Article type
Year
Abstract
Background: Large variation in results of individual studies (heterogeneity) decreases certainty in the effect estimates from meta-analyses. Authors have addressed the interpretation of heterogeneity, as assessed by I2, primarily in meta-analysis evaluating binary outcomes.
Objectives: We compared the distribution of heterogeneity in meta-analyses of binary and continuous outcomes, and explored hypotheses explaining the difference in distributions.
Methods: We searched citations in MEDLINE and Cochrane databases for meta-analyses of randomized trials published in 2012 that reported a measure of heterogeneity in the analysis of either binary or continuous outcomes. Two reviewers independently performed eligibility screening and data abstraction. We evaluated the distribution of I2 in meta-analyses of binary and continuous outcomes and explored the association of number of studies included and distribution of heterogeneity separately for continuous and binary outcomes. We tested the hypothesis that I2 increases with an increasing number of studies meta-analyzed and increasing precision of study effect estimate using bivariate Spearman rank correlation.
Results: After full-text screening, we selected 671 meta-analyses evaluating 557 binary and 352 continuous outcomes. Heterogeneity, as assessed by I2, proved higher in continuous than in binary outcomes: the proportion of continuous and binary outcomes reporting an I2 of 0% was 34% versus 52% respectively and reporting an I2 of 60% to 100% was 39% versus 14%. In continuous - but not binary outcomes - I2 increased with larger number of studies included in a meta-analysis. Increased precision and sample size do not explain the larger I2 found in meta-analyses of continuous outcomes with a larger number of studies.
Conclusions: Meta-analyses evaluating continuous outcomes showed substantially higher I2 than meta-analyses of binary outcomes. Results suggest differing standards for interpreting I2 in continuous versus binary outcomes may be appropriate.
Objectives: We compared the distribution of heterogeneity in meta-analyses of binary and continuous outcomes, and explored hypotheses explaining the difference in distributions.
Methods: We searched citations in MEDLINE and Cochrane databases for meta-analyses of randomized trials published in 2012 that reported a measure of heterogeneity in the analysis of either binary or continuous outcomes. Two reviewers independently performed eligibility screening and data abstraction. We evaluated the distribution of I2 in meta-analyses of binary and continuous outcomes and explored the association of number of studies included and distribution of heterogeneity separately for continuous and binary outcomes. We tested the hypothesis that I2 increases with an increasing number of studies meta-analyzed and increasing precision of study effect estimate using bivariate Spearman rank correlation.
Results: After full-text screening, we selected 671 meta-analyses evaluating 557 binary and 352 continuous outcomes. Heterogeneity, as assessed by I2, proved higher in continuous than in binary outcomes: the proportion of continuous and binary outcomes reporting an I2 of 0% was 34% versus 52% respectively and reporting an I2 of 60% to 100% was 39% versus 14%. In continuous - but not binary outcomes - I2 increased with larger number of studies included in a meta-analysis. Increased precision and sample size do not explain the larger I2 found in meta-analyses of continuous outcomes with a larger number of studies.
Conclusions: Meta-analyses evaluating continuous outcomes showed substantially higher I2 than meta-analyses of binary outcomes. Results suggest differing standards for interpreting I2 in continuous versus binary outcomes may be appropriate.