Our primary results relied on Cox regression modeling with propensity score matching. As to the ongoing question of the relative merits of propensity score versus conventional
regression approaches, systematic reviews (26,31) have found that the two approaches generate similar parameter estimates and patterns of statistical significance in a large majority of instances (~85%–90%), with propen-sity score matching methods tending to yield more conservative results—a pattern we found in our head-to-head comparisons of the two approaches (see Table S1 in the online data supplement). At this time, however, there do not appear to be any criterion standards to determine whether either of the approaches provides a truer result (26).