Table 4

Random-effects three-level meta-analysis suggests significant irreproducibility

Meta-analysis data
Number of included effects of TMZ on U-87 MG viability644
Number of included articles the effects were reported in101
Effect of TMZ
Overall weighted mean effect of TMZ
(U-87 MG viability reduction compared to untreated control)
33.8% (30.0% to 37.7%)
Investigation of heterogeneity
Test whether heterogeneity is present
 Q (df=643)134 066.1
 P value<0.001
 Total I299.5%
Within-articles variance of the true effect
 τ23.6% (3.2% to 4.1%)
 τ19.0% (17.9% to 20.3%)
 I2 (=proportion of total variance)56.6%
Between-articles variance of the true effect (~irreproducibility)
 τ22.8% (1.9% to 3.9%)
 τ16.6% (13.9% to 19.8%)
 I2 (=proportion of total variance)42.9%
  • Random-effects three-level meta-analysis using the raw data the effects were calculated with as first level, the effect sizes within each article as second level and the articles the effects were reported in as third level. τ2: estimator of the variance of true effects (level-two variance: within-articles variance; level-three variance: between-articles variance: representant of irreproducibility); τ: square root of τ2; I2: proportion of within-articles and between-articles variance, respectively, of the total observed variance, including sampling error. τ2 estimator: restricted-maximum likelihood. Cochran’s Q was used as the test for heterogeneity using a χ2 distribution; values in parenthesis show CIs with significance level set at 0.05.

  • df, degree of freedom; U-87 MG, Uppsala-87 Malignant Glioma; TMZ, temozolomide.