Modeling and validating Bayesian accrual models on clinical data and simulations using adaptive priors

Abstract

Slow recruitment in clinical trials leads to increased costs and resource utilization, which includes both the clinic staff and patient volunteers. Careful planning and monitoring of the accrual process can prevent the unnecessary loss of these resources. We propose two hierarchical extensions to the existing Bayesian constant accrual model: the accelerated prior and the hedging prior. The new proposed priors are able to adaptively utilize the researcher’s previous experience and current accrual data to produce the estimation of trial completion time. The performance of these models, including prediction precision, coverage probability, and correct decision-making ability, is evaluated using actual studies from our cancer center and simulation. The results showed that a constant accrual model with strongly informative priors is very accurate when accrual is on target or slightly off, producing smaller mean squared error, high percentage of coverage, and a high number of correct decisions as to whether or not continue the trial, but it is strongly biased when off target. Flat or weakly informative priors provide protection against an off target prior, but are less efficient when the accrual is on target. The accelerated prior performs similar to a strong prior. The hedging prior performs much like the weak priors when the accrual is extremely off target, but closer to the strong priors when the accrual is on target or only slightly off target. We suggest improvements in these models and propose new models for future research.

A novel evaluation of optimality for randomized controlled trials

“…Needing an accrual rate goal, we use the accrual model in Gajewski, Simon, and Carlson (2008) and Jiang, Simon, Mayo, and Gajewski (2015). Define the average time elapsed between subsequent participants accrued to trial as 1/λ (in weeks).…”

Real-time prediction of clinical trial enrollment and event counts: A review

“…Gajewski in 2008 [21] proposed the use of interim data to refine enrollment predictions, a technique that appeared earlier in [46] (explicitly) and in [6] (implicitly). Jiang et al studied a range of prior distributions, including "adaptive" (data-based) priors, for improving Bayesian accrual predictions [25].…”

Bayesian accrual modeling and prediction in multicenter clinical trials with varying center activation times

“…The prior distribution for μ is given by the inverse gamma where J is the total number of planned centers and T Site is the length of the planned period of center activations. Similar to the approach of Jiang et al, 12 Q is the prior probability that the study will successfully activate all centers in the planned timeframe and is measured on a 0‐1 scale. Additionally, Q is totally based on prior knowledge and experience, which could be modified at any time during the accrual process.…”

“…As data collection progresses, the weight will shift from prior to the observed data. Jiang et al 12 applied this prior to random accrual times for patients in a single‐center trial. For a single‐center trial, it also can be treated as all the centers have identical center activation time under the multicenter scenario.…”

“…Jiang et al 12 proposed a simple method to predict completion time via a single‐center time prediction model. Under this scenario, a multicenter trial could be treated as a large single‐center study.…”