Abstract:
|
Propensity score (PS) methods (matching, covariate adjustment, subclassification, and inverse probability treatment weighting) have been widely applied in observational studies to address issues such as confounding, classification bias, and failure to abide by the intention to treat principle with binary treatment groups. However, the evaluation and application of PS methods with an ordinal treatment (e.g. antibiotics: none, 1-2 and >2) are limited. In this work, we first develop a general class of inverse probability treatment weighted estimators (IPTW) of the average treatment effects in this ordinal case. We then evaluate their performance through extensive simulations, in comparison with existing PS methods such as matching, subclassification and covariate adjustment. Finally, we apply the IPTW estimators to a population-based birth cohort study of mother-child dyads to assess the cumulative effect of environmental exposures of ordinal levels on the risk of childhood asthma.
|