|| Checking for direct PDF access through Ovid
We commend Snowden et al1 for evaluating Oregon's 2011 hard-stop policy to limit elective early-term deliveries. We are concerned, however, that their results may not necessarily reflect the effects of the new policy. The before-and-after design used in the study (which tests whether the adverse outcome rate after the introduction of a new policy is significantly different than the rate before) does not allow the policy's effects to be disentangled from other influences on maternal and neonatal health.2 In particular, if the rate of an adverse outcome is rising steadily throughout the study period for reasons unrelated to the policy change, the average rate during the study's later years will be higher than the average rate during earlier years. Even if the new policy has no effects, a before-and-after analysis can estimate a significantly higher rate postpolicy. Regression adjustment will not prevent this bias if women's measured characteristics do not fully explain temporal changes in outcomes.Using 2008–2012 data from our previously published cohort of low-risk, term, planned cesarean deliveries,3 we can replicate Snowden's increased risk of chorioamnionitis when comparing rates before with those after 2011 (2.32% compared with 1.24%, P=.042). However, this increase cannot be attributed to a new policy, because no policy changes occurred at the institution during this period. Instead, the significant before-and-after difference reflects a steady temporal increase in chorioamnionitis (0.87%, 0.96%, 1.78%, 1.97% and 3.8% in 2008–2012, respectively). Before-and-after studies of pediatric medical emergency teams have shown similar bias, where significant reductions in in-hospital mortality after the introduction of the team were replicated in an institution with no pediatric medical emergency team.4Obstetric policies should be evaluated using approaches that are specifically intended to isolate policy effects from underlying time trends, such as time-series or difference-in-differences analyses.2,5 These designs can be implemented easily using standard statistical software and used with data in which only month and year of birth is available (eg, national vital statistics). Policy evaluations using rigorous methods are critical for supporting the uptake of policies that truly improve maternal and newborn health and revising policies that do not.