Incorporating external control data into randomized clinical trials (RCTs) may enhance trial efficiency and reduce the required sample size, according to a recent study published in BMC Medical Research Methodology. The research focused on solid tumors, utilizing data from two trials in non-small cell lung cancer (NSCLC) to evaluate the impact of integrating external control data on treatment effect estimation.
The study re-analyzed data from two RCTs, supplementing the control arms with external control data. Baseline characteristics were generally similar between the RCT and external control groups, although some differences were noted in gender and chemotherapy usage. Five different methods were used to incorporate external control data, including pooled Cox regression, propensity score matching, and Bayesian approaches such as fixed power prior and Conditional Weighted Average Power Prior (CWAPP).
Impact on Overall Survival Analysis
In Trial 1, which initially showed non-statistically significant overall survival (OS), incorporating external control data shifted the hazard ratio estimates closer to 1, suggesting a more neutral treatment effect. All methods using external controls produced confidence intervals with smaller widths. Specifically, the CWAPP method estimated subject-specific weights, averaging 0.53 after transformation, and contributed data from approximately 458 external control subjects out of a possible 869.
Trial 2, which demonstrated statistically significant OS in the original analysis, showed that incorporating external control data using pooled Cox, propensity score matching, and CWAPP further moved the hazard ratio estimate away from the null hypothesis. Again, all methods narrowed the confidence intervals. The CWAPP method in this trial contributed data from about 130 external control subjects out of 248.
Simulation Studies and Type 1 Error
Bootstrap analyses emulating trial scenarios were conducted to assess the performance of each method. In Trial 1 simulations, all five methods provided treatment hazard ratios consistent with the original RCT findings, but the Bayesian methods, particularly CWAPP, yielded noticeably smaller confidence intervals. All methods controlled the bootstrap-estimated type-1 error at the nominal level, with CWAPP achieving a type-1 error rate of 0.03.
Similar simulations for Trial 2 showed consistent treatment hazard ratios across all methods. The methods utilizing external controls demonstrated higher statistical power than the traditional Cox model. Both Bayesian methods exhibited higher power compared to frequentist methods, with CWAPP achieving the highest power (0.89 versus 0.80 for the fixed power prior method).
CWAPP Method Shows Promise
The CWAPP method consistently provided the most precise estimates, indicated by the smallest average confidence interval widths, and the highest statistical power in simulation studies. This suggests that CWAPP could be a valuable tool for optimizing trial designs and potentially accelerating drug development in solid tumors.
"Incorporating external control data, especially via Bayesian methods like CWAPP, can optimize trial design," the researchers noted. The study highlights the potential of leveraging external data sources to enhance the efficiency and informativeness of clinical trials, ultimately benefiting patients by expediting the development of new therapies.
