人類学
オープンアクセス
ISSN: 2167-0870
ISSN: 2167-0870
ルイ・ジャコブ、マリオン・カセレス、モルガン・ジル、レア・ポールマルク、シルヴィー・シェブレ
Objectives: The analysis of Adverse Events (AE) is an important aspect of the assessment of new treatments. Data on AE are often reported through individual frequency rates, ignoring potential sources of heterogeneity due to either treatment course or individuals. We aimed to illustrate how Bayesian modelling may achieve reliable information using data of a randomized clinical trial evaluating chemotherapies against acute promyelocytic leukaemia (APL2006 trial). Methods: We first performed in 2015 a medical literature search to illustrate the need for improvement in AE reporting. We then used the APL2006 trial data to apply Bayesian hierarchical models on AE counts. Results: Only five over the 10 intended journals were found to have published results from RCTs in the study period. Median trial sample size was 523, ranging from 50 up to 20,870 with efficacy results mostly positive (in 61%). Although 39 (89%) articles briefly report AE information in the abstract, the analysis of AE data was poorly reported or even performed. In the APL2006 trial, 522 (97%) of the 538 patients received a total of 4,203 chemotherapy courses. A total of 3,584 AEs were recorded on 2,242 (53.3%) courses in 520 (99.6%) patients, that is, in all but 2 patients from arm A. Therefore, the rate of patients experiencing AE was poorly informative while the mean AE counts per patient were preferred. Besides the randomization arm, the various exposures– as summarized by the number of administered courses and the type of chemotherapy course, appeared as potential sources of variability. Bayes analysis of these AE counts, using Poisson-Gamma models with non-informative priors allowed to depict the heterogeneity in AE count across arms. Conclusion: We showed the interests of Bayes modeling to provide information on the adverse events distribution in a randomized clinical trial. Trial registration number and trial register: APL2006, NCT00378365.