Table 3 Meta-analyses of the proportions of associations for each criterion and level of evidence (random effects)
| n/N associations (crude proportion) |
Proportions [95% CI] from meta-analysis of 57 topicsa | I2 | Range of proportions across topics | |
|---|---|---|---|---|
| Level of evidence: | ||||
| Convincing | 99/3744 (2.6%) | 1.3% [1.0%; 2.2%] | 73.9% | 0–16.7% |
| Highly suggestive | 253/3744 (6.7%) | 4.6% [2.9%; 6.6%] | 85.7% | 0–33.3% |
| Suggestive | 440/3744 (11.8%) | 11.0% [8.5%; 13.8%] | 83.9% | 0–50% |
| Weak | 1497/3744 (40.0%) | 39.1% [34.8%; 43.5%] | 86.2% | 0–71.4% |
| Non-significant | 1455/3744 (38.9%) | 34.7% [29.2%; 40.3%] | 90.8% | 0–80.7% |
| Criteria: | ||||
| Statistical significance | ||||
| P < 10-6 | 762/2289 (33.3%) | 29.0% [24.9%; 33.3%] | 74.8% | 0–66.7% |
| P < 0.001 | 1377/2289 (60.2%) | 58.6% [54.1%; 63.0%] | 73.3% | 0–100% |
| Cases > 1000 (or > 20,000 participants for continuous outcomes) | 1182/2107 (56.1%) | 65.3% [56.9%; 73.2%] | 94.9% | 1.7–100% |
| Largest study with P < 0.05 | 1343/1781 (75.4%) | 74.9% [71.2%; 78.4%] | 63.6% | 28.6–100% |
| 95% prediction interval that excluded the null | 642/2136 (30.1%) | 30.3% [26.5%; 34.2%] | 71.0% | 9.0–100% |
| Absence of large heterogeneity (I2<50%) | 1050/2277 (46.1%) | 46.6% [41.8%; 51.3%] | 79.5% | 0–88.2% |
| No evidence of small study effects (P > 0.10) | 1628/2164 (75.2%) | 75.3% [72.2%; 78.3%] | 63.2% | 40–100% |
| No evidence of excess significance (P > 0.10) | 1599/2052 (77.9%) | 77.7% [72.6%; 82.5%] | 83.0% | 33.3–100% |