A syndromic surveillance system for clinical and non-clinical health data

Abstract

The data used in health care and public health monitoring and surveillance are likely to be nonstationary and highly auto-correlated. These data may come from clinical and nonclinical sources, the latter being increasingly used in an attempt to supplement traditional surveillance. We propose an alternative approach applicable to multiple sets of health data that combines a method of preconditioning (ARMA filtering), a forecasting method (the method of analogues) and a monitoring method (the change-point). This procedure is applied to Influenza-Like Illness (ILI) surveillance in France using both clinical data (time series of weekly ILI incidence at the national level and in 22 regions) and nonclinical data (over-the-counter and prescribed weekly medication sales data at the national level). The outcomes show that multivariate monitoring provides an improved ability to recognize warning signals and drug sales data provide an indirect but earlier indication of influenza outbreaks. The proposed method has the advantage of being both practical and relatively simple to implement and could be easily adapted to develop adequate surveillance tools for specific diseases.

Keywords:

• Syndromic surveillance
• statistical process control
• change-point
• influenza-like illness
• time series model
• the Method of Analogues