Data-driven trial design: use of target trial emulation to evaluate eligibility criteria in asthma and COPD

Clinical trial eligibility criteria are crucial to ensuring valid estimation of therapeutic treatment effects and robust measurement of safety. However, highly restrictive criteria have a negative impact on study generalizability and patient accrual.

Following the target trial emulation framework, we applied a modified Trial Pathfinder approach to emulate dupilumab trials for chronic obstructive pulmonary disease (COPD; NCT03930732) and asthma (NCT02414854), using Optum Market Clarity electronic health records and annualized exacerbation rate ratio (AERR) as the primary endpoint. We attempted to match the rigor of the original trial effect estimates using doubly robust, augmented inverse probability-weighted estimates for AERR and corresponding confidence intervals using the bootstrap method. We computed E-values to assess the potential impact of unmeasured confounding and applied empirical assessments of improvement in the generalizability of the study population under data-driven eligibility criteria.

We identified dupilumab new-user cohorts for COPD (N = 388,051) and asthma (N = 915,154) in electronic health records. The number of original criteria successfully emulated was 29 (out of 47) for COPD and 21 (out of 38) for asthma. After application of all emulated eligibility criteria, cohort sample sizes were markedly reduced (COPD: N = 4,333; asthma: N = 115,761). Analysis with Shapley values retained nine criteria for COPD and 10 for asthma. After applying this data-driven set of criteria, cohort sample sizes increased (COPD: N = 118,739; asthma: N = 219,580) while AERR point estimates changed by less than 10% with a narrower confidence interval width.

Our findings demonstrate the use of trial emulation and Trial Pathfinder for data-driven identification of eligibility criteria for relaxation or removal in non-oncology therapeutic areas. Thorough assessment for unmeasured confounding is essential for credible causal inference. Lastly, the comparison of sample characteristics provided valuable insights into how data-driven eligibility criteria enhance the generalizability of clinical trials.