Table 3.
Multivariable logistic regression models to predict mortality 6 months after traumatic brain injury fitted separately in each of the 15 studies. We show the estimated regression coefficients with associated standard errors for the 15 studies. A two‐stage multivariate meta‐analysis provided pooled estimates of the between‐study variance parameter tau and prediction intervals for the regression coefficients. The between versus within‐study heterogeneity is summarized in I 2 estimates
| Study | Intercept | Age | Motor score | Pupillary reactivity | Hypoxia | Hypotension | CT class | tSAH |
|---|---|---|---|---|---|---|---|---|
| 1 | −1.22 (0.09) | 0.20 (0.05) | −0.39 (0.08) | 0.41 (0.11) | 0.36 (0.20) | 1.03 (0.21) | 0.56 (0.10) | 1.01 (0.17) |
| 2 | −1.40 (0.10) | 0.21 (0.07) | −0.40 (0.08) | 0.36 (0.11) | 0.46 (0.18) | 0.75 (0.19) | 0.34 (0.10) | 0.74 (0.17) |
| 3 | −1.35 (0.22) | 0.28 (0.09) | −0.28 (0.12) | 0.71 (0.23) | −0.36 (0.58) | 0.97 (0.35) | 0.47 (0.15) | 0.70 (0.37) |
| 4 | −1.34 (0.09) | 0.20 (0.06) | −0.14 (0.07) | 0.74 (0.11) | 0.68 (0.24) | 0.22 (0.23) | 0.33 (0.10) | 0.82 (0.18) |
| 5 | −1.73 (0.10) | 0.21 (0.05) | −0.52 (0.06) | 0.52 (0.08) | 0.33 (0.16) | 0.77 (0.17) | 0.38 (0.08) | 0.54 (0.14) |
| 6 | −1.41 (0.19) | 0.30 (0.09) | −0.45 (0.13) | 0.82 (0.17) | 0.00 (0.38) | −0.60 (0.63) | 0.38 (0.08) | 0.95 (0.29) |
| 7 | −0.93 (0.11) | 0.43 (0.05) | −0.30 (0.09) | 1.01 (0.12) | 0.07 (0.21) | 1.21 (0.22) | 0.36 (0.11) | 0.70 (0.19) |
| 8 | −0.73 (0.12) | 0.47 (0.07) | −0.42 (0.10) | 0.57 (0.12) | 0.36 (0.27) | 1.31 (0.25) | 0.43 (0.12) | 0.63 (0.21) |
| 9 | −1.28 (0.35) | 0.38 (0.16) | −0.23 (0.22) | 0.34 (0.26) | −0.40 (0.54) | 0.71 (0.59) | 0.64 (0.28) | 0.72 (0.63) |
| 10 | −1.41 (0.12) | 0.40 (0.05) | −0.45 (0.09) | 0.80 (0.12) | 0.54 (0.23) | 0.73 (0.24) | 0.31 (0.11) | 0.81 (0.19) |
| 11 | −1.44 (0.11) | 0.22 (0.06) | −0.40 (0.09) | 0.43 (0.11) | 0.21 (0.23) | 0.34 (0.30) | 0.38 (0.11) | 0.97 (0.19) |
| 12 | −1.49 (0.17) | 0.24 (0.10) | −0.39 (0.11) | 0.68 (0.14) | 0.35 (0.28) | 0.83 (0.34) | 0.36 (0.21) | 0.60 (0.26) |
| 13 | −1.43 (0.14) | 0.22 (0.09) | −0.42 (0.11) | 0.68 (0.14) | −0.04 (0.34) | 0.26 (0.30) | 0.52 (0.14) | 0.76 (0.24) |
| 14 | −1.61 (0.11) | 0.17 (0.07) | −0.34 (0.09) | 0.29 (0.16) | 0.06 (0.23) | 0.46 (0.26) | 0.53 (0.11) | 0.42 (0.21) |
| 15 | −2.07 (0.18) | 0.52 (0.07) | −0.59 (0.15) | 0.91 (0.16) | 0.33 (0.28) | 0.54 (0.37) | 0.29 (0.14) | 0.47 (0.26) |
| Pooled | −1.35 (0.07) | 0.28 (0.03) | −0.38 (0.03) | 0.61 (0.06) | 0.27 (0.07) | 0.71 (0.10) | 0.40 (0.03) | 0.72 (0.06) |
| Estimated τ | 0.25 | 0.09 | 0.07 | 0.17 | 0.08 | 0.27 | 0.06 | 0.08 |
| 95% Prediction interval | [−1.92, −0.78] | [0.08, 0.48] | [−0.55, −0.20] | [0.21, 1.01] | [0.05, 0.50] | [0.08, 1.34] | [0.25, 0.55] | [0.50, 0.93] |
| I 2 | 84% | 67% | 35% | 65% | 0% | 49% | 0% | 2% |
Age was analyzed as a continuous predictor, per 10 years; Motor score, pupillary reactivity, and CT class were analyzed as continuous predictors, coded as in Figure 1. Hypoxia, hypotension, and tSAH were binary predictors. For interpretation of the baseline risk (the intercept α), we standardized predictors by subtracting the overall means of predictor values.