Modelling the utility of body temperature readings from primary care consults for SARS surveillance in an army medical centre.

INTRODUCTION: There is interest in surveillance systems for outbreak detection at stages where clinical presentation would still be undifferentiated. Such systems focus on detecting clusters of syndromes in excess of baseline levels, which may indicate an outbreak. We model the detection limits of a potential system based on primary care consults for the detection of an outbreak of severe acute respiratory syndrome (SARS). MATERIALS AND METHODS: Data from an averaged-sized medical centre were extracted from the Patient Care Enhancement System (PACES) [the electronic medical records system serving the Singapore Armed Forces (SAF)]. Thresholds were set to 3 or more cases presenting with particular syndromes and a temperature reading of >or=38oC (T >or=38). Monte Carlo simulation was used to insert simulated SARS outbreaks of various sizes onto the background incidence of febrile cases, accounting for distribution of SARS incubation period, delay from onset to first consult, and likelihood of presenting with T >or=38 to the SAF medical centre. RESULTS: Valid temperature data was available for 2,012 out of 2,305 eligible syndromic consults (87.2%). T >or=38 was observed in 166 consults (8.3%). Simulated outbreaks would peak 7 days after exposure, but, on average, signals at their peak would consist of 10.9% of entire outbreak size. Under baseline assumptions, the system has a higher than 90% chance of detecting an outbreak only with 20 or more cases. CONCLUSIONS: Surveillance based on clusters of cases with T >or=38 helps reduce background noise in primary care data, but the major limitation of such systems is that they are still only able to confidently detect large outbreaks.