Monitoring potential adverse event rate differences using data from blinded trials: the canary in the coal mine

Abstract

The development of drugs and biologicals whose mechanisms of action may extend beyond their target indications has led to a need to identify unexpected potential toxicities promptly even while blinded clinical trials are under way. One component of recently issued FDA rules regarding safety reporting requirements raises the possibility of breaking the blind for pre-identified serious adverse events that are not the clinical endpoints of a blinded study. Concern has been expressed that unblinding individual cases of frequently occurring adverse events could compromise the overall validity of the study. However, if external information is available about adverse event rates among patients not receiving the test product in populations similar to the study population, then it may be possible to address the potential for elevated risk without unblinding the trial. This article describes a Bayesian approach for determining the likelihood of elevated risk suitable binomial or Poisson likelihoods that applies regardless of the metric used to express the difference. The method appears to be particularly appropriate for routine monitoring of safety information for project development programs that include large blinded trials. Copyright © 2016 John Wiley & Sons, Ltd.

Bayesian detection of potential risk using inference on blinded safety data

“…While there is no unique way to formulate the Bayesian problem for blinded analysis, we think our modeling approach is particularly advantageous as it transparently allows formulation of the prior that reflects the true knowledge or uncertainty of the relative risk and also allows for conditional inference with a fixed value or a range of background rates. Similarly, as Gould and Wang and others, we assumed a constant hazard to use a Poisson model, and future work may be needed when the constant hazard assumption does not hold. We also added the functionality in our model of allowing for different allocation ratios, and we more directly incorporated considerations of exposure.…”

“…The proposed approach explicitly models and puts a prior on the signal parameter (relative risk) and thus enables direct inference on signal detection. While there is no unique way to formulate the Bayesian problem for blinded analysis, [4][5][6] we think our modeling approach is particularly advantageous as it transparently allows formulation of the prior that reflects the true knowledge or uncertainty of the relative risk and also allows for conditional inference with a fixed value or a range of background rates. Similarly, as Gould and Wang 6 and others, we assumed a constant hazard to use a Poisson model, and future work may be needed when the constant hazard assumption does not hold.…”

“…In the case of unblinded data, the Data and Safety Monitoring Board could use a similarly derived Bayesian safety boundary for the difference of AESIs rates for the treatment and control groups. After Yao et al, a few Bayesian approaches have been proposed for blinded monitoring of adverse events . Wen et al also proposed an approach using a combined event rate so that the study would remain blinded.…”

“…Schnell and Ball proposed an approach using separate priors for the treatment group rate (weak prior) and control group rate (strong prior) with a Poisson model which is traditionally used for modeling rare events. Gould and Wang constructed “posterior distributions for metrics of relative treatment effect based on blinded data given (i) current values of the number of patients in a trial; (ii) the total number of patients reporting an adverse event; (iii) a prior distribution for the control group event rate that reflects what is known about the event rate from external data; and (iv) a prior distribution for the particular metric used.”…”

“…After Yao et al, a few Bayesian approaches have been proposed for blinded monitoring of adverse events. [4][5][6] Wen et al 4 also proposed an approach using a combined event rate so that the study would remain blinded. Schnell and Ball 5 proposed an approach using separate priors for the treatment group rate (weak prior) and control group rate (strong prior) with a Poisson model which is traditionally used for modeling rare events.…”