On the Two-Step Hybrid Design for Augmenting Randomized Trials Using Real-World Data

Area of Science:

• Clinical Trials Methodology
• Biostatistics
• Real-World Data Analysis

Background:

• Hybrid clinical trials integrate real-world data (RWD) with randomized controlled trials (RCTs).
• These trials are crucial for rare diseases where patient recruitment is challenging.
• A key assumption is the exchangeability of RWD and RCT control arms, which if violated, can introduce bias and affect statistical accuracy.

Purpose of the Study:

• To propose and evaluate novel methods for controlling Type I error rates in hybrid clinical trials.
• To assess the performance of these methods under varying degrees of exchangeability between RWD and RCT data.
• To compare the proposed methods against existing approaches like the Yuan et al. (2019) and Bayesian power prior methods.

Main Methods:

• Four new methods were developed to control Type I error under the exchangeability assumption.
• Methods involve variance estimation, numerical critical value determination, and Type I error rate splitting.
• The performance was evaluated using a hypothetical amyotrophic lateral sclerosis (ALS) scenario, assessing Type I error and statistical power.

Main Results:

• The proposed methods and the Bayesian power prior approach effectively control Type I error and increase power when exchangeability holds.
• The Yuan et al. (2019) method showed an increased Type I error.
• When exchangeability is violated, all methods struggle; however, the proposed methods demonstrate limited Type I error inflation (6-8%) compared to others.

Conclusions:

• The proposed methods offer robust Type I error control in hybrid trials under the exchangeability assumption.
• These methods, along with the Bayesian power prior, enhance statistical power when RWD and RCT data are exchangeable.
• The proposed methods provide a more controlled inflation of Type I error when the exchangeability assumption is not met, offering a safer approach for hybrid trial design.