In Reply Retrospective studies relying on data derived from administrative claims and electronic health records often have incomplete capture of some patient-level covariates. These data are typically not captured to support research but to document clinical care and billing. Nevertheless, modern statistical techniques, such as sequential stratification used in our study, reduce biases induced by nonrandom treatment assignment.
We did not directly adjust for smoking status or borderline personality disorder, but imbalance in these variables across our 2 groups is likely reduced or eliminated because they are correlated with other variables (eg, age, sex, race/ethnicity, diabetes, and Diagnostic Cost Group) on which we did match patients. Evidence in support of this untestable assumption can be seen in Table 1 in our article in which many covariates (eg, marital status, coronary artery disease, alcohol abuse, and substance abuse) were well-balanced despite not matching on them.
Arterburn DE, Eid G, Maciejewski ML. Long-term Survival Following Bariatric Surgery in the VA Health System—Reply. JAMA. 2015;313(14):1474-1475. doi:10.1001/jama.2015.2586