Using Bayesian pre-trial simulations to optimize the design of adaptive clinical trials in childhood nephrotic syndrome

Background

Randomized controlled trials (RCTs) are often infeasible in rare pediatric diseases. Adaptive trials can increase trial efficiency while maintaining scientific validity. Our aim was to determine the optimal design of a Bayesian adaptive RCT in childhood nephrotic syndrome using simulation.

Methods

We used simulation to evaluate candidate Bayesian adaptive clinical trial designs for a planned non-inferiority RCT comparing low-dose vs. standard-dose steroids for childhood nephrotic syndrome relapses. Each design had a unique combination of adaptive settings (stopping thresholds, futility margin, initial recruitment, and interim analysis frequency). We simulated 10,000 RCTs for each design to estimate operating characteristics (power, type 1 error rate, and mean sample size). The best designs were tested under plausible RCT conditions (varying treatment effect, recruitment rate, and prior distributions).

Results

We simulated 10,000 trials for each of 540 adaptive RCT designs with unique combinations of adaptation settings (5.4 million simulated trials). For the top three designs, we simulated another 10,000 trials under 40 different RCT conditions (1.2 million simulated trials). The optimal trial design was associated with the lowest mean sample size, smallest probability of an inconclusive trial, and a type 1 error rate < 5 %. Compared to a frequentist RCT, using this Bayesian adaptive design with an informative prior decreased sample size by 71 % (n = 198 vs. n = 682).

Conclusions

Bayesian trial simulation was used to optimize the design of an adaptive RCT in childhood nephrotic syndrome, lowering estimated sample size. Adaptive designs can reduce barriers to conducting RCTs in rare pediatric diseases.