M-20: Quality and Report of Subgroup Analysis in Randomized Controlled Trials

Poster Presenter

Aurélien Grolleau

MSc Student
Université Claude Bernard - Institut De Pharmacie Industrielle De Lyon
France

Objectives

To describe the quality and the use of reporting subgroups in randomized controlled trials (RCT)

Method

Co-Authors: Grevat Emmanuelle - Gutiérrez Beatriz ; We performed a systematic review of 100 consecutive RCT reporting subgroups analysis in the following journals: NEJM, Lancet, JAMA, BMJ. The publications between January 2014 and December 2015 were reviewed by two independent investigators.

Results

44% of these RC were funded by industry. The main areas were cardiovascular (38%), oncologic (25%) and infectious (22%) diseases. Among the 100 RCT, 44% trials reported the results for 6-10 subgroups and 45% for >10 subgroups. Regarding the information reported in subgroups, 82% of the trials mentioned the number of patients and 56% the number of events by subgroups. 68% of the RCT were clear about prespecified or post-hoc subgroups analyses in the protocol. Subgroups were defined a priori in 40% of RCT. 22% of the trials reported information about statistical power calculation and only 16% reported information about the anticipation of alpha risk inflation. The interaction test was reported in 51 % of the trials encompassing 15% of RCT with a positive interaction. Among these 15 RCT finding interaction, 11 RCT were reported in abstract and 8 RCT claimed heterogeneity.

Conclusion

Despite the high impact factors of selected journals, the reporting of subgroup analysis is still neither uniform nor complete leading to ambiguities and imprecisions for physicians. Only one half of RCT is in accordance of guidelines for the design and analysis of subgroup analysis. These results are in accordance of a previous systematic review performed in 2007. Our results highlight the need to improve the design and report of subgroups in order to aid clinical decision and strengthen the design of future clinical trials.