Objectives: Heterogeneity in meta-analysis is usually assessed by a chi2 test, but it is also important to know the extent of any differences. Accordingly, I propose two measures of heterogeneity.

Methods: One measure quantifies the systematic inter-trial variation. It is the systematic variance of the (log) odds ratios, calculated from the chi2 statistic, converted to a standard deviation, expressed as a percentage and termed the geometric systematic variation (GSV). The other summarises the association between trial size and effect size. It is the ratio of the overall odds ratio for the large trials (top third ranked by number of patients) to that for the remainder (RLS: ratio of large to small). These measures were calculated for meta-analyses of eight interventions; three were tested in mega-trials.

Results: Meta-analyses of magnesium (myocardial infarction) and aspirin (pre-eclampsia) had high GSVs (about 50%), high RLSs (>2) and their results conflicted with those from subsequent mega-trials, in contrast, that for streptokinase had an RLS close to unity, a low GSV (20%) and its results were confirmed by mega-trials. Beta-blockers and a comparison of two forms of heparin were examples of meta-analyses showing little inter-trial variation, while heparin and antiplatelet therapy as thromboprophylaxes were examples showing large variation.

Discussion: High values for GSV and/or RLS suggest a lack of internal consistency. A high RLS may indicate publication bias while a large GSV may reflect differences in efficacy of the intervention among patient groups. The latter may be a property of the condition and/or the intervention under study. In conjunction with the number and size of trials and the clinical context, these measures can aid in interpreting results from meta-analyses, and may encourage a more widespread exploration of heterogeneity.