Survey
* Your assessment is very important for improving the workof artificial intelligence, which forms the content of this project
* Your assessment is very important for improving the workof artificial intelligence, which forms the content of this project
International Biometric Society PHASE II TRIAL DESIGN WITH BAYESIAN RESPONSE ADAPTIVE COVARIATE BALANCED RANDOMIZATION USING LONGITUDINAL PATIENT OUTCOMES Tomoyoshi Hatayama*1,2 Satoshi Morita*3 1: Department of Biostatistics and Epidemiology, Graduate School of Medicine, Yokohama City University, Yokohama, Japan 2: Department of Biostatistics, ACRONET Corporation, Tokyo, Japan 3: Department of Biomedical Statistics and Bioinformatics, Kyoto University Graduate School of Medicine, Kyoto, Japan [Background] The response adaptive randomization (RAR) method is used to increase the number of patients assigned to more efficacious treatment arms in clinical trials. In many trials evaluating longitudinal patient outcomes, RAR methods based only on the final measurement may not benefit significantly from RAR due to its delayed initiation. Alternatively, initiation of RAR can be hastened by using longitudinal patient outcomes evaluated previous to the last measurement (longitudinal RAR). In addition, when treatments are compared, it is important to ensure any observed treatment effect is attributable to the treatment itself rather than to any particular patient characteristic. [Aim] We propose a Bayesian covariate balanced (CB) RAR method that improves RAR performance by accounting for longitudinal patient outcomes while controlling baseline covariates imbalance. [Methods] We use a Bayesian linear mixed effects model to analyze longitudinal continuous patient outcomes for calculating longitudinal RAR based allocation probability. To control the covariate imbalance during RAR, we use the idea of the biased coin design where the assignment that will result in the minimum covariate imbalance is given a higher probability [1]. We use a combination of the two probability of RAR based allocation probability and probability of biased coin design as the allocation probability for assigning more patients to presumably superior treatment while controlling covariate imbalance (longitudinal CBRAR). Additionally, it is known that RAR implementation can lead to excessively large patient allocation imbalances and may result in substantial loss of statistical power. Therefore, we aim to mitigate the loss of statistical power due to large patient allocation imbalances by embedding penalties into the longitudinal RAR based allocation probability calculation. We evaluated the operating characteristics of our proposed longitudinal CBRAR using extensive simulation. [Results] Simulation results showed that our proposed longitudinal CBRAR method assigned more patients to the presumably superior treatment arm compared to the RAR method based only on the final measurement and longitudinal CBRAR method achieved smaller covariate imbalance than any randomization method without covariate adjustment. In addition, the embedded penalty effectively worked to prevent extreme patient allocation imbalances. Although our longitudinal CBRAR method decreased the statistical power compared to a completely randomization (CR) method, our method more desirably performed in reducing the number of patients who would not be benefited. [Conclusion] This study suggests that accounting for longitudinal outcome measurements in calculating patient allocation probabilities may lead to improving the performance of a RAR method. Moreover, proposed method improves RAR by controlling covariate imbalance between treatment arms. [Reference] 1. Ning J, Huang X. Response-adaptive randomization for clinical trials with adjustment for covariate imbalance. Stat Med 2010; 29: 1761-68. International Biometric Conference, Florence, ITALY, 6 – 11 July 2014