Each week, we will be publishing layman’s abstracts of new articles from our prestigious portfolio of journals in statistics. The aim is to highlight the latest research to a broader audience in an accessible format.
The article featured today is from Pharmaceutical Statistics, with the full article now available to read on Early View here.
Harun, N, Liu, C, Kim, M‐O. Critical appraisal of Bayesian dynamic borrowing from an imperfectly commensurate historical control. Pharmaceutical Statistics. 2020; 1– 13. https://doi.org/10.1002/pst.2018
Historical data may be available before a large trial from patient registries, observational/natural history study, small pilot/randomized trials, and/or large trials on “standard care” group. If the studies are similar, incorporating historical information may increase the probability to detect treatment effect when it truly exists (power), conclude the trial early with correct inference, reduce both sample size and trial cost, and most importantly avoid exposure of patients to the inferior treatment/no treatment (placebo). On the contrary, borrowing information from dissimilar studies have been reported to inflate the probability to erroneously detect a treatment effect (type I error) and reduced power. Bayesian inference assigns uncertainty on the true treatment effect rather than assuming it to be fixed in contrast to the classical Frequentist approach. Regulatory agencies require calibration of Bayesian designs to meet Frequentist operating characteristics, that is, keeping type I error and power fixed. Calibration results in implicit borrowing of historical information and thus designs that do not borrow are likewise negatively impacted. Based on simulated data from two motivating stroke trials, the performance of a selective response-adaptive design while borrowing was evaluated and were compared with a design that does not incorporate historical information under similar assumptions. Increased power, greater probability to stop early correctly, and smaller average sample size were observed with borrowing compared to not borrowing irrespective of how similar trial outcomes were. The type I error was inflated while both borrowing and not borrowing using dissimilar historical controls. Although slightly larger type I error inflation was noted with borrowing, there was greater probability to stop early correctly for efficacy, thus, favoring borrowing.