Jingbo Zhai1, Hongbo Cao1, Ming Ren2, Wei Mu3, Sisi Lv4, Jinhua Si5, Hui Wang1, Jing Chen6, Hongcai Shang7. 1. Research Institute of Traditional Chinese Medicine, Tianjin University of Traditional Chinese Medicine, 88 Yuquan Street, Nankai District, Tianjin 300193, China. 2. Baokang Hospital, Tianjin University of Traditional Chinese Medicine, 88 Yuquan Street, Nankai District, Tianjin 300193, China. 3. Second Affiliated Hospital, Tianjin University of Traditional Chinese Medicine, 816 Zhenli Street, Hebei District, Tianjin 300150, China. 4. Modern Educational Technology and Information Center, Tianjin University of Traditional Chinese Medicine, Tianjin, China. 5. Library of Tianjin University of Traditional Chinese Medicine, 88 Yuquan Street, Nankai District, Tianjin 300193, China. 6. Baokang Hospital, Tianjin University of Traditional Chinese Medicine, 88 Yuquan Street, Nankai District, Tianjin 300193, China. Electronic address: cjshcsyc@126.com. 7. Key Laboratory of Chinese Internal Medicine of Ministry of Education and Beijing, Dongzhimen Hospital, Beijing University of Chinese Medicine, Beijing 100700, China. Electronic address: shanghongcai@foxmail.com.
Abstract
OBJECTIVES: N-of-1 trials can be aggregated to estimate population treatment effects using hierarchical Bayesian models. It is very important to report core items in hierarchical Bayesian analysis. In this study, we assessed reporting of items in hierarchical Bayesian analysis for aggregating N-of-1 trials to estimate population treatment effects. STUDY DESIGN AND SETTING: This was a systematic literature review of aggregating N-of-1 trials by hierarchical Bayesian models to estimate population treatment effects. A comprehensive search was performed to collect eligible articles. Pilot studies, formal N-of-1 trials and reports in which the data were reanalyzed using hierarchical Bayesian methods, were included. The information of reported items related with hierarchical Bayesian analysis was extracted by two independent reviewers. The guideline "ROBUST," developed for reporting Bayesian analysis of clinical studies, was published in Journal of Clinical Epidemiology in 2005. We assessed the included reports using ROBUST criteria and 18 other important items. RESULTS: After careful screening, 11 studies were identified to be eligible for inclusion. There were three pilot studies, four formal trials, and four reports in which the data were reanalyzed using hierarchical Bayesian methods. The number of reported items in ROBUST criteria ranged from six to seven, with a median number of six. Five of eleven included articles reported all items of the ROBUST criteria. But for justification and sensitivity analysis in prior distribution items, other items were reported in all of the included articles. Software and analysis data set items were reported the most frequently in additional items excluded from the ROBUST criteria. Less than half of the studies reported the other additional items. CONCLUSION: Reporting of core items in hierarchical Bayesian analysis for aggregating N-of-1 trials to estimate population treatment effects is suboptimal. A PRISMA-like guidance on reviews of Bayesian N-of-1 trials may be required in the future.
OBJECTIVES: N-of-1 trials can be aggregated to estimate population treatment effects using hierarchical Bayesian models. It is very important to report core items in hierarchical Bayesian analysis. In this study, we assessed reporting of items in hierarchical Bayesian analysis for aggregating N-of-1 trials to estimate population treatment effects. STUDY DESIGN AND SETTING: This was a systematic literature review of aggregating N-of-1 trials by hierarchical Bayesian models to estimate population treatment effects. A comprehensive search was performed to collect eligible articles. Pilot studies, formal N-of-1 trials and reports in which the data were reanalyzed using hierarchical Bayesian methods, were included. The information of reported items related with hierarchical Bayesian analysis was extracted by two independent reviewers. The guideline "ROBUST," developed for reporting Bayesian analysis of clinical studies, was published in Journal of Clinical Epidemiology in 2005. We assessed the included reports using ROBUST criteria and 18 other important items. RESULTS: After careful screening, 11 studies were identified to be eligible for inclusion. There were three pilot studies, four formal trials, and four reports in which the data were reanalyzed using hierarchical Bayesian methods. The number of reported items in ROBUST criteria ranged from six to seven, with a median number of six. Five of eleven included articles reported all items of the ROBUST criteria. But for justification and sensitivity analysis in prior distribution items, other items were reported in all of the included articles. Software and analysis data set items were reported the most frequently in additional items excluded from the ROBUST criteria. Less than half of the studies reported the other additional items. CONCLUSION: Reporting of core items in hierarchical Bayesian analysis for aggregating N-of-1 trials to estimate population treatment effects is suboptimal. A PRISMA-like guidance on reviews of Bayesian N-of-1 trials may be required in the future.
Authors: Harry P Selker; Theodora Cohen; Ralph B D'Agostino; Willard H Dere; S Nassir Ghaemi; Peter K Honig; Kenneth I Kaitin; Heather C Kaplan; Richard L Kravitz; Kay Larholt; Newell E McElwee; Kenneth A Oye; Marisha E Palm; Eleanor Perfetto; Chandra Ramanathan; Christopher H Schmid; Vicki Seyfert-Margolis; Mark Trusheim; Hans-Georg Eichler Journal: Clin Pharmacol Ther Date: 2021-10-20 Impact factor: 6.903