Abstract
BackgroundAlthough subgroup analyses in clinical trials may provide evidence for individualised medicine, their conduct and interpretation remain controversial.MethodsSubgroup effect can be defined as the difference in treatment effect across patient subgroups. Cumulative subgroup analysis refers to a series of repeated pooling of subgroup effects after adding data from each of related trials chronologically, to investigate the accumulating evidence for subgroup effects. We illustrated the clinical relevance of cumulative subgroup analysis in two case studies using data from published individual patient data (IPD) meta-analyses. Computer simulations were also conducted to examine the statistical properties of cumulative subgroup analysis.ResultsIn case study 1, an IPD meta-analysis of 10 randomised trials (RCTs) on beta blockers for heart failure reported significant interaction of treatment effects with baseline rhythm. Cumulative subgroup analysis could have detected the subgroup effect 15 years earlier, with five fewer trials and 71% less patients, than the IPD meta-analysis which first reported it. Case study 2 involved an IPD meta-analysis of 11 RCTs on treatments for pulmonary arterial hypertension that reported significant subgroup effect by aetiology. Cumulative subgroup analysis could have detected the subgroup effect 6 years earlier, with three fewer trials and 40% less patients than the IPD meta-analysis. Computer simulations have indicated that cumulative subgroup analysis increases the statistical power and is not associated with inflated false positives.ConclusionsTo reduce waste of research data, subgroup analyses in clinical trials should be more widely conducted and adequately reported so that cumulative subgroup analyses could be timely performed to inform clinical practice and further research.Electronic supplementary materialThe online version of this article (doi:10.1186/s12916-016-0744-x) contains supplementary material, which is available to authorized users.
Highlights
Subgroup analyses in clinical trials may provide evidence for individualised medicine, their conduct and interpretation remain controversial
Case study 1: beta blockers for heart failure Published in 2014, an individual patient data (IPD) meta-analysis of 10 Randomised controlled trial (RCT) found that the use of beta blockers reduced all-cause mortality for patients with heart failure and sinus rhythm, but not for those with heart failure plus atrial fibrillation at baseline [16]
Using reported hazard ratios for patients with sinus rhythm and those with atrial fibrillation, we estimated the ratio of hazard ratios (RHR) between the two subgroups in each of the 10 RCTs (Fig. 1)
Summary
Subgroup analyses in clinical trials may provide evidence for individualised medicine, their conduct and interpretation remain controversial. Randomised controlled trials (RCTs) provide the most valid evidence on effects of treatments and healthcare interventions, and results of RCTs are usually reported as estimated average effects. Subgroup analysis is often used in RCTs to investigate differences in treatment effect between patients with different characteristics [3]. A study found that subgroup analyses were reported in 44% of the 469 RCTs published in core clinical journals in 2007 [4]. It has been recommended that only a small number of pre-specified subgroup analyses in RCTs should be conducted using appropriate statistical methods [6, 7]
Sign-in/Register to view highlights & summary Sign-in/Register to access full text options 
Talk to us
Join us for a 30 min session where you can share your feedback and ask us any queries you have
Schedule a call
