemporal Modeling Illustrated by events of porcine pleuropneumonia as presumptive diagnosis, different models were evaluated adding diverse sequential Clinical Events of Porcine extensions. Initially, a basic model was evaluated accounting Pleuropneumonia as Presumptive for endemic and epidemic parameters with annual seasonality Diagnosis variation and overall trend. Then, other covariates, such as Counts of some clinical events grouped by clinical sign or county neighborhood effect or population fraction of each presumptive diagnosis evidenced an overall trend and/or annual county, were tested on the endemic or epidemic parameters, seasonality over time (e.g., clinical events such as porcine and finally random effects were tested to eventually account for pleuropneumonia as presumptive diagnosis). To get a better unobserved heterogeneity of counties. understanding of the observed patterns for different groups, The most appropriate model was selected by comparing the endemic-epidemic multivariate time series models for infectious values obtained from the Akaike Information Criterion and disease counts were used (8–13). This approach considers that choosing the lowest one (14, 15). Frontiers in Veterinary Science | www.frontiersin.org 4 February 2020 | Volume 7 | Article 68 Alba-Casals et al. Monitoring Clinical Events in Swine Farms Reporting Information in Near Real Time (24), “psych” (25), “maptools” (26), “rgeos” (27), “foreign” (28), The GSP application allowed not only recording and integration “plotrix” (29), “sp” (30), “rgdal”(31), and “spdep” (32). of data on clinical events, but also immediate feedback to The reproducible reports were built in Latex format (33) and veterinary practitioners on trends and spatiotemporal evolution compiled with RStudio (20). of events at county and regional level. In addition, reproducible documents were created in “pdf ” RESULTS format to report the updated information extracted from the analyses of data. All the stakeholders and veterinarians who Coverage Assessment collaborated in the network received these reports with detailed Between January 2012 and March 2018, a total of 55 practitioners results related to reported clinical events and spatial and temporal out of 90 volunteered to report clinical events (i.e., veterinary evolution of patterns of clinical signs and presumptive diagnoses. It is important to notice that these reports did not show raw information. In order to protect the privacy of the participating TABLE 1 | Coverage and number of swine farms reporting by production type and number of clinical events recorded by the GSP monitoring system between stakeholders, the information was summarized by region or January, 2012 and March, 2018. county without giving exact details on individual farms. Production type Coverage No. farms in the No. No. clinical (%) target farms events (%) population (%) reporting Software Used for the Development of the (%) Web Application and the Implementation of Fattening 25 4,677 (69) 1,494 (88) 9,014 (85) Analyses Sow farm 16 664 (10) 103 (6) 947 (9) The application of GSP was developed using the following Farrow-to-finish 8 647 (1) 50 (3) 504 (5) software: HTML5 (16), Java script (17), and Angular Nursery 10 253 (4) 20 (1) 45 (0.4) 2 (18), and the working environment of IONIC 4 Gilt development unit 6 177 (3) 10 (0.6) 107 (1) Continuous flow 4 173 (3) 6 (0.4) 24 (0.2) (https://ionicframework.com). finisher The analyses of this study were carried out using the Multiplication 11 85 (1) 9 (0.5) 12 (0.1) statistical software R (19) jointly with RStudio as a development Boar stud 0 34 (0.5) 0 (0) 0 (0) environment (20). The plotting of multiple time series, Boar development unit 0 17 (0.2) 0 (0) 0 (0) the mapping and the modeling were performed using the Continuous flow 11 9 (0.1) 1 (0.1) 1 (0.01) “surveillance” package. This package has been broadly used to nursery monitor public health data in diverse European institutions (16). Others 0 5 (0.1) 0 (0) 0 (0) In addition, other R packages were used to make calculations Total – 6,741 (100) 1,693 10,654 (100) and graphs: “doBy” (21), “gdata” (22), “ggplot2” (23), “lattice” (25.1) FIGURE 2 | Location of the target swine population and the coverage achieved by the GSP system by county between January, 2012 and March, 2018 in Aragon, Catalonia and Navarra. Frontiers in Veterinary Science | www.frontiersin.org 5 February 2020 | Volume 7 | Article 68 Alba-Casals et al. Monitoring Clinical Events in Swine Farms FIGURE 3 | Frequency of clinical signs