Table 3 Characteristics of 68 meta-epidemiological (ME) studies quantitatively synthesized a difference of treatment effect estimatesa
Statistical analysis | No. of ME studyb | % (95% CI) |
|---|---|---|
Which model the ME study used to combine? | ||
 Two-step approach: within-meta-analysis comparison and combination | 43 | 63 (52 to 75) |
 Logistic regression | 7 | 10 (3 to 18) |
 Two-step approach: within-trial comparison and combination | 5 | 7 (1 to 14) |
 Bayesian multilevel model | 5 | 7 (1 to 14) |
 Meta-regression | 3 | 4 (0 to 9) |
 More than one modelc | 5 | 7 (1 to 14) |
Accounted for clustering of trials within meta-analysisd | 53/61 | 87 (78 to 96) |
Used random effect models to account for variability across meta-analyses/trials | ||
 Yese | 43/61 | 70 (59 to 82) |
 Noe | 18/61 | 30 (18 to 41) |
Adjusted meta-confounders | 54 | 79 (70 to 89) |
 Based on subgroup analysis solelye | 32/54 | 59 (46 to 73) |
 Based on multiple variable analysis solelye | 6/54 | 11 (2 to 20) |
 Bothe | 16/54 | 30 (17 to 42) |
 If yes, as main analysise | 21/54 | 39 (26 to 52) |
Assessed the heterogeneity during analysis | 59 | 87 (78 to 95) |
 Based on qualitative domain solely (chi-square test)e | 4/59 | 7 (0 to 13) |
 Based on quantitative domain solely (I2, τ2, φ2, F test)e | 20/59 | 34 (22 to 46) |
 Bothe | 35/59 | 59 (46 to 72) |
Whether the author clearly reported the direction of interpretation of results | ||
 Yes | 60 | 88 (80 to 96) |
 No | 8 | 12 (4 to 20) |
Subgroup analyses performed | 48 | 71 (60 to 82) |
 Based on trial-level characteristicse | 30/48 | 62 (48 to 77) |
 Based on meta-analysis-level characteristicse | 1/48 | 2 (0 to 6) |
 Bothe | 17/48 | 35 (21 to 50) |