Machine learning for syndromic surveillance using veterinary necropsy reports

Abstract

The use of natural language data for animal population surveillance represents a valuable opportunity to gather information about potential disease outbreaks, emerging zoonotic diseases, or bioterrorism threats. In this study, we evaluate machine learning methods for conducting syndromic surveillance using free-text veterinary necropsy reports. We train a system to detect if a necropsy report from the Wisconsin Veterinary Diagnostic Laboratory contains evidence of gastrointestinal, respiratory, or urinary pathology. We evaluate the performance of several machine learning algorithms including deep learning with a long shortterm memory network. Although no single algorithm was superior, random forest using feature vectors of TF-IDF statistics ranked among the top-performing models with F1 scores of 0.923 (gastrointestinal), 0.960 (respiratory), and 0.888 (urinary). This model was applied to over 33,000 necropsy reports and was used to describe temporal and spatial features of diseases within a 14-year period, exposing epidemiological trends and detecting a potential focus of gastrointestinal disease from a single submitting producer in the fall of 2016.

Research perspectives on animal health in the era of artificial intelligence

“…AI also includes methods to mobilise information available on the web. For example, semi-automatic data mining methods enabled to identify emerging signals for international surveillance of epizooties [ 74 ] or to analyse veterinary documents such as necropsy reports [ 75 , 76 ]. Methods from the field of natural language processing can compensate the scarcity of data by extracting syntactic and semantic information from textual records, triggering alerts on new emerging threats that could have been missed otherwise.…”

“…Such research is just starting in AH [ 46 ] and must be extended as part of the development of agro-ecology, facing current societal demand for product quality and respect for ecosystems and their biodiversity on one side, animal well-being and ethics on the other side, and more generally international health security.

Figure 4 Identifying relevant strategies to control bovine paratuberculosis at a regional scale (adapted from [ 76 ] ) . Classically, identifying relevant strategies means defining them a priori and comparing them, e.g., by modelling.

…”

Autoregressive Models Applied to Time-Series Data in Veterinary Science

“…These studies were focused mostly on livestock (26). The temporal unit of data collection was most commonly day (17) or month (16), and the median period (years) covered by the datasets analyzed was 10 (IQR [5][6][7][8][9][10][11][12][13][14][15][16].…”

Ir-Man

“…All of the reports listed above applied information retrieval techniques to biomedical text pertaining to humans. In the veterinary domain, Bollig et al [2] used a machine learning based approaches for extraction of different pathologies from free text. Furrer et al [7] built a text mining tool for veterinary surveillance by linking terms identified in necropsies to existing ontologies.…”

“…The ability to effectively confirm or rule out the presence of a pathology based on descriptions of abnormalities would allow for clearer understanding of the problems facing marine mammals. Information retrieval approaches have been applied extensively in human pathology [3,8,11,21,25,26], and other animal pathologies [2,7,13], however, no work currently exists for marine mammal pathology free text.…”