Table 4.
Depression Identification Algorithm Operating Characteristicsa
| Patient Subset/Characteristic | Algorithm 1 | Algorithm 2 |
|---|---|---|
| Algorithm “depressed,”n (%) | ||
| • True positive | 115 (25) | 63 (14) |
| • False positive | 119 (26) | 40 (9) |
| Algorithm “non-depressed,”n (%) | ||
| • True negative | 225 (48) | 304 (65) |
| • False negative | 6 (1) | 58 (12) |
| All patients, % | ||
| • Sensitivity | 95.0 | 52.1 |
| • Specificity | 65.4 | 88.4 |
| • Positive predictive valueb | 49.1 | 60.6 |
| • Negative predictive valueb | 97.4 | 84.0 |
a
Members were identified from 1997 Managed Care Organization administrative data by Algorithm 1; reported data (n=465) were obtained from chart review of stratified random subset; comparison standard: physician diagnosis of depression in medical record.
b
Based upon 26% prevalence of physician-recognized depression as determined by sampling plan; because the high prevalence of physician-recognized depression is related to the sampling strategy of this study, predictive values over a range of lower prevalence rates are presented in Figures 1A and 1B.