Advances in Epidemiological Methods and Utilisation of Large Databases: A Methodological Review of Observational Studies on Central Nervous System Drug Use in Pregnancy and Central Nervous System Outcomes in Children

Introduction Studies have used various epidemiological approaches to study associations between central nervous system (CNS) drug use in pregnancy and CNS outcomes in children. Studies have generally focused on clinical adverse effects, whereas variations in methodologies have not received sufficient attention. Objective Our objective was to review the methodological characteristics of existing studies to identify any limitations and recommend further research. Methods A systematic literature search was conducted on observational studies listed in PubMed from 1 January 1946 to 21 September 2017. Following independent screening and data extraction, we conducted a review addressing the trends of relevant studies, differences between various data sources, and methods used to address bias and confounders; we also conducted statistical analyses. Results In total, 111 observational studies, 25 case–control studies, and 86 cohort studies were included in the review. Publications dating from 1978 to 2006 mainly focused on antiepileptic drugs, but research on antidepressants increased from 2007 onwards. Only one study focused on antipsychotic use during pregnancy. A total of 46 studies obtained data from an administrative database/registry, 20 from ad hoc disease registries, and 41 from ad hoc clinical samples. Most studies (58%) adjusted the confounding factors using general adjustment, whereas only a few studies used advanced methods such as sibling-matched models and propensity score methods; 42 articles used univariate analyses and 69 conducted multivariable regression analyses. Conclusion Multiple factors, including different study designs and data sources, have led to inconsistent findings in associations between CNS drug use in pregnancy and CNS outcomes in children. Researchers should allow for study designs with clearly defined exposure periods, at the very least in trimesters, and use advanced confounding adjustment methodology to increase the accuracy of the findings.