Table 3.
Performance of DAP model and baseline models within 365 days after discharge on 10-fold external validation on PHS for cohorts