Author:

Keywords:

Science & Technology, Life Sciences & Biomedicine, Physical Sciences, Health Care Sciences & Services, Mathematical & Computational Biology, Medical Informatics, Statistics & Probability, Mathematics, Bayesian statistics, commensurate prior, historical data, meta-analytic-predictive prior, power prior, ACUTE MYELOID-LEUKEMIA, COLONY-STIMULATING FACTOR, POWER PRIOR DISTRIBUTIONS, INFORMATION, MODELS, CYTARABINE, INDUCTION, PRIORS, Adult, Bayes Theorem, Data Interpretation, Statistical, Databases, Factual, Female, History, 21st Century, Humans, Leukemia, Myeloid, Acute, Male, Middle Aged, Outcome Assessment, Health Care, Randomized Controlled Trials as Topic, 0104 Statistics, 1117 Public Health and Health Services, 4202 Epidemiology, 4905 Statistics

Abstract:

Data of previous trials with a similar setting are often available in the analysis of clinical trials. Several Bayesian methods have been proposed for including historical data as prior information in the analysis of the current trial, such as the (modified) power prior, the (robust) meta-analytic-predictive prior, the commensurate prior and methods proposed by Pocock and Murray et al. We compared these methods and illustrated their use in a practical setting, including an assessment of the comparability of the current and the historical data. The motivating data set consists of randomised controlled trials for acute myeloid leukaemia. A simulation study was used to compare the methods in terms of bias, precision, power and type I error rate. Methods that estimate parameters for the between-trial heterogeneity generally offer the best trade-off of power, precision and type I error, with the meta-analytic-predictive prior being the most promising method. The results show that it can be feasible to include historical data in the analysis of clinical trials, if an appropriate method is used to estimate the heterogeneity between trials, and the historical data satisfy criteria for comparability.