Abstract

In observational longitudinal studies, confounding can significantly affect the estimation of causal associations. Several methods, mainly based on propensity score analysis, are increasingly employed to mitigate confounding bias in these studies and estimate treatment effects. Additional challenges arise in observational longitudinal studies, particularly regarding the potential presence of time-varying exposures and confounders. Such problems can render conventional causal inference methods ineffective, especially when multiple related outcome measures are involved. This chapter applies propensity score matching (PSM), inverse probability of treatment weighting (IPTW), and propensity score subclassification methods in real-life data applications. We extend the IPTW methods to estimating causal effects in longitudinal observational studies with time-varying exposures and multivariate outcomes. The analysis of 2010 and 2015–2016 Malawi Demographic Health Survey (MDHS) datasets revealed differing estimates of maternal effects on birth weight using three propensity methods: propensity score matching (PSM), inverse probability of treatment weighting (IPTW), and subclassification. While PSM and IPTW showed a negative significant effect, subclassification indicated a positive association. All methods linked maternal HIV with increased prenatal deaths. For the 2015–2016 MDHS, they indicated reduced birth weight and higher perinatal death rates for children born to HIV-infected mothers. Additionally, a bivariate MSM-IPTW analysis found smaller estimated effects of HIV awareness on condom use and multiple sexual partners compared to a univariate MSM-IPTW model.

Original language	en
Pages (from-to)	283-312
Publication status	Published - 2025