Bridging differences across cohorts in the relationship between prenatal secondhand smoke exposure and birth weight: Transportability of effect estimates in the ECHO consortium

Abstract

Background and Aim: Transportability of effect estimates in epidemiology can potentially help investigators explain the differences in effect estimates for the same exposure-outcome relationships observed across different cohorts. In this study we aim to explain potential differences in the relationship between prenatal secondhand smoke exposure (SHS) and birth weight across different cohorts of the Environmental influences on Child Health Outcomes (ECHO) consortium that may be due to population differences in sociodemographic characteristics. Methods: We assessed transportability of effect estimates across 5 different cohorts in the ECHO consortium consisting of 6582 mother-offspring dyads. We first estimated individual cohort effects using Targeted Maximum Likelihood Estimation adjusting for several sociodemographic covariates. We then proceeded to estimate transported effects from one cohort to each of the remaining cohorts using a robust non-parametric estimation approach. We compared the transported effect estimates to the original effect estimates for these cohorts with any decrease in differences between cohort-specific estimates potentially attributable to different sociodemographic variable distributions. Results: Individual cohort effects associated with SHS exposure varied across the 5 cohorts with a range of strongly suggestive results for harmful effects of exposure (149.0 gram decrease in birthweight (95% confidence interval (CI): -254.4, -43.5), to results on the protective side of the null albeit with very wide confidence intervals. Transported effect estimates partially explained differences in 3 out of the 4 cohort pairs, explaining 7% to 97% of the differences in the effect point estimates. Conclusion: Differences in sociodemographic characteristics across different populations may partially explain differences observed in the relationship between prenatal SHS exposure and birth weight across cohorts in the ECHO consortium. This framework could have broader implications in in multi-cohort consortiums by aiding in the comparability of cross-cohort results. Keywords: transportability, causal inference, machine learning, secondhand smoke, birth outcomes