key: cord-0974806-z5c0kx5t authors: Tizzani, Michele; Muñoz-Gómez, Violeta; De Nardi, Marco; Paolotti, Daniela; Muñoz, Olga; Ceschi, Piera; Viltrop, Arvo; Capua, Ilaria title: Integrating digital and field surveillance to complement efforts to manage epidemic diseases of livestock: African swine fever as a case study date: 2021-05-27 journal: bioRxiv DOI: 10.1101/2021.05.27.445948 sha: 4c8cdaa6155a5b8b98950fb5c88722f8693a95f8 doc_id: 974806 cord_uid: z5c0kx5t The SARS-CoV-2 pandemic has unveiled the importance of stakeholders and ordinary citizens in managing infectious disease emergencies. Efficient management of infectious diseases requires a top-down approach which must be complemented with a bottom-up response to be effective. Here we investigate a novel approach to surveillance for transboundary animal diseases using African Swine fever as a model. We were able to collect data at a population level on information-seeking behavior and at a local level through a targeted questionnaire-based survey to relevant stakeholders such as farmers and veterinary authorities. Our study shows how information-seeking behavior and resulting public attention during an epidemic, can be addressed through novel data streams from digital platforms such as Wikipedia. We also bring evidence on how field surveys aimed at local workers (e.g. farmers) and public authorities remain a crucial tool to assess more in-depth preparedness and awareness among front-line actors. We conclude that they should be used in combination to maximize the outcome of surveillance and prevention activities for selected transboundary animal diseases. African Swine Fever (ASF) is a transboundary animal disease and its impact on global markets can 33 potentially be catastrophic, threatening the economy from local to global level (1). Despite decades of international control efforts, the disease is still spreading in various regions of the 35 world with different epidemiological dynamics. These are significantly influenced by the natural 36 environment as this influences the density of susceptible species and vectors (i.e. ticks) but also 37 significantly by human behavior (2). In particular, human behavior plays a key role in the transmission 38 and geographic spread of the ASF virus (3,4) through infringement or low compliance with biosecurity 39 measures. Among all, the practice of swill (scraps of meat that are potentially infectious) feeding 40 mainly in backyard farms is known to be a driver of infection. Additionally, underreporting of ASF 41 suspected cases linked to movements of contaminated pork products and infected pigs have 89 We used the Wikipedia application programming interface (API) (22) to collect the number of visits 90 per day of Wikipedia articles normalized with the total monthly access to Wikipedia from each 91 targeted country. We selected the Wikipedia articles between 2015/02 and 2020/05 specific for ASF, 92 namely 'African_swine_fever_virus', and the relative translations in the languages of interest, (S1 93 Table) . For most of the countries of the study, the language is highly indicative of the location. On the 94 other hand, a weighted normalization factor for the number of views was necessary to account for 95 the multilingualism of some of them, like Belgium and India. More specifically, we weighted the 96 number of daily accesses to a single article from a Wikipedia project (S1 Table) p, S p (d), with the total 97 number of monthly accesses from a country, c, to the related Wikipedia project T c p (d) , such that the 98 daily pageviews from a given Wikipedia project and country was described by equation (1): where the denominator is the total number of views of the Wikipedia-specific project. The total 103 volume of views at day, d, from a country, c, is then given by the sum over all the articles and projects, 104 p, given by equation (2): English the summary of the full text of each article to implement a topic modeling analysis. To analyze the correlation between ASF media coverage in a given country and online users' 117 collective response as measured through the Wikipedia pageview, we introduced two regression 118 models as shown in Equation number 4. The first one is a simple linear regression, while the second 119 one contains a memory kernel to account for "memory effects" (e.g. loss of interest) in the public 120 response to media coverage. In the latter, the cumulative news articles volume time series were 121 weighted with an exponential decaying term (9,11) introducing the variable in equation (3): where τ is a free parameter that sets the memory timescale. We tuned τ in the range of [1, 60] 124 optimizing the results of the linear regression for the adjusted R 2 , and showing only the best results. Finally, the two models considered are shown in Equation (4): where y t is the number of country-specific Wikipedia pageviews, and u t is the error term. More details 129 on the diagnostics for the two models can be found in the Supplementary Information. In the two 130 models, the independent variables are either the news volume or the news volume plus a memory 131 term. Finally, to qualitatively explore the content of the digital news, we analyze the prevalent topics in the 133 news articles through an unsupervised topic modeling approach (23). Topic modeling is a statistical 134 method that is particularly effective for classifying, clustering, and arranging textual data in latent 135 themes. It has been extensively applied in the literature to extract groups of coherent information 136 from a list of documents(9,23-25). We used a well-known probabilistic framework, the latent Dirichlet 137 allocation (LDA) (26). We cleaned and lemmatized the text using the python "spacy" library (27) while 138 the number of topics was chosen through a grid search of the parameter for the Latent Dirichlet Algorithm from the scikit-learn python library (28). The process led originally to twenty-five topics 140 which were lately grouped into 5 main broad topics. All the authors (the majority of which have a 141 background in veterinary sciences) were involved in the qualitative annotation of the resulting topics. The news analyzed in this activity was focused on Estonia to qualitatively assess the coverage of the 143 problem in this specific country where we have also carried out more in-depth analysis by means of 144 field surveys described in the next section. In Table 2 we show the results of the two regressions models and in Table 3 Table) . In experiencing the COVID 19 pandemic, it is clear that epidemic diseases are to be managed at the 286 general population level as well as at the local level, and that these two approaches must be synergic In the case of ASF, specific sub-groups of the population that are at the forefront of an epidemic 298 preparedness, such as pig farmers, differ from the general population in the sense that they are 299 considerably more aware and interested in a topic that affects them directly, and this is reflected in 300 the results of the questionnaire. Not the same can be said for the general population, given that digital 301 information-seeking behavior peaked only after generalist news coverage. For these reasons, both 302 approaches (digital and field-oriented) are valuable and needed from a public health perspective to 303 understand how interest, risk perception, and awareness in the general population and specific 304 interest groups evolve during an epidemic and how they may affect public opinion as well as the 305 preparedness. First, we exploited the pervasiveness of digital data to assess the awareness among the general 307 population as measured through Wikipedia page views in reaction to the exposure to news about ASF 308 outbreaks at an international scale. Then, we examined the preparedness, awareness, and 309 information-seeking in localized areas through questionnaires to farmers and health authorities, with 310 a focus on the Estonian context. Expectedly our results show that, at an international scale, public interest rapidly declines after an 312 initial attention peak which occurs after exposure to news coverage of a specific outbreak. In 313 particular, the public activity profile as measured through the access to the Wikipedia pages shows 314 nonlinear dependencies and memory effects in the relation between information seeking, media 315 pressure, and disease dynamics. Modelling the 404 global economic consequences of a major African swine fever outbreak in China Epidemiological considerations on 407 African swine fever in Europe Small-scale pig farmers' behavior, silent 409 release of African swine fever virus and consequences for disease spread African Swine Fever Epidemiology and Control 416 6. OIE and FAO. Standing Group of Experts on African swine fever in Europe. Depository on 417 African swine fever Information-seeking 420 behaviour for epilepsy: An infodemiological study of searches for Wikipedia articles. Epileptic 421 Disord Asthma information seeking via Wikipedia between Collective response to the 426 media coverage of COVID-19 Pandemic on Reddit and Wikipedia Public Anxiety and 429 Information Seeking Following the H1N1 Outbreak : Blogs , Newspaper Articles Wikipedia Visits Public Anxiety and Information Seeking Following the H1N1 Outbreak : Blogs 431 , Newspaper Articles , and Wikipedia Visits The impact of news exposure on collective 433 attention in the United States during the 2016 Zika epidemic Using knowledge, attitudes 436 and practice (KAP) surveys on lymphatic filariasis to prepare a health promotion campaign for 437 mass drug administration in Alor District Dissanayake 440 DRA, et al. Knowledge, attitudes, and practices survey of rabies in a community in Sri Lanka Supporting 443 control policies on African swine fever in Ukraine through a knowledge, attitude, and practice 444 survey targeting backyard farmers Development of 446 African swine fever epidemic among wild boar in Estonia-two different areas in the 447 epidemiological focus Epidemiological analysis of 450 the 2015-2017 African swine fever outbreaks in Estonia ASF situation in Estonia Surveillance and prevention of transmission to/from domestic pigs To sample or not to 458 sample? Detection of African swine fever in wild boar killed in road traffic accidents Self-declaration by Estonia as a country free from African swine fever in domestic 461 and captive wild pigs Situational updates of ASF in Asia and the Pacific Public Anxiety and Information Seeking 468 Following the H1N1 Outbreak: Blogs, Newspaper Articles, and Wikipedia Visits Health communication through news 471 media during the early stage of the covid-19 outbreak in China: Digital topic modeling 472 approach An overview of topic modeling and its current 474 applications in bioinformatics Health Monitoring on Social 476 Media over Time Latent Dirichlet Allocation: Extracting Topics from Software 478 Engineering Data Natural language understanding with Bloom embeddings, convolutional 480 neural networks and incremental parsing Scikit-learn: 483 Machine Learning in Python e-estonia-We have built a digital society and we can show you how Testing Hypotheses in Nested Regression Models How to demonstrate 490 freedom from African swine fever in wild boar-Estonia as an example The biosecurity battle with African Swine Fever on Farms Mfuleni pig farmers hit hard by African Swine Fever outbreak African Swine Fever Policies: Do We Address Virus and Host Adequately? Front Vet Sci African swine fever ( ASF ) virus in wild boar in Belgium : Situation and detailed 502 information Shiny from R Studio Autocorrelations by country Linear Regression-Diagnostics 510 S1 Table. Total pageviews and news by selected country The null 512 hypothesis is that residuals are homoscedastic, hence a p-value < 0.05 indicates heteroscedasticity 513 S3 Questionnaire to Estonian farmers. Selected questions analyzed in this study are marked with a 514 star* 515 S4 Questionnaire to the Estonian veterinary authorities Biosecurity aspects that Estonian farmers mentioned having invested resources due to ASF Public institutions and private organizations mentioned by the Estonian veterinary 520 authorities as part of the risk management strategy Target groups, information material, and communication channels mentioned by the 522 Estonian veterinary authorities S9 GDELT query for data collection