Introduction: Subgroup analyses are frequently encountered in randomized trials (RCT) and those of meta-analyses of RCTs. Several authors have recommended the use of a test of homogeneity to estimate the probability that an observed interaction might have arisen due to chance. There has been inadequate study on whether the results of the tests of homogeneity depend on the choice of the measure of effect of an intervention.

Objective: To determine whether the results of the tests of homogeneity are different with the odds ratio and risk difference as measures of effect and to determine whether the observed difference, if any, influences the interpretation of results.

Methods: We extracted data from a report of a RCT of endoscopic surgery of primary supratentorial intracerebral haemorrhage for the reported subgroups based on age, size of haematoma, level of consciousness and site of haematoma. We conducted chi-square tests of homogeneity for the various subgroups using risk difference and odds ratio as measures of effect. P-values were interpreted as follows: >0.10 as "no interaction", between 0.5 and 0.10 as "probable interaction", and <0.5 as "definite interaction". The validity of the results was determined using opinion of experts blinded to the results of the tests.

Results: Tests based on RD showed no interaction; those based on odds ratio showed a definite interaction, indicating a benefit for patients with haematoma volume of >50 ml. Experts fully agreed on two interactions of which one was detected by odds ratio method and none by RD method. The interaction so detected reversed the reported interpretation of the findings and was closer to the experts' opinion.

Discussion: In applying tests of homogeneity, the choice of measure of effect should receive careful consideration. In some situations, tests of homogeneity based on odds ratio may be more valid than the ones based on risk difference.