Randomized controlled trials are the gold standard for measuring the effect of a treatment or intervention. Unfortunately, it is not feasible to conduct a randomized controlled trial to test all research questions, whether due to cost, achieving sufficient subject sizes or when administering an arm of the trial would be unethical. To understand the effects of therapeutics, policy changes, and other interventions where it is not possible to administer a clinical trial, researchers have developed approaches that attempt to simulate clinical trials in observational data. Despite sophisticated statistical methodologies, it is not clear whether it is possible to reliably simulate a randomized controlled in observational data.
We aim to quantify one potential driver of these different results, differences between the clinical trial and real-world populations.
In Aim 1, we compare trials that have individual level data available to three real- world data sources.
In Aim 2, we develop methodologies to infer most likely individual-level statistics from aggregate trial statistics using real world data. Finally, in Aim 3 we compare neurological trials that do not release individual level data to real world data. We then estimate the transportability of treatment estimates across different populations including: the population eligible for the trial in RWD and the population ineligible for the trial but receiving the treatment in RWD. This allows for the study of indication drift and treatment heterogeneity. By uncovering differences between these groups, we may be able to identify groups that are underrepresented in clinical trials to help reduce healthcare disparities. The K99/R00 award will allow me to gain expertise in using regulatory sciences (with mentor Dr. Florence Bourgeois and advisor Dr. Deborah Schrag) for biologic discovery (with mentors Dr. Tianxi Cai and Dr. Isaac Kohane) within neurology (with mentors Dr. Page Pennell and Dr. Clemens Scherzer). My background in statistics, informatics, genetics, and machine learning with clinical data sources ideally positions me for the proposed project. The proposed training plan, mentoring and project will provide a strong foundation for a successful transition to independent research.
Clinical trial populations are a subset of the overall population and may be biased through several factors. To better understand disease, reduce healthcare disparities for underrepresented groups, assess drug approvals, and ensure safety as the indications for a drug expand it is critical to quantify how representative clinical trials are of the real-world population. Identifying enriched and underrepresented groups offers the opportunity to identify subpopulations who may have heterogeneous treatment effects and/or underlying physiologies.
