Using propensity score-based weighting in the evaluation of health management programme effectiveness

When the randomized controlled trial is unfeasible, programme evaluators attempt to emulate the randomization process in observational studies by creating a control group that is essentially equivalent to the treatment group on known characteristics and trust that the remaining unknown characteristics are inconsequential and will not bias the results. In recent years, adjustment procedures based on the propensity score, such as matching and subclassification, have become increasingly popular. A new technique that has particular appeal for evaluating health management programmes uses the propensity score to create a weight based on the subject's inverse probability of receiving treatment. This weighting mechanism removes imbalances of pre-intervention characteristics between treated and non-treated individuals, and is then used within a regression framework to provide unbiased estimates of treatment effects. This paper presents a non-technical introduction of this technique by illustrating its implementation with data from a recent study estimating the impact of a motivational interviewing-based health coaching on patient activation measure scores in a chronically ill group of individuals. Because of its relative simplicity and tremendous utility, propensity-score weighting adjustment should be considered as an alternative procedure for use with observational data to evaluate health management programme effectiveness.