Article type
Year
Abstract
Background: Credibility of subgroup analyses in randomized trials (RCTs) depends on several factors, in particular subgroup prespecification. Little is known about subgroup planning in trial protocols and their reporting in subsequent publications.
Objectives: To describe (i) the planning of subgroup analyses in RCTs, and (ii) to determine the agreement between planning of subgroups and their reporting in subsequent publications.
Methods: We established a multicentre cohort of RCTs based on protocols approved by six research ethics committees from 2000 to 2003 in Switzerland, Germany, and Canada. From included protocols we extracted data on study design and planning of subgroup analyses. We identified subsequent full publications by literature searches and surveys of trialists. We calculated the proportion of protocol/publications that planned/reported subgroup analyses, provided hypotheses, anticipated direction of effect, and planning/application of an interaction term. We investigated the agreement between subgroups stated in protocols and reported in publications using 2 × 2 tables. We used logistic regression to investigate whether characteristics of subgroup reporting and trial initiation (industry vs. investigator) were associated with subgroup prespecification.
Results: We included 894 protocols and 520 journal publications. The table summarizes the proportions of planned/reported subgroup analyses and the agreement between planning and reporting. In 96 publications authors stated that at least 1 of their reported subgroup analyses was prespecified, but only 51 (53%) corresponding protocols reported subgroup planning. In 13 instances the number of subgroup analyses in protocol and corresponding publication(s) was identical. Reported subgroup hypothesis, direction of effect, and interaction term were not associated with subgroup prespecification in the protocol. Industry-initiated trials were more likely to prespecify subgroup analyses.
Conclusions: Trial protocols commonly describe planned subgroup analyses, but hypotheses, anticipated direction of effects, and planning interaction tests are rarely specified. Systematic review authors cannot rely on statements of subgroup prespecification in RCT reports.
Objectives: To describe (i) the planning of subgroup analyses in RCTs, and (ii) to determine the agreement between planning of subgroups and their reporting in subsequent publications.
Methods: We established a multicentre cohort of RCTs based on protocols approved by six research ethics committees from 2000 to 2003 in Switzerland, Germany, and Canada. From included protocols we extracted data on study design and planning of subgroup analyses. We identified subsequent full publications by literature searches and surveys of trialists. We calculated the proportion of protocol/publications that planned/reported subgroup analyses, provided hypotheses, anticipated direction of effect, and planning/application of an interaction term. We investigated the agreement between subgroups stated in protocols and reported in publications using 2 × 2 tables. We used logistic regression to investigate whether characteristics of subgroup reporting and trial initiation (industry vs. investigator) were associated with subgroup prespecification.
Results: We included 894 protocols and 520 journal publications. The table summarizes the proportions of planned/reported subgroup analyses and the agreement between planning and reporting. In 96 publications authors stated that at least 1 of their reported subgroup analyses was prespecified, but only 51 (53%) corresponding protocols reported subgroup planning. In 13 instances the number of subgroup analyses in protocol and corresponding publication(s) was identical. Reported subgroup hypothesis, direction of effect, and interaction term were not associated with subgroup prespecification in the protocol. Industry-initiated trials were more likely to prespecify subgroup analyses.
Conclusions: Trial protocols commonly describe planned subgroup analyses, but hypotheses, anticipated direction of effects, and planning interaction tests are rarely specified. Systematic review authors cannot rely on statements of subgroup prespecification in RCT reports.