Discriminative validity of the clinical rule predictions for the onset of hazardous alcohol drinking over 12 months in primary care
| Risk algorithms | Number of items | N | C-index (95% CI) | Hedges’ g (95% CI) | Predicted probabilitya | Frequencyb N (%) | Sensitivity Specificity | LR+ LR– | PPV NPV | Shrinkage factorc |
|---|---|---|---|---|---|---|---|---|---|---|
| predictAL-10d | 10 | 2264 | 0.886 (0.854 to 0.918) | 1.694 (1.460 to 1.928) | ≥2.72% | 495 (21.86) | 0.83 0.80 | 4.15 0.21 | 0.13 0.99 | 0.9595 |
| predictAL-17e | 17 | 2264 | 0.886 (0.853 to 0.919) | 1.729 (1.495 to 1.963) | ≥2.92% | 454 (20.05) | 0.80 0.82 | 4.44 0.25 | 0.13 0.99 | 0.9595 |
| predictAL-7f | 7 | 2264 | 0.819 (0.772 to 0.866) | 1.292 (1.052 to 1.532) | ≥3.31% | 539 (23.81) | 0.71 0.79 | 3.38 0.37 | 0.10 0.99 | 0.8615 |
| Clinical rule predictions excluding the variable sexual abuses in childhood | ||||||||||
| predictAL-9d | 9 | 2278 | 0.880 (0.847 to 0.913) | 1.658 (1.425 to 1.892) | ≥3.01% | 475 (20.85) | 0.79 0.81 | 4.16 0.26 | 0.13 0.99 | 0.9629 |
| predictAL-16e | 16 | 2278 | 0.883 (0.850 to 0.916) | 1.697 (1.463 to 1.931) | ≥2.78% | 483 (21.20) | 0.80 0.81 | 4.21 0.25 | 0.13 0.99 | 0.9639 |
| predictAL-6f | 6 | 2278 | 0.803 (0.752 to 0.854) | 1.273 (1.042 to 1.505) | ≥2.60% | 668 (29.32) | 0.70 0.72 | 2.5 0.42 | 0.08 0.98 | 0.8563 |
| Clinical rule predictions including only the AUDIT | ||||||||||
| AUDIT-C | 3 | 2288 | 0.775 (0.721 to 0.830) | 1.211 (0.980 to 1.442) | ≥2.34% | 517 (22.60) | 0.75 0.68 | 4.16 0.26 | 0.07 0.89 | 0.9819 |
| AUDIT | 10 | 2288 | 0.781 (0.725 to 0.836) | 1.254 (1.042 to 1.485) | ≥2.20% | 525 (22.95) | 0.75 0.68 | 4.21 0.25 | 0.07 0.89 | 0.9822 |
↵a Predicted probability of hazardous alcohol drinking at 12 months, cutoff point where Youden’s J statistic (J = Sensitivity + Specificity − 1) was greater: ‘optimal threshold’.
↵b Number of primary care attendees above the optimal threshold.
↵c Copas’ shrinkage factor estimates overfitting of the prediction models (shrinkage = 1 indicates that there is no overestimation).
↵d Including the AUDIT-C (three items).
↵e Including the AUDIT (10 items).
↵f Excluding any AUDIT. AUDIT = Alcohol Use Disorders Identification Test. LR+ = positive likelihood ratio. LR− = negative likelihood ratio. PPV = positive predictive value. NPV = negative predictive value.