333 Adapting an interrupted time series design to voluntarily reported surveillance data: Advantages of statistical interactions in reducing bias

Abstract

Objectives To describe the method we use to identify temporal associations between events such as changes in legislation and changes in the incidence of work-related ill-health (WRI) using surveillance data and to show some examples applying this method. Methods The Health and occupation reporting network (THOR) collects reports of work-related ill-health from clinical specialists. Previously we have published a method to analyse time trends in the incidence of WRI using a 2 level negative binomial regression model with beta distributed random effects1. The model also controls for calendar time, reporter type (monthly or annual reporter) and first month as a new reporter. One variable that influences reporting to the THOR surveillance scheme is the length of membership time i.e. reporters tend to report fewer cases after longer membership time resulting in an inherent downward trend in incidence. In an attempt to mitigate this effect, alongside other factors affecting trends in reporting that are not directly related to the incidence of WRI, we have employed a segmented interrupted time series design and included statistical interaction terms in the model. Briefly time periods describing the time periods pre and post-event, and groups representing cases and comparators are prospectively defined. Groups are usually defined by occupation and/or suspected agent. Comparisons are made of the estimated change in incidence per reporter according to inclusion or exclusion within a group. Results This method has been applied to estimate the effect of events anticipated to influence the incidence of short latency WRI (e.g. asthma, dermatitis). Examples will be shown. Conclusions THOR data can contribute to the evaluation of the impact of events such as changes in legislation or interventions.