Subgroup analysis claims in haematologic phase III trials lack credibility, says study

21 Jul 2021
The role of scientists in medicine to be particularly important because of their ability to conduct large-scale research involving animal or human trials and draw reasonable conclusions.

The credibility of subgroup claims found in haematology randomized clinical trials (RCTs) seems to be lacking even when the claims are robust, suggests a study, noting that information about subgroup difference must be interpreted with caution.

The investigators performed a systematic review of Medline, including haematology phase III RCTs published between January 2013 and October 2019, to examine the appropriateness of the use and interpretation of subgroup analysis in these trials. They also obtained information related to trial characteristics, subgroup analysis, and claims of subgroup difference.

A total of 1,622 studies were identified in the initial search, of which 98 had reports on subgroup analyses. Twenty-four of such studies reported 46 claims of subgroup difference. Of these claims, 44 were for the primary outcomes: 25 were considered strong claims, while 17 were considered suggestions of a possible effect.

Of the 44 claims, 38 (86.36 percent) included subgroup variables for the primary outcome measured at baseline, 15 (34.09 percent) used a subgroup variable as a stratification factor at randomization, and 11 (25 percent) prespecified the hypothesis in a clear manner. The subgroup effect was one of a small number of hypothesized effects tested (≤5) for 17 claims (38.64 percent).

In addition, 18 claims (40.91 percent) had a test of interaction performed, 11 (25 percent) documented replication of a subgroup effect with previously related studies, 10 (22.72 percent) identified the consistency of a subgroup effect across outcomes, and eight (18.18 percent) provided a biological rationale for the effect.

Notably, majority of the claims for the primary outcome (n=34, 77.27 percent) met four or fewer of the 10 credibility criteria.

Editor's Recommendations