Leveraging Real-World Data, Including the Electronic Health Record, to Fill in Clinical Knowledge Gaps in Treating Pregnant People

Abstract

There is a substantial knowledge gap in our understanding of treatment for pregnant people due to their exclusion from clinical trials. Despite this, the majority of pregnant people take prescription medication. The electronic health record (EHR), with its data on longitudinal phenotypes, confounders, and prescription orders, emerges as a promising resource for investigating medical treatments during pregnancy.This dissertation addresses knowledge gaps regarding treatments for pregnant women through a comprehensive approach employing hypothesis-generating and testing methods on EHR data, with four primary aims. In Aim 1, we used propensity score matching to expedite the generation and prioritization of hypotheses regarding drug effect signals associated with preterm birth risk. Aim 2 validated drug effect signals, particularly the association between Selective Serotonin Reuptake Inhibitors and preterm birth risk, using traditional pharmacoepidemiology methods. Aims 1 and 2 present a streamlined workflow for detecting known and unknown drug effect signals and validating by traditional epidemiological methods. In Aim 3, we investigated the COVID-19 antithrombotic therapy guidelines for pregnant women, which lack evidence in favor of or against their clinical application. We examined the prophylactic anticoagulant prescription rate among hospitalized pregnant women with COVID-19 and its impact on risks related to coagulopathy, COVID-19, and maternal-fetal health outcomes. Surprisingly, a low (7.0%) prescribing rate for prophylactic anticoagulants was observed across healthcare systems. In Aim 4, we leveraged EHRs to assess comorbidities during pregnancy, exploring the likelihood of adverse maternal-fetal health outcomes among pregnant individuals with immune-mediated inflammatory diseases (IMIDs). The association between IMIDs and elevated risk of adverse pregnancy outcomes depended on the specific type of IMIDs and presence of comorbidities. Consequently, this dissertation introduces innovative approaches to aid in filling in the knowledge gap in treating pregnant individuals, focusing primarily on preterm birth, with potential applications to a broader spectrum of maternal-fetal outcomes requiring attention. Despite the rich longitudinal data EHR offered, we were limited by some inherent challenges when used for research purposes. Challenges included misclassification of variables, a range of good and bad standardization in the type and quality of data collected, and limited information derived from structured data. In corresponding chapters, we described these challenges and potential alternatives for future direction for each aim.