Table 3.
Algorithm approaches to diagnosis of alcohol problems in primary care in hypothetical 1000 attendees
| Sensitivity (95% CI) | Specificity (95% CI) | PPV (95% CI) | NPV (95% CI) | Overall accuracy (95% CI) | Positive clinical utility index (95% CI) | Negative clinical utility index (95% CI) | Questions asked per 100 attendees | |
|---|---|---|---|---|---|---|---|---|
| Combined algorithms | ||||||||
| SQ then AUDIT | 0.337 (0.272 to 0.403) | 0.989 (0.981 to 0.996) | 0.886 (0.810 to 0.955) | 0.856 (0.834 to 879) | 0.858 (0.837 to 0.880) | 0.298 (0.292 to 0.303) ‘v poor’ |
0.847 (0.847 to 0.847) ‘excellent’ |
311 |
| 2QQ then AUDIT | 0.540 (0.471to 0.609) | 0.983 (0.973 to 0.992) | 0.885 (0.829 to 0.942) | 0.895 (0.875 to 0.915) | 0.894 (0.875 to 0.913) | 0.478 (0.474 to 0.482) ‘poor’ |
0.880 (0.879 to 0.880) ‘excellent’ |
536 |
| SQ then CAGE | 0.387 (0.319 to 0.455) | 0.988 (0.980 to 0.995) | 0.886 (0.819 to 0.952) | 0.866 (0.844 to 0.888) | 0.867 (0.846 to 0.888) | 0.343 (0.337 to 0.348) ‘v poor’ |
0.855 (0.885 to 0.885) ‘excellent’ |
184 |
| 2QQ then CAGE | 0.620 (0.553 to 0.687) | 0.981 (0.972 to 0.991) | 0.892 (0.841 to 0.944) | 0.912 (0.893 to 0.931) | 0.909 (0.891 to 0.926) | 0.553 (0.549 to 0557) ‘fair’ |
0.895 (0.894 to 0.895) ‘excellent’ |
334 |
| Single applications | ||||||||
| CAGE Single application | 0.615 (0.548 to 0.682) | 0.915 (0.896 to 0.934) | 0.644 (0.576 to 0.712) | 0.905 (0.885 to 0.925) | 0.855 (0.833 to 0.876) | 0.396 (0.392 to 0.400) ‘poor’ |
0.828 (0.828 to 0.828) ‘excellent’ |
400 |
| AUDIT Single application | 0.71 (0.647 to 0.773) | 0.91 (0.890 to 0.930) | 0.664 (0.600 to0.727) | 0.926 (0.908 to 0.944) | 0.870 (to 0.849 to 0.890) | 0.471 (0.468 to 0.475) ‘poor’ |
0.843 (0.842 to 0.843) ‘excellent’ |
1000 |
| SQ Single application | 0.545 (0.476 to 0.614) | 0.873 (0.849 to 0.896) | 0.517 (0.449 to 0.584) | 0.885 (0.862 to 0.907) | 0.807 (0.782 to 0.831) | 0.282 (0.278 to 0.285) ‘v poor’ |
0.772 (0.771 to 0.772) ‘excellent’ |
100 |
| 2QQ Single application | 0.87 (0.823 to 0.917) | 0.798 (0.770 to 0.825) | 0.518 (0.464 to 0.571) | 0.961 (0.946 to 0.976) | 0.812 (0.787 to 0.836) | 0.451 (0.448 to 0.453) ‘poor’ |
0.766 (0.766 to 0.767) ‘excellent’ |
200 |
Table shows hierarchical screening algorithms applied to 1000 hypothetical primary care attendees, of whom 200 have alcohol problems. Calculations from www.clinicalutlity.co.uk. 2QQ = two questions. NPV = negative predictive value. PPV = positive predictive value. SQ = single question.