Using the Bayesian detection of potential risk using inference on blinded safety data (BDRIBS) method to support the decision to refer an event for unblinded evaluation

Abstract

In the Sponsor Responsibilities-Safety Reporting Requirements and Safety Assessment for IND and Bioavailability/Bioequivalence Studies: Draft Guidance for Industry (June 2021) the Food and Drug Administration recommends that sponsors develop a Safety Surveillance Plan as a key element of a systematic approach to safety surveillance and describes two possible approaches to assess the aggregate safety data. One approach regularly analyzes unblinded serious adverse events (SAEs) by treatment group. The alternative approach prespecifies estimated background rates for anticipated SAEs in the study population (e.g., myocardial infarctions in an older adult population). If the event rate in the blinded data from the study population exceeds a "trigger rate," then an unblinded analysis by treatment group is conducted. The Bayesian detection of potential risk using inference on blinded safety data (BDRIBS) method has been previously described and offers a quantitative approach for assessing blinded events. In this article we provide a procedural workflow for blinded review of safety data that is consistent with the unblinding "trigger approach" for aggregate safety review. In addition, this publication contextualizes the use of BDRIBS within the broader safety surveillance framework, extends the method to allow for multiple studies, and offers examples of its use in various settings via an R-Shiny application that allows for dynamic visualization and assessment.

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

“…This process is in line with the US FDA guideline for IND safety reporting. 2 On this note, it is worth emphasising the point also made in Waterhouse et al 8 that although the statistical methodology is one important aspect, blinded safety signal detection is a multidisciplinary activity that requires engagement from members in safety, clinical, epidemiology, statistics, programming, and potential other teams. Team inputs are crucial, particularly in the areas of selection of adverse events for monitoring, selection of relevant historical studies, setting up discrimination threshold values for flagging potential safety signals, and strategy for further review or unblinding if a potential signal is identified.…”

“…Bayesian methods for routine safety monitoring in clinical studies have been discussed in the literature with several different approaches. [5][6][7][8][9][10][11] Schnell et al discussed a Bayesian hierarchical exposure-time model, using separate priors for experimental treatment and control together with blinded data to make inferences on the rate of selected adverse events in the experimental treatment arm. They utilised a Poisson model, accounting for all observed blinded occurrences of the event.…”