Guidelines on how to assess the validity of results presented in subgroup analysis of clinical trials

Abstract

In observational studies, identification of associations within particular subgroups is the usual method of investigation. As an exploratory method, it is the bread and butter of epidemiological research. Nearly everything that has been learned in epidemiology has been derived from the analysis of subgroups. In a randomized clinical trial, the entire purpose is the comparison of the test subjects and the controls, and when there is particular interest in the results of treatment in a certain section of trial participants, a subgroup analysis is performed. These subgroups are examined to see if they are liable to a greater benefit or risk from treatment. Thus, analyzing patient subsets is a natural part of the process of improving therapeutic knowledge through clinical trials. Nevertheless, the reliability of subgroup analysis can often be poor because of problems of multiplicity and limitations in the numbers of patients studied. The naive interpretation of the results of such examinations is a cause of great confusion in the therapeutic literature. We emphasize the need for readers to be aware that inferences based on comparisons between subgroups in randomized clinical trials should be approached more cautiously than those based on the main comparison. That is, subgroup analysis results derived from a sound clinical trial are not necessarily valid; one must not jump to conclusions and accept the validity of subgroup analysis results without an appropriate judgment.

Invasive candidiasis in cancer patients: observations from a randomized clinical trial

“…The median (range) duration of study drug was 10.0 (2-28) days for caspofungin and 12.0 (2-28) days for amphotericin B. Six (18%) caspofungin recipients were switched to oral fluconazole for a median (range) of 8.5 (4)(5)(6)(7)(8)(9)(10)(11)(12)(13)(14)(15)(16)(17)(18)(19)(20) days; nine (22%) amphotericin B recipients were switched to oral fluconazole for a median (range) of 6.0 (2-15) days. Table 2).…”

“…Definitive conclusions about relative efficacy can not be made from a small post-hoc subgroup analysis. [13][14][15] Because of its broad-spectrum fungicidal activity against Candida species and lack of nephrotoxicity, caspofungin may provide another useful therapeutic option for selected cancer patients with invasive candidiasis. 7,[9][10][11][32][33][34][35] Further data pertinent to the treatment of cancer patients with invasive candidiasis with caspofungin come from a recently completed, double-blind, randomized clinical trial of empirical anti-fungal therapy for persistently febrile neutropenic patients.…”

“…Since this exploratory subgroup analysis was not specified in the original protocol, only point estimations of response rates are presented. [13][14][15] Formal comparisons between groups were not performed.…”

A systematic review of individual patient data meta-analyses on surgical interventions

“…Others showed that the proportion of studies that used interaction tests for at least one of their subgroups ranges from 10% to 56% [6,43-48], which is slightly lower compared to our findings. So far, few studies mentioned the importance of the power of subgroup analyses [6,44,50,51], and reported that many reports put too much emphasis on subgroup analyses that commonly lacked statistical power. This is in agreement with the results of the present study.…”

Extended Evaluation of Recombinant Human Activated Protein C United States Trial (ENHANCE US)

“…No randomization occurred, and no individual was blinded to the use of drotrecogin alfa (activated). The advantages of large, randomized clinical trials vs observational studies [17][18][19][20] and the limitations of subgroup analyses [21][22][23] have been welldocumented. Small sample sizes in this study and among analyzed patient subgroups resulted in large 95% CIs.…”