Table 1.

Recognized pitfalls and solutions for subgroup analysis as described in the literature

Statistical issue	Potential solutions
Increased type I error rate (false-positive findings)	Limit the number of analyses performed to those that are pre-specified and biologically plausible Adjust for multiplicity
Increased type II error rate (low power to detect interesting findings)	Design study to examine particularly relevant subgroups
Biased estimates of treatment	Limit to performing proper analyses
Findings are difficult to interpret	Emphasize overall findings Utilize formal tests of interaction to appropriately assess heterogeneity of effects Do not perform subgroup analyses if overall findings are negative Report the number of tests performed Report the number of tests that were specified a priori and the number that were specified a posteriori Report all findings—positive and negative