Developing a propensity score protocol for evaluating the oral health consequences of methamphetamine use

Conventional propensity-score matching strategies do not always produce covariate balance across treatment groups comparing a hard-to-reach population subclass to a representative sample from the general population. We describe techniques for combining propensity-score matching with stratification into propensity-score-based subclasses to produce samples with balanced distributions of characteristics from a cross-sectional cohort of methamphetamine users and nonusers from the 1999-2004 National Health and Nutrition Examination Survey. Resulting outcome estimates are weighted between subclasses for interpretable comparisons. We consider features of an approach facilitating between-group comparisons, including the matching ratio, number of subclasses, analysis strategy, and conceptual underpinnings of weight selection.