Table 2. Factors to consider when interpreting credibility of claims of significance from subgroup analyses.

	Criteria to consider claims on subgroup analyses	Interpretation
1	Can chance explain the subgroup differences?	Instead of focusing on each subgroup in isolation, compare summary point estimate and confidence intervals across subgroups—if point estimate is similar, and confidence intervals are overlapping, and statistical test of interaction. For example, if subgroup 1 has effect estimate (RR) 0.78 with 95% CI 0.40–1.05, and subgroup 2 has effect estimate 0.75 with 95% CI 0.38–0.95, then the correct interpretation would be that there is NO difference in subgroups, rather than that effects are significant in subgroup 2 and not in subgroup 1; results may not be statistically significant in subgroup 1 due to small sample size or low event rate in subgroup 1, rather than true differences in efficacy of intervention in subgroups
2	Is the subgroup difference consistent across studies and suggested by comparisons within rather than between studies?	Findings from subgroup analyses are credible if observed in multiple individual studies, rather than just at summary level
3	Was the subgroup difference one of a small number of a priori hypotheses in which the direction was accurately prespecified?	Prespecified, hypotheses-driven subgroup analyses to explain heterogeneity across studies are more plausible, rather than post hoc assessment, which may be positive due to multiple statistical comparisons
4	Is there a strong preexisting biological rationale supporting the apparent subgroup effect?	Subgroup claims are more credible if supported by strong external, biological evidence from preclinical studies or studies of surrogate outcomes

CI, confidence interval; RR, relative risk.