Table 2. Prevalence and potential impact of publication bias.
| Intervention | Reported Effect Size (95%Cl) | Bias with Egger Regression | Bias with METATRIM | Additional %Studies Considered “Missing” | METATRIM Adjusted Effect Size (95%Cl) | Absolute Overstatement of Efficacy | Relative Overstatement of Efficacy |
| Estrogens | 26.7% (20.4%–33.0%) | + | + | 24 | 11.9% (4.6%–19.2%)a | 14.8% (8.0%–21.6%) | 124.4% |
| FK506 | 32.0% (27.8%–36.3%) | + | + | 30 | 21.9% (17.5%–26.3%)a | 10.1% (5.8%–14.4%) | 46.1% |
| Growth factors | 29.7% (25.9%–33.4%) | + | + | 14 | 25.1% (21.2%–28.9%)a | 4.6% (0.9%–8.3%) | 18.3% |
| Hypothermia | 43.5% (40.1%–47.0%) | + | + | 20 | 35.4% (31.7%–39.1%)a | 8.1% (4.5%–11.6%) | 22.9% |
| IL1-RA | 38.2% (31.2%–45.1%) | + | + | 36 | 25.4% (18.4%–32.4%)a | 12.8% (5.9%–19.7%) | 50.4% |
| Melatonin | 42.1% (35.7%–48.5%) | + | + | 14 | 41.0% (34.8%–47.3%) | 1.1% (−5.1% to 7.4%) | 2.7% |
| Minocycline | 30.9% (24.1%–37.6%) | + | − | 0 | No adjustment | — | |
| Nicotinamide | 29.2% (23.0%–35.5%) | + | + | 24 | 21.8% (14.9%–28.6%)a | 7.4% (0.8%–13.9%) | 33.9% |
| NOS donors | 21.4% (13.7%–29.1%) | + | + | 25 | 14.0% (6.4%–21.6%)a | 7.4% (−0.1% to 14.9%) | 52.9% |
| NOS inhibitors | 22.2% (17.1%–27.3%) | + | + | 13 | 14.7% (8.9%–20.6%)a | 7.5% (2.0%–13.0%) | 51.0% |
| NXY-059 | 43.8% (34.7%–52.8%) | + | − | 0 | No adjustment | — | |
| Piracetam and related compounds | 29.6% (16.1%–44.4%) | + | − | 0 | No adjustment | ||
| Stem cells | 29.6% (23.7%–35.4%) | + | − | 0 | No adjustment | — | |
| Tirilazad | 31.9% (23.1%–40.7%) | + | − | 0 | No adjustment | — | |
| tPA | 22.5% (19.2%–25.9%) | + | + | 5 | 19.9% (16.4%–23.3%) | 2.6% (−0.7% to 6.0%) | 13.1% |
| Other Thrombolytics | 46.6% (35.7%–57.5%) | + | − | 0 | No adjustment | - | |
| Pooled analysis | 31.3% (29.7%–32.8%) | + | + | 214 b | 23.8% (22.2%–25.5) a | 7.5% (5.9%–9.1%) | 31.1% |
Duval and Tweedie nonparametric trim-and-fill provides an estimate of the number of unpublished studies, and provides an estimate of what the observed efficacy might have been had these studies been available. Where no adjustment is made there are either not enough data to infer the number of missing studies or there is no publication bias.
a
p<0.05 versus unadjusted estimate of efficacy.
b
This (214) is the estimate of missing studies in the pooled analysis of the total dataset rather than the sum of missing studies from the individual drug datasets (205), and suggests that a further nine studies are missing, probably from those reviews where no adjustment was made because the analysis of publication bias was underpowered for smaller reviews.