Abstract
ObjectiveTo improve the ability of syndromic surveillance systems to detectunusual events.IntroductionSyndromic surveillance systems are used by Public Health England(PHE) to detect changes in health care activity that are indicative ofpotential threats to public health. By providing early warning andsituational awareness, these systems play a key role in supportinginfectious disease surveillance programmes, decision making andsupporting public health interventions.In order to improve the identification ofunusualactivity, wecreated new baselines to modelseasonally expectedactivity inthe absence of outbreaks or other incidents. Although historicaldata could be used to model seasonality, changes due to publichealth interventions or working practices affected comparability.Specific examples of these changes included a major change in theway telehealth services were provided in England and the rotavirusvaccination programme introduced in July 2013 that changed theseasonality of gastrointestinal consultations. Therefore, we needed toincorporate these temporal changes in our baselines.MethodsWe used negative binominal regression to model daily syndromicsurveillance, allowing for day of week and public holiday effects.To account for step changes in data caused by changes in healthcaresystem working practices or public health interventions we introducedspecific independent variables into the models. Finally, we smoothedthe regression models to provide short term forecasts of expectedtrends.The new baselines were applied to PHE’s four syndromicsurveillance systems for daily surveillance and public-facing weeklybulletins.ResultsWe replaced traditional surveillance baselines (based on simpleaverages of historical data) with the regression models for dailysurveillance of 53 syndromes across four syndromic surveillancesystems. The improved models captured current seasonal trends andmore closely reflected actual data outside of outbreaks.ConclusionsSyndromic surveillance baselines provide context forepidemiologists to make decisions about seasonal disease activity andemerging public health threats. The improved baselines developedhere showed whether current activity was consistent with expectedactivity, given all available information, and improved interpretationwhen trends diverged from expectations.