Observational studies of prenatal antidepressant safety are hindered by methodological concerns, including susceptibility to surveillance bias. Some studies address potential bias by using alternative strategies to operationalize study comparison groups. In a meta-analysis of the association between prenatal antidepressant exposure and autism risk, the authors examined the utility of comparison group operationalization in reducing surveillance bias.A systematic search of multiple databases through August 2017 was conducted, selecting controlled observational studies of the association of prenatal antidepressant exposure with autism. Study quality was assessed using the Newcastle-Ottawa Scale. Random-effects meta-analysis produced summary effect measures with 95% confidence intervals stratified by comparator group composition, antidepressant class, and trimester of exposure.Fourteen studies were included, with 13 reporting results using a population-based comparison group, five using a psychiatric control group, and four using a discordant-sibling control group. Eight of the 14 studies were rated poor because of inadequate control for prenatal depression and maternal ethnicity. Autism risk estimates after prenatal exposure to any antidepressant were decidedly different for population-based designs (hazard ratio=1.42, 95% CI=1.18, 1.70; odds ratio=1.58, 95% CI=1.25, 1.99) compared with psychiatric control (hazard ratio=1.14, 95% CI=0.84, 1.53; odds ratio=1.24, 95% CI=0.93, 1.66) and discordant-sibling (hazard ratio=0.97, 95% CI=0.68, 1.37; odds ratio=0.85, 95% CI=0.54, 1.35) designs. Findings for prenatal exposure to selective serotonin reuptake inhibitors were similar. Meta-regression of population-based studies demonstrated that despite statistical adjustment, ethnicity differences remained a significant source of study heterogeneity.In this meta-analysis, neither psychiatric control nor discordant-sibling designs supported an association between prenatal antidepressant exposure and autism. Discordant-sibling designs effectively addressed surveillance bias in pharmacovigilance reports derived from national registries and other large databases.