A Bayesian method for safety signal detection in ongoing blinded randomised controlled trials.

Area of Science:

• Clinical Trials
• Biostatistics
• Pharmacovigilance

Background:

• Sponsors must minimize participant risk through safety monitoring.
• FDA regulations mandate aggregate safety evaluation for Investigational New Drug (IND) applications, including blinded studies.
• Estimating the probability of higher adverse event rates in experimental versus control arms is crucial for safety.

Purpose of the Study:

• To develop a Bayesian statistical method for estimating adverse event rate differences in blinded clinical studies without unblinding.
• To enable reliable detection of safety signals for earlier risk mitigation.

Main Methods:

• A Bayesian approach was developed using a meta-analytic predictive prior for the control arm and an uninformative prior for the experimental arm.
• A mixture likelihood function was employed to account for patient allocation to either arm in ongoing blinded studies.
• The method's performance was evaluated using simulations, assessing its ability to detect safety signals.

Main Results:

• The Bayesian method reliably detects safety signals, with improved performance for larger sample sizes and more common adverse events.
• The model's performance is robust, showing minimal degradation with heterogeneous or non-constant hazards in historical data.
• The approach successfully estimates the probability of increased adverse event rates without requiring study unblinding.

Conclusions:

• This novel Bayesian method facilitates effective safety monitoring in ongoing blinded clinical studies, allowing for earlier risk identification and mitigation.
• The methodology can be adapted for informative priors on both arms or for incorporating predictive covariates.
• Effective safety monitoring requires integrating statistical methods with multidisciplinary medical judgment.