Updating insights into rosiglitazone and cardiovascular risk through shared data: individual patient and summary level meta-analyses

This paper by Wallach and colleagues (BMJ 2020;368:l7078, doi:10.1136/bmj.l7078, published 5 February 2020) has been corrected to acknowledge discrepancies in the number of fatal adverse events between the event numbers reported in this study and in other reports for four of the 34 trials for which individual patient level data were used (ADOPT, AVA102670, AVA102672, and AVD100521). For three of the 34 trials (AVA102670, AVA102672, and AVD100521), the number of fatal events differed by 28, 24, and 11, respectively. For the 34th trial, ADOPT, the clinical study report redacted all information on causes of death, but the overall number of deaths differed greatly from the total number of deaths that we identified using the originally shared individual patient level data (96 v 5). Because of the large number of meta-analyses conducted and presented in this paper—each including varying numbers of trials and data types (individual patient level data v summary data), examining multiple different endpoints, and using an array of statistical approaches—the investigators and editors determined that only the primary analysis would be repeated using the alternate number of events, finding that the conclusions regarding cardiovascular specific and mortality risk associated with rosiglitazone were not meaningfully different (web table).

Web table