Keywords: machine learning, treatment effect heterogeneity, comparative effectiveness research
Postmarket comparative effectiveness and safety analyses of therapeutic treatments typically involve large observational cohorts. We propose robust machine learning estimation techniques with parameters defined in a marginal structural model for treatment effect heterogeneity in implantable medical device evaluations where there are more than two unordered treatments. We isolate the effects of individual drug-eluting stents as well as effects of manufacturer stents on a composite outcome for a priori specified effect modifiers. This flexible approach accommodates a large number of covariates from clinical databases in the estimation of a complex target parameter. Data from the Massachusetts Data Analysis Center (Mass-DAC) percutaneous coronary intervention cohort is used to assess the composite outcome of 10 drug-eluting stents and of 3 manufacturers among adults implanted with at least one drug-eluting stent in Massachusetts with effect modification by sex, diabetes status, and chronic renal failure status.