In a recent article in the Archives, Kerns et al1 found that high school students of an urban public school district who occasionally used school-based health centers (SBHCs) had a roughly 33% lower dropout rate than non-SBHC users.
The apparent association between lower dropout rates and SBHC visits might have been exaggerated by their methods, which seem to have introduced time-dependent (or “immortal time”) bias. This bias, apparently common in clinical studies, arises from modeling a time-dependent treatment in a hazards regression as a fixed baseline covariate instead of as a time-varying covariate.2 In this case, Kerns et al ignored the timing of SBHC visits, modeling the treatment as a fixed covariate representing the frequency of the subject's SBHC visits, averaged over his or her entire high school career.