with their spatiotemporal distribution reported by week and by county in swine farms of Northeastern Spain between January, 2012 and March, 2018. participation of 61%). These practitioners covered 33 counties farms with 8 and 4%, respectively. The composition of the of the 62 existing counties and reported a median of 5 clinical swine target population was slightly different and comprised events by farm with a range between 1 and 136. The 33 accessible 11 production types in which the fattening farms were the counties comprised 4,207 swine farms, which represented 62% most abundant (69%), followed by sow farms (10%), and of 6,741 farms of the target swine population. The counties farrow-to-finish farms (10%). Moreover, it is interesting to of Western Catalonia and Aragon were the most represented. remark that the median size of farms that reported clinical Over this period, 10,654 clinical events were reported from 1,693 events were larger than in the target population, mainly farms (i.e., 25 and 40% of target and accessible swine population, in fattening, continuous flow finisher, and sow farms. This respectively). Figure 2 shows the location of the target swine demonstrates that farms from integrated large-operations with population and the coverage achieved by the GSP system by highly specialized facilities were more likely to report problems region and county. Appendix lists the numerical county codes to the system. with their corresponding names. Table 1 summarizes the coverage and number of swine farms Most of the clinical events were reported in fattening and clinical events by production type reported by the GSP farms (88%), followed by sow farms and farrow-to-finish monitoring system. Frontiers in Veterinary Science | www.frontiersin.org 6 February 2020 | Volume 7 | Article 68 Alba-Casals et al. Monitoring Clinical Events in Swine Farms TABLE 2 | Summary of clinical signs and presumptive diagnoses differentiating growing pigs and sows (period: January, 2012–March, 2018). Type of Type of affected No. of No. of No. of counts by Degree of Presumptive diagnoses No. of No. of clinical animal affected clinical week (median severity presumptive confirmed signs farms signs and range) diagnoses diagnoses Respiratory Growing pigs 88% 1,231 5,698 17 (1–41) 75% mild, Swine Influenza 1212 19 23% moderate, Mycoplasmosis 1029 3 2% severe Pasteurellosis 731 3 PRRS 875 14 Porcine pleuropneumonia 678 8 Glässer’s disease 680 1 Rhinitis 31 0 Porcine circovirus associated 6 2 disease Sows 12% 118 751 4 (1–12) 1% mild, Swine Influenza 67 2 96% moderate, Mycoplasmosis 12 0 3% severe PRRS 82 15 Porcine pleuropneumonia 80 2 Glässer’s disease 189 1 Rhinitis 0 0 Porcine circovirus associated 1 1 disease Digestive Growing pigs 87% 1,115 4,256 13 (1–31) 73% mild, Colibacilosis 1231 7 25% moderate, Unspecific diarrhea 599 0 2% severe Clostridiosis 670 2 Swine dysentery 440 14 Salmonellosis 314 4 Ileitis 137 1 Ulcers 71 0 Any parasitosis 33 0 Rectal prolapse 4 0 Sows 13% 126 662 3 (1–14) 4% mild, Colibacilosis 315 3 93% moderate, Unspecific diarrhea 42 0 3% severe Clostridiosis 1 2 Swine dysentery 73 12 Salmonellosis 93 4 Ileitis 6 3 Neurological Growing pigs 82% 708 1,778 5 (1–16) 75% mild, Glässer’s disease 584 0 23% moderate, Streptococcosis 226 2 3% severe Sows 18% 32 379 3 (1–9) 98% moderate Glässer’s disease 157 0 2% severe Streptococcosis 217 0 Reproductive Growing pigs 84% 328 575 2 (1–10) 92% mild, PRRS 573 57 5% moderate, Porcine circovirus associated 0 0 disease 2% severe Sows 16% 71 90 1 (1–3) 17% mild, PRRS 87 57 60% moderate, Porcine circovirus associated 2 1 disease 23% severe Leptospirosis 1 0 Locomotor Growing pigs 98% 429 654 2 (1–11) 67% mild, Streptococcosis 618 0 32% moderate, Glässer’s disease 56 0 1% severe (Continued) Frontiers in Veterinary Science | www.frontiersin.org 7 February 2020 | Volume 7 | Article 68 Alba-Casals et al. Monitoring Clinical Events in Swine Farms TABLE 2 | Continued Type of Type of affected No. of No. of No. of counts by Degree of Presumptive diagnoses No. of No. of clinical animal affected clinical week (median severity presumptive confirmed signs farms signs and range) diagnoses diagnoses Sows 2% 9 11 1 (1–2) 91% moderate, Streptococcosis 11 0 9% severe Dermatological Growing pigs 96% 95 112 1 (1–4) 83% mild, Exudative epidermitis 77 0 15% moderate, PRRS 1 0 2% severe Sows 4% 5 5 1 (1–1) 60% moderate, Exudative epidermitis 5 0 40% severe FIGURE 4 | Number of clinical events with porcine pleuropneumonia as presumptive diagnosis (January, 2012–March, 2018). (A) Counts reported by week. (B) Cumulative counts mapped by county. During the first semester of 2012 the number of reports Spatiotemporal Descriptive Analyses of was relatively low, but in the second semester of that Clinical Signs and Presumptive Diagnoses year the level of reporting increased substantially being From 10,654 clinical events a total of 14,971 clinical signs were subsequently sustained throughout the whole study period (see reported, most of them in growing pigs (85%). In general, the Figure 3). degree of severity of clinical events detected in growing pigs Frontiers in Veterinary Science | www.frontiersin.org 8 February 2020 | Volume 7 | Article 68 Alba-Casals et al. Monitoring Clinical Events in Swine Farms FIGURE 5 | Clinical events of porcine pleuropneumonia as presumptive diagnosis reported by week and by county in Northeastern Spain between January, 2012 and March, 2018. FIGURE 6 | Summary of the fitted model to data of suspected clinical events of porcine pleuropneumonia recorded by the GSP system between January, 2012 and March, 2018: coefficients and resulting multiplicative effect of seasonality on the endemic component. farms (nursery and fattening pigs) was milder than in sow farms reported in clinical events with digestive signs was colibacilosis, (sows and/or nursery pigs). In both growing pigs and sow farms, while the most frequent suspicion for neurological signs was respiratory clinical signs were the most frequent, followed by Glässer’s disease, PRRS for reproductive signs, streptococcosis for digestive, neurological, locomotor, reproductive, and finally locomotor signs, and exudative epidermitis for dermatological dermatological signs. A combination of several clinical signs signs. Less than 1% of those presumptive diagnoses were was observed in 30% of these events (n = 3,182). The most confirmed by laboratory diagnosis. Table 2 summarizes the frequent combination was digestive and neurological signs (7%), clinical signs by type of affected animals (i.e., growing pigs or followed by respiratory and neurological (5%), respiratory sows), the number of affected farms, the degree of severity and and locomotor (4%), and respiratory and reproductive the associated presumptive and confirmed diagnoses. (4%). Figure 3 summarizes the frequency of all clinical signs Next, using porcine pleuropneumonia as example of reported with their respective spatiotemporal distribution by presumptive diagnosis, we illustrate how the clinical events county and week. grouped by each presumptive diagnosis were represented The main presumptive diagnoses associated with respiratory spatiotemporally. It is important to note that the clinical signs clinical signs were diseases that belong to the porcine respiratory of porcine pleuropneumonia are quite pathognomonic, and complex (such as swine influenza, PRRS, and mycoplasmosis), thus clinical suspicions were a useful measure of the pattern followed by Glasser’s disease, pasteurellosis, and porcine of disease. The disease was suspected by the veterinarians on pleuropneumonia. The most frequent presumptive diagnosis 758 occasions, most often in pigs on fattening farms (88%). Frontiers in Veterinary Science | www.frontiersin.org 9 February 2020 | Volume 7 | Article 68 Alba-Casals et al. Monitoring Clinical Events in Swine Farms FIGURE 7 | Illustration of layout of the GSP web application to visualize the area and the trend of reported reproductive clinical events. Figure 4 illustrates the time series of the number of clinical coefficients of the fitted model and the resulting multiplicative events reported by week and the cumulative counts of events effect of seasonality on the endemic component for the reported by county (see Figure 4). suspected clinical events of swine pleuropneumonia are shown Moreover, using multivariate surveillance time series, trend in Figure 6. at county level can be visually assessed and compared among counties (see Figure 5). Reporting Information in Near Real Time Figures 4, 5 show that, except for 2012, throughout the Finally, with the aim of providing a continuous feedback to whole period of study the overall trend of reporting was stakeholders and veterinarians, and communicate information quite stable with a seasonal increase each winter. From the on the health status of the population in near real time, accessible population 19 out of the 33 counties reported at the system produced different reports. Directly from the web least one suspicion of porcine pleuropneumonia. Although application the user could get the area where the clinical the number of weekly counts by county were relatively events were reported during the last 3 months and the trend low (i.e., maximum 3), those counties of Western Catalonia (see Figure 7). and Aragon that had more swine farms also reported more Moreover, every month the stakeholders received a brief consistently suspicions of porcine pleuropneumonia (e.g., county report summarizing the information of the clinical events labeled as 26). reported by age, type of farms, and counties. Each year they also received a very detailed report of the monitoring conducted and the results of the models that evaluate the pattern of some clinical Spatiotemporal Modeling for Porcine signs and presumptive diagnoses. Pleuropneumonia as Presumptive Diagnosis DISCUSSION The data on porcine pleuropneumonia as presumptive diagnosis was fitted using a negative binomial model. This model included Traditionally, in Spain, the reporting of many endemic swine an endemic component with a marked seasonality that increased diseases through passive surveillance has been very disperse between January and March and an epidemic component. and scarce, and the information required for making proper In this case, the influence of random effects structures decisions in health management at population level has often could not be assessed due to the lack of convergence. The been poor (4). The development of new user-friendly and Frontiers in Veterinary Science | www.frontiersin.org 10 February 2020 | Volume 7 | Article 68 Alba-Casals et al. Monitoring Clinical Events in Swine Farms standardized digital methods to report and analyze data on In addition, the spatiotemporal descriptive analyses of clinical events can help to determine the frequency and the retrospective data from clinical events allowed assessment of evolution of diseases at population level (5, 34). Recent initiatives the evolution of different clinical signs and endemic diseases, have been carried out to monitor some endemic diseases in swine comparison of different subpopulations and identification of populations using these methods. Although many authors have groups of farms, areas, or periods with higher incidence of pointed out the potential for near real-time monitoring, system specific problems. This long-term monitoring could help to development faces several technical, social and communication determine the baseline frequency of clinical signs or endemic challenges in order to define data standards and data-sharing diseases, assess the influence of different factors on disease agreements (1–3, 5). Implementation involves a multidisciplinary presence, and predict clinical events. In our study, the results approach and the participation of the swine sector. This of these analyses showed that the most frequent clinical was the case with the GSP system, built in collaboration signs reported from the accessible population were respiratory, with researchers in swine diseases, lab technicians, computer followed by digestive and neurological. Moreover, as example of scientists, practitioners, epidemiologists and with the support a more detailed analysis, by combining data of clinical signs and of producers of Northeastern Spain, all of whom were fully presumptive diagnoses, we observed that the practitioners mainly committed to the initiative. associated these signs with diseases of the respiratory complex The coverage assessment of the GSP system allowed (such as swine influenza, mycoplasmosis, or PRRS), followed by identification of the accessible population under monitoring. pasteurellosis, porcine pleuropneumonia, and Glässer’s disease. This study, based on data collected over more than 6 years, However, it is important to note that due to the small number shows that implementation was gradual during the first year of samples (<1% of events), most of these presumptive diagnoses (2012), while in subsequent years the practitioners registered were not confirmed by the laboratory. The main reason why clinical events on a regular basis. The spatial coverage of swine practitioners did not take samples was that they believed the target swine population was partial and varied between that the laboratory confirmation would not change the medical counties (see Figure 2) and types of production (see Table 1). interventions to undertake at farm level; and thus, they preferred Despite this limitation, the monitoring system achieved the to avoid extra-costs and logistical difficulties. The monitoring collection of data from an important proportion of fattening of presumptive diagnoses without laboratory confirmation could farms from integrated large operations in 33 counties, mainly result in false alerts being raised. To minimize this limitation, concentrated in Western Catalonia and Aragon. In total 55 out we suggest identifying subpopulations frequently affected by of 90 swine practitioners in Northeastern Spain volunteered clinical signs or endemic disease suspicions, communicating the to participate. Veterinarians usually worked in a specific area information to practitioners and recommending submission of and this could lead to spatiotemporal clustering of reporting samples for laboratory confirmation. Furthermore, the reporting of clinical events. In each county, the clinical events were of presumptive diagnoses was defined as a closed list of possible reported by veterinarians from different companies so, to ensure endemic diseases and the option to report other endemic diseases standardized reporting, the same training was provided to all the (not included in the list) or exotic diseases was not considered. system participants. To improve this reporting, we suggest including the option of To improve the coverage in areas where the app was not “other suspicion” within an open field, where the veterinarian used, our suggestions are to hold more meetings explaining could record other diagnoses or findings. the benefits of the information provided by this kind of The reproducible reports created by the GSP system provided monitoring and try to sort out the problems that prevent updated and continuous information to the practitioners and practitioners from participating. On the other hand, since this swine stakeholders who participated. These reports showed system did not allow differentiating if a specific farm was visually the frequency of health problems at county and regional not visited or the veterinarian did not detect any clinical level, allowed the identification of their spatial distribution and event, we recommended adding a field in the application to progress, and helped the decision-making of where and when record all visits of the practitioner, even when no disease was actions were necessary. Practitioners and swine stakeholders observed. We believe that the recording of all the clinical benefited from sharing information of clinical events occurring inspections carried out would help improve the assessment of the in the neighboring areas to plan control measures against coverage of the system and serve to demonstrate the absence of these infections at farm level. Moreover, this system facilitated endemic diseases. communication within the swine sector in Northeastern Spain The spatiotemporal descriptive analyses from clinical events and promoted co-operation. However, at this initial stage, the reported at farm level allowed the identification in near real- interventions to undertake in the event of alert at population level time of the most frequent clinical signs and presumptive had not been agreed upon by different practitioners and private suspicions at county and regional level. This easily accessible and stakeholders, so the system was not ready to be used to plan current information could be useful to veterinary clinicians and specific actions. A future potential use of this system would be as stakeholders for decision making. For example, if a practitioner surveillance system in order to detect outbreaks or aberrations. knew that the incidence of an endemic disease had increased in Nevertheless, for directing effective control actions, we still need neighboring farms, he/she could decide to implement or modify to gradually build more trust in the current monitoring and preventive measures (e.g., vaccination) or take samples in other achieve a better consensus and commitment from the whole swine farms to confirm the presence or absence of infection. swine sector. Frontiers in Veterinary Science | www.frontiersin.org 11 February 2020 | Volume 7 | Article 68 Alba-Casals et al. Monitoring Clinical Events in Swine Farms CONCLUSIONS provided data. All data have been analyzed in aggregate form guaranteeing their privacy and security. Overall, we believe that the kind of monitoring system described in this study provides very useful information to detect and monitor the trend of the most frequent endemic diseases, identify AUTHOR CONTRIBUTIONS specific health problems and to enhance communication within AA-C conceived the research, performed the statistical analyses, the swine sector. A consensual and broad implementation of developed the reproducible reports, and wrote the article. EA the system on the whole target population could shorten the conceived the research, designed the app, obtained the data, and response time to prevent and control certain diseases, decreasing provided the layout to show the results in the app. VT and JB productive, and sanitary losses. conceived the research, built, coordinated, and motivated the Further research could be directed at identifying disease network of veterinary clinicians that participated in the GSP characteristics and modeling other covariates of interest at system. EN conceived the research, performed the laboratory company or county level to further benefit endemic disease tests, and interpreted the results. SN interpreted the laboratory control within the swine industry. test results and wrote the article. LF conceived the research, reviewed the statistical analyses and reproducible reports, guided DATA AVAILABILITY STATEMENT the study, and wrote the article. The datasets generated for this study are available on request to the corresponding author. ACKNOWLEDGMENTS ETHICS STATEMENT This study was carried out with the support of the Porcine Sanitation Group of Lleida, Spain (Grup de Sanejament Porcí- The approval of our study was not required as per the local GSP) in collaboration with all the veterinary clinicians working legislation. This research did not involve any specific clinical in the field. We would like to thank the editor and the reviewers study using animal experimentation. This work has been written for their constructive comments and efforts toward improving in agreement with the veterinary clinicians and stakeholders who our manuscript. REFERENCES 11. Held L, Paul M. Modeling seasonality in space-time infectious disease surveillance data. Biom J. (2012) 54:824–43. doi: 10.1002/bimj.2012 1. Dórea FC, Sanchez J, Revie CW. Veterinary syndromic surveillance: current 00037 initiatives and potential for development. Prev Vet Med. (2011) 101:1–17. 12. Meyer S, Held L, Höhle M. Spatio-temporal analysis of epidemic phenomena doi: 10.1016/j.prevetmed.2011.05.004 using the R package surveillance. arXiv preprint arXiv:1411.0416. (2014). 2. Dórea FC, Vial F. Animal health syndromic surveillance: a systematic 13. Paul M, Meyer S. hhh4: an endemic-epidemic modelling framework for literature review of the progress in the last 5 years (2011–2016). Vet Med Res infectious disease counts. Package Surveillance. (2016). Available online at: Rep. (2016) 7:157–70. doi: 10.2147/VMRR.S90182 https://cran.r-project.org/web/packages/surveillance/vignettes/hhh4.pdf 3. Dupuy C, Bronner A, Watson E, Wuyckhuise-Sjouke L, Reist M, Fouillet A, 14. McDonald RP. An index of goodness-of-fit based on noncentrality. J Classif. et al. Inventory of veterinary syndromic surveillance initiatives in Europe (1989) 6:97–103. doi: 10.1007/BF01908590 (triple-S project): current situation and perspectives. Prev Vet Med. (2013) 15. Höhle M, Meyer S, Paul M. Surveillance: Temporal and Spatio-Temporal 111:220–9. doi: 10.1016/j.prevetmed.2013.06.005 Modeling and Monitoring of Epidemic Phenomena. R Package Version (2015). 4. Rodríguez-Prieto V, Vicente-Rubiano M, Sánchez-Matamoros A, Rubio- 16. Lubbers P, Albers B, Salim F, Pye T. Pro HTML5 Programming. New York, NY: Guerri C, Melero, M, Martínez-López B, et al. Systematic review of Apress (2011). surveillance systems and methods for early detection of exotic, new and re- 17. Flanagan D. JavaScript: The Definitive Guide. O’Reilly Media, Inc. (2006). emerging diseases in animal populations. Epidemiol Infect. (2015) 143:2018– 18. Fain Y, Moiseev A. Angular 2 Development With TypeScript. Manning 42. doi: 10.1017/S095026881400212X Publications Co. (2016). 5. Lambert MÈ, Audet P, Delisle B, Arsenault J, D’Allaire S. Porcine reproductive 19. R Development Core Team. R: A Language and Environment for Statistical and respiratory syndrome virus: web-based interactive tools to support Computing. Vienna: R Foundation for Statistical Computing (2012). surveillance and control initiatives. Porcine Health Manag. (2019) 5:10. 20. Van der Loo MP. Learning R Studio for R Statistical Computing. Packt doi: 10.1186/s40813-019-0117-x Publishing Ltd. (2012). 6. Magapor Swine News. The Swine Sector in Spain and Its Evolution. 21. Højsgaard S, Halekoh U. doBy: Groupwise Statistics, LSmeans, Linear Available online at: https://www.magapor.com/en/el-sector-porcino-espanol/ Contrasts, Utilities. R Package, Version 4.5–15. (2016). (accessed September 05, 2018). 22. Warnes GR, Bolker B, Gorjanc G, Grothendieck G, Korosec A, Lumley 7. Allue E. Swine Census. Lleida (2017). Available online at: http://www.gsplleida. T, et al. gdata: Various R Programming Tools for Data Manipulation. R net/es Package Version 2.18.0. (2017). Available online at: https://CRAN.R-project. 8. Held L, Höhle M, Hofmann M. A statistical framework for the analysis org/package=gdata of multivariate infectious disease surveillance counts. Stat Model. (2005) 23. Wickham H. ggplot2: Elegant Graphics for Data Analysis. Springer-Verlag New 5:187–99. doi: 10.1191/1471082X05st098oa York (2016). 9. Paul M, Held L, Toschke AM. Multivariate modelling of infectious disease 24. Sarkar D. Lattice: Multivariate Data Visualization With R. New York, NY: surveillance data. Stat Med. (2008) 27:6250. doi: 10.1002/sim.3440 Springer (2008). 10. Paul M, Held L. Predictive assessment of a non-linear random effects model 25. Revelle W. psych: Procedures for Personality and Psychological Research. for multivariate time series of infectious disease counts. Stat Med. (2011) Evanston, IL: Northwestern University. (2018). Available online at: https:// 30:1118–36. doi: 10.1002/sim.4177 CRAN.R-project.org/package=psych Version = 1.8.12. Frontiers in Veterinary Science | www.frontiersin.org 12 February 2020 | Volume 7 | Article 68 Alba-Casals et al. Monitoring Clinical Events in Swine Farms 26. Bivand R, Lewin-Koh N. maptools: Tools for Handling Spatial Objects. R 33. Lamport L. LATEX: A Document Preparation System: Package Version 0.9–5. (2019). Available online at: https://CRAN.R-project. User’s Guide and Reference Manual. Addison-Wesley org/package=maptools (1994). 27. Bivand R, Rundel C. rgeos: Interface to Geometry Engine - Open Source 34. VanderWaal K, Morrison RB, Neuhauser C, Vilalta C, Perez, (’GEOS’). R Package Version 0.4–3. (2019). Available online at: https://CRAN. AM. Translating big data into smart data for veterinary R-project.org/package=rgeos epidemiology. Front Vet Sci. (2017). 4:110. doi: 10.3389/fvets.2017. 28. R Core Team. foreign: Read Data Stored by ‘Minitab’, ‘S’, ‘SAS’, ‘SPSS’, ‘Stata’, 00110 ‘Systat’, ‘Weka’, ‘dBase’, .... R Package Version 0.8–70. (2017). Available online at: https://CRAN.R-project.org/package=foreign Conflict of Interest: The authors declare that the research was conducted in the 29. Lemon J. Plotrix: a package in the red light district of R. R News. (2006) 6: 8–12. absence of any commercial or financial relationships that could be construed as a 30. Bivand RS, Pebesma E, Gomez-Rubio V. Applied Spatial Data Analysis With potential conflict of interest. R. 2nd ed. (2013). New York, NY: Springer. Available online at: http://www. asdar-book.org/ Copyright © 2020 Alba-Casals, Allue, Tarancon, Baliellas, Novell, Napp and Fraile. 31. Bivand R, Keitt T, Rowlingson B. rgdal: Bindings for the ’Geospatial’ Data This is an open-access article distributed under the terms of the Creative Commons Abstraction Library. R Package Version 1.4–3. (2019). Available online Attribution License (CC BY). The use, distribution or reproduction in other forums at: https://CRAN.R-project.org/package=rgdal is permitted, provided the original author(s) and the copyright owner(s) are credited 32. Bivand RS, Wong DWS. Comparing implementations of global and that the original publication in this journal is cited, in accordance with accepted and local indicators of spatial association. TEST. (2018) 27:716–48. academic practice. No use, distribution or reproduction is permitted which does not doi: 10.1007/s11749-018-0599-x comply with these terms. Frontiers in Veterinary Science | www.frontiersin.org 13 February 2020 | Volume 7 | Article 68 Alba-Casals et al. Monitoring Clinical Events in Swine Farms APPENDIX TABLE A1 | List of numerical county codes and corresponding names covered by the GSP monitoring system. Codes County names 0 ALT URGELL 1 ANOIA 2 BAIX CAMP 3 BAIX EBRE 4 BAJO ARAGON 5 BAJO CINCA 6 BORJA 7 CALATAYUD 8 CASPE 9 CONCA DE BARBERA 10 DAROCA 11 EGEA DE LOS CABALLEROS 12 GARRIGUES 13 HOYA DE HUESCA 14 LA ALMUNIA DE DONA GODINA 15 LA LITERA 16 MONEGROS 17 NOGUERA 18 PALLARS JUSSA 19 PLA D’URGELL 20 PRIORAT 21 RIBARGORZA 22 RIBERA ALTA ARAGON 23 RIBERA BAJA 24 RIBERA D’EBRE 25 SEGARRA 26 SEGRIA 27 SERRANIA DE MONTALBAN 28 SOBRARBE 29 SOLSONES 30 SOMONTANO 31 URGELL 32 ZARAGOZA Frontiers in Veterinary Science | www.frontiersin.org 14 February 2020 | Volume 7 | Article 68