Design of a Bayesian adaptive phase 2 proof-of-concept trial for BAN2401, a putative disease-modifying monoclonal antibody for the treatment of Alzheimer's disease

Abstract

Introduction: Recent failures in phase 3 clinical trials in Alzheimer's disease (AD) suggest that novel approaches to drug development are urgently needed. Phase 3 risk can be mitigated by ensuring that clinical efficacy is established before initiating confirmatory trials, but traditional phase 2 trials in AD can be lengthy and costly.

Methods: We designed a Bayesian adaptive phase 2, proof-of-concept trial with a clinical endpoint to evaluate BAN2401, a monoclonal antibody targeting amyloid protofibrils. The study design used dose response and longitudinal modeling. Simulations were used to refine study design features to achieve optimal operating characteristics.

Results: The study design includes five active treatment arms plus placebo, a clinical outcome, 12-month primary endpoint, and a maximum sample size of 800. The average overall probability of success is ≥80% when at least one dose shows a treatment effect that would be considered clinically meaningful. Using frequent interim analyses, the randomization ratios are adapted based on the clinical endpoint, and the trial can be stopped for success or futility before full enrollment.

Discussion: Bayesian statistics can enhance the efficiency of analyzing the study data. The adaptive randomization generates more data on doses that appear to be more efficacious, which can improve dose selection for phase 3. The interim analyses permit stopping as soon as a predefined signal is detected, which can accelerate decision making. Both features can reduce the size and duration of the trial. This study design can mitigate some of the risks associated with advancing to phase 3 in the absence of data demonstrating clinical efficacy. Limitations to the approach are discussed.

Keywords: Adaptive trial; Alzheimer's disease; Bayesian analysis; Clinical trial simulation; Interim analysis; Monoclonal antibody.

Fig. 1

Simulating futility boundaries in multiple dose and effect scenarios. Futility boundaries ranging from 2.5% to 15% were simulated for each scenario. The results for two scenarios are shown: (A) null scenario and (B) a dose-response scenario in which one dose has a robust effect. Robust indicates a dose-response in which the percentage reduction in decline relative to placebo for the 2.5-mg bimonthly, 5-mg bimonthly, 10-mg bimonthly, 5-mg monthly, and 10-mg monthly doses are 17%, 33%, 50%, 17%, and 33%, respectively. Null scenario simulations showed that with a boundary of 15%, the cumulative probability of declaring futility at the 13th IA would be 54% (A). However, using the same boundary, futility could be declared 13% of the time in a scenario where a single dose had a robust response (B), which was considered too risky. A boundary of 2.5% would reduce the probability of declaring futility to nearly zero in the event of a robust response for one dose (B) but would only permit stopping for futility in the null scenario 13% of the time (A). A boundary of 7.5% would permit stopping for futility 32% of the time in the null scenario (A), while only incorrectly declaring futility 4% of the time when one dose actually had a robust effect (B). Based on these simulation results, a futility boundary of 7.5% was chosen as providing the most acceptable balance of trial efficiency and risk. For the first three IAs (i.e., at 196, 250, and 300 subjects randomized), a more conservative futility boundary of 5% was chosen to further reduce the possibility of inappropriately stopping early for futility at very early time points in the trial, when the decision would be based on more limited data.

Fig. 2

Simulating early success boundaries in multiple dose and effect scenarios. Early success boundaries ranging from 85% to 99% were simulated for each scenario. The results for two scenarios are shown: (A) null scenario and (B) dose-response scenario in which one dose has a robust effect. Robust indicates a dose-response in which the percentage reduction in decline relative to placebo for the 2.5-mg bimonthly, 5-mg bimonthly, 10-mg bimonthly, 5-mg monthly, and 10-mg monthly doses are 17%, 33%, 50%, 17%, and 33%, respectively. With a success boundary of 85%, the cumulative probability of declaring early success at the 13th IA was 79% if one dose had a robust effect (B), but this boundary would result in success being declared 29% of the time in the null scenario (A), which was considered too risky. A success boundary of 99% would reduce the incorrect declaration of success in the null scenario to 3% (A) but would only permit stopping early for success when one dose had a robust effect 28% of the time (B). A success boundary of 95% would declare early success 56% of the time when one dose had a robust effect (B), while incorrectly declaring success 16% of the time in the null scenario (A). Based on these simulation results, an early success boundary of 95% was chosen as providing the most acceptable balance of trial efficiency and risk.

Fig. 3

One simulation of a robust dose-response scenario, in which the trial is stopped early for success at 550 subjects. Robust dose-response scenario used is the linear dose-response scenario with the largest response having ≥50% reduction in decline relative to placebo. During the study, IAs are carried out after 196 subjects are randomized (A), after 250 subjects, and every 50 subjects thereafter (B) until 800 subjects are recruited, after which IAs are carried out every 3 months, until 52-week data are available for all subjects. Based on the results of each IA, the study can be stopped early for success or futility. If the study continues, the randomization ratios are adapted based on the effects of each dose on the ADCOMS (C). If the study achieves early success, recruitment is stopped (D), and the study continues with enrolled subjects until 52-week data are collected for all subjects (E). Robust indicates a dose-response in which the percentage reduction in decline relative to placebo for the 2.5-mg bimonthly, 5-mg bimonthly, 10-mg bimonthly, 5-mg monthly, and 10-mg monthly doses are 17%, 33%, 50%, 17%, and 33%, respectively.