(35) Approaches that use historical controls to meet modern needs in cystic fibrosis clinical trials

Background: CF transmembrane conductance regulator protein (CFTR) modulators are effective in reducing the disease burden of CF, and the overall proportion of patients who are eligible for CFTR modulators is just over 90% with the expansion of eligibility to those with a single copy of a F508del mutation. This broad success creates logistical challenges for clinical trials of new CF therapeutics for the small number of patients that remain ineligible for modulators. New logistically feasible trial designs must be developed to assess efficacy of CF therapeutics and CFTR restorative therapies. Novel study designs that utilize readily available external data sources, such as previous trial controls, need to be rigorously developed and understood to meet these modern challenges.

Methods: In the context of a pulmonary exacerbation simulation study, we will examine an adaptive trial design and three analytic approaches that use historical data. The adaptive trial methodology contains a planned interim analysis at which congruency between the historical and active controls is assessed. If congruent, historical controls will be used in the analysis to reduce the need for active controls and reduce the active trial sample size. If incongruent, the historical controls are thrown away and more active controls are recruited. The three analytic methods include inverse probability weighting (IPW), propensity score-based power priors, and commensurate priors. We will assess a simulated time to event outcome with Poisson regression and compare analytic approaches regarding their ability to accurately estimate treatment effect and to reduce required active trial sample size.

Results: Early, limited simulation shows that the adaptive trial design and all three analytic approaches can improve over naïve approaches under certain conditions. IPW can achieve unbiased results when most covariates measuring active trial and historical population differences were available. A hazard ratio (HR) of 0.63 (0.50, 0.80) was obtained when historical participants were used exclusively as the control arm. The expected HR was 0.55 (0.34, 0.86). Additionally, failure to adjust for differences between current and historical study populations results in biased estimates of efficacy and poor coverage. All methods will be compared in terms of coverage, power, and bias.

Conclusions: Evaluation of trial designs and analytic approaches that leverage historical data is critical as these tools will be necessary to meet the challenges modern CF clinical trials. As traditional placebo-controlled trials become increasingly less feasible, it will be critical for researchers to make use of efficient and accurate alternative designs.

Acknowledgements:

References: