Translating Big Data into Smart Data for Veterinary Epidemiology

VanderWaal, Kimberly; Morrison, Robert B.; Neuhauser, Claudia; Vilalta, Carles; Perez, Andres M.

Of the Vs of big data, velocity represents the largest departure from traditional data processing in veterinary epidemiology, but it also has the most potential for revolutionizing the field, particularly in regard to monitoring and surveillance (2, 3, 7, 15, 19, 54) . Sources of high velocity data include digitized records from clinics and diagnostic laboratories, analysis of Google search trends and social media, and mortality and abattoir data (2, 3, 15, 49, (54) (55) (56) . Analysis of data through time allows for the establishment of baselines to which emerging data can be compared (49) . The typical values for a metric that relates directly or indirectly to disease (e.g., incidence or production levels) are summarized for a particular population or spatial location, and deviations outside the normal variation of the metric can be used as an indicator of an outbreak (57, 58) . Time-series analysis provides an additional method for mathematically quantifying temporal patterns, incorporating seasonality and long-term trends (58). Short-term predictions can then be made in terms of expected incidence or prevalence over time. Departures from expectations can be considered "anomalies" and may serve as early warnings for emerging threats or altered disease dynamics that warrant further investigation or intervention (54, 56, 58) . For example, a recent initiative involving 700 veterinary hospitals in the US tracked the daily proportion of patients with certain clinical signs, contrasting new data with averages from a retrospective period of time. As a proof-of-concept, the system was able to rapidly detect a simulated outbreak scenario and generate an outbreak alert (19) . Similarly, daily condemnation rates of pig carcasses at abattoirs in Canada were evaluated to detect aberrations in the data stream that may signal a disease outbreak. Aberrations detected in a retrospective analysis coincided with several documented disease outbreaks in swine, thus demonstrating the potential timeliness of a syndromic surveillance system based on abattoir data streams (59) . Finally, Guernier et al. (54) found that Google search trends could be used to track the occurrence of tick paralysis in companion animals in Australia, and certain search terms could potentially be used as early indicators of high-risk periods. These examples highlight the potential for high velocity and high volume data to enhance surveillance capabilities. Such applications may be particularly relevant for syndromic surveillance, where the causative agent may not be identified and the focus of the analysis is on tracking suites of clinical signs, or a syndrome, that may be associated with an endemic or emerging disease (49) .