Evaluating health management programmes over time: application of propensity score-based weighting to longitudinal data

Health management programmes are generally evaluated as point treatment studies in which only a baseline and outcome measurement are used in the analysis, even when multiple observations for each individual are available. By summarizing observations into two distinct measurements the evaluator loses any ability to discern patterns of change in the outcome variable over time in relation to the intervention. There are several statistical models available to evaluate longitudinal data that are typically regression-like in form and designed to adjust for clustering at the individual level. Most evaluators of longitudinal studies tend to adjust for the effect of time-dependent confounding by including these covariates as independent variables in the model. However, this standard adjustment approach is likely to provide biased estimates. In this paper we describe the application of the propensity score-based weighting technique to longitudinal data to estimate the effect of treatment on an outcome. This method reweights each treatment pattern to represent the entire population at each time point and provides an unbiased treatment effect. We illustrate the technique using data from a disease management programme and demonstrate its superiority over standard analytical adjustments in correcting for time-dependent confounding for each time period under study.