Choice of effect measure for meta-analyses of dichotomous outcomes influenced the identified heterogeneity and direction of small-study effects.

OBJECTIVES: The aim of this study was to compare the use of odds ratio (OR), risk ratio (RR), and risk difference (RD) in meta-analyses of dichotomous outcomes and assess their influence on their results. STUDY DESIGN AND

SETTING: Initially, we included meta-analyses from a meta-epidemiologic database and reanalyzed them with OR, RR, and RD as summary metric. The primary outcomes were the effects of metric choice on the (1) statistical significance, (2) heterogeneity, and (3) Egger's test for publication bias. Additionally, meta-analyses that originally used OR were reanalyzed using RR to assess the differences in their results.

RESULTS: In the 235 meta-analyses (147 reviews) that were included, the conclusions in terms of significance rarely changed. On the other hand, use of OR displayed the lowest I(2) values (median 42%), followed by RR (+5.1%) and RD (+15.0%). The Egger's test was most often significant with RR (32%), followed by RD (29%) and OR (24%). Substitution of RR for OR led to a change of the observed effects in 3%, change of between-study heterogeneity in 6% to 24%, and change in Egger's test results in 7% of the cases, respectively.

CONCLUSION: The choice of metric for meta-analyses of dichotomous outcomes might influence the identified between-study heterogeneity and the results of Egger's test.