Deriving Real World Insights to Inform Trials for Pregnant Individuals Using Electronic Health Records and Machine Learning

Pregnant women have been routinely excluded from trials due to ethical and safety concerns. Nearly 95% of clinical trials submitted to the FDA excluded pregnant women. By contrast, almost all pregnant women take at least one medication in pregnancy, but trial evidence on the safety and efficacy of these medications is largely missing. Such evidence gaps present challenges for clinicians who advise treatment for pregnant women, and patients who are making decisions in the absence of data to inform safety and efficacy in pregnancy. 1) The COVID-19 pandemic further escalated these gaps for pregnant women. Acquiring SARS-CoV-2 infection during pregnancy is associated with an increased risk of mortality and obstetric complications. SARS-CoV-2 infection acquired in pregnancy is not only associated with worse perinatal outcomes but also exhibits risks of getting post-acute sequelae of SARS-CoV-2 infection (PASC), or Long COVID. By contrast, all the initial COVID-19 vaccine trials and treatment studies, including Paxlovid, etc., excluded pregnant women, despite their higher risk of COVID-19 complications and risk of long-term complications. 2) Other examples are coexisting type 2 diabetes or living with obesity in pregnancy which are associated with increased risk of adverse pregnancy outcomes. Though glucagon-like peptide 1 (GLP-1) agonists and sodium-glucose co-transporter-2 (SGLT2) inhibitors are increasingly used in treating diabetes and obesity in the general population there is a paucity of data on safety in pregnancy limiting evidence-based decision making for this population, 3) Other examples include but are not limited to Selective Serotonin Reuptake Inhibitors (SSRIs) with perinatal mood and anxiety disorders affecting an estimated 10-20% of reproductive-aged females. Our proposal aims to bridge these critical evidence gaps for pregnant women with RWD and advanced machine learning methods, with applications including but not limited to generating real-world evidence on the safety and effectiveness of antivirals (e.g., Paxlovid), antidiabetic drugs (e.g., GLP-1, SGLT-2, metformin), and antidepressants (e.g., SSRIs) in pregnancy. To achieve our objectives, we will leverage the large-scale electronic health record (EHR) data from patients in New York City (~19 million) and Florida (~19 million). There are three specific aims for this study: (1) Develop NLP-assisted computable phenotyping algorithms and tools to extract pregnant cohorts, key patient characteristics, and outcomes relevant to pregnant trials, (2) Develop a federated target trial emulation framework for estimating the real-world safety and effectiveness outcomes of interventions from distributed CRNs, and (3) Develop an RWE-based trial design system consisting of causal AI and a prototype toolbox for assessing and optimizing eligibility criteria to inform pregnancy trials.