[Influence of data quality on early warning sensitivity of syndromic surveillance system based on medical...

Objective: To evaluate the influence of data quality on the sensitivity of early warning syndromic surveillance system based on medical institutions in Qianjiang, Hubei province and explore the relationship between data quality and sensitivity of early warning of the system. Methods: The delay reporting rate and underreporting rate were calculated for the evaluation of the data quality. Data obtained from semi-synthetic simulated outbreak and area under the curve (AUC) were used in combination to test the sensitivity of early warning of various models and select the optimal model. Time-series generalized additive model (GAM) was used to analyze the curve fitting and threshold effect between data quality and early warning sensitivity of the system. Results: A total of 179 905 cases were reported from April 1, 2012 to January 31, 2014, in which 8 744 were not reported timely (16.45%). Averagely 416 reporting were delayed in each month. There were 2 566 cases which were underreported (4.83%). Compared with other early warning models, i.e. Cumulative Sum (CUSUM), Shewhart, Exponentially Weighted Moving Average (EWMA), Early Aberration Reporting System (EARS-3C), the MA model had the maximum area under the curve (AUC=0.93), and the difference was significant (P<0.001). The early warning sensitivity ranged from 84.89% to 97.25% during the operation period of the syndromic surveillance system. Underreporting had influence on early warning sensitivity, when underreporting rate was over 2.78%, the sensitivity would decrease obviously. No obvious associations were observed between the delay reporting rate and early warning sensitivity of the system. Conclusion: The data quality had influence on the early warning sensitivity of the syndromic surveillance system based on medical institution in Qianjiang. In the context of this study, underreporting had the main influence on the sensitivity of early warning.