| Subgroup analysis in clinical trials can be used to identify important treatment differences by subgroup. NIH guidelines encourage subgroup analyses by age, sex or race, depending upon the evidence for differences. The aims of this study were to: (1) identify what types of subgroups are analyzed and differences reported; and (2) compare subgroup analyses by manual review of clinical trial publications with electronic word identification.;We conducted a MEDLINE search of clinical trials published from 1990 to 1999. From 37,999 abstracts, we reviewed a random sample of 3,800 abstracts for eligibility, and included 145 trials in our study. We manually reviewed each trial for report of subgroup analysis and subgroup difference, and characterized the subgroups analyzed.;We also obtained electronic files (PDF or HTML) of 140 of 145 articles. We compared strategies of identifying subgroup analysis found through electronic word identification of the abstract or article with manual review. We calculated the sensitivity and specificity of the first three strategies compared to all other strategies, for various combinations of terms.;In 35% (47/145) of trials at least one subgroup analysis was reported. We found subgroup differences in 45% (21/47) of trials with at least one subgroup analysis. In the 21 trials, there were 38 reported subgroup differences: 2 by center, 4 by gender, 2 by age, 28 by disease-related or biological factors, and 2 by composite subgroups (e.g., gender by genotype); none by race/ethnicity.;We found that electronic word identification can capture 75% of the articles identified by manual review for subgroup analysis, but the false positive rate was 68%, and the specificity was only 20%. Refining the search algorithm improved the specificity to 80%, with a decrease in the sensitivity to 67%.;In a broad sample of trial publications, subgroup analyses and differences were not commonly reported; few were based on demographic factors. Identifying subgroup analyses in clinical trials is a time-consuming process due to inconsistent reporting and terminology use. We recommend that clinical trial reports include consistent terminology, clear accounting of subgroup analyses performed, and methodology, such as whether the subgroup hypotheses were a priori or ad hoc. |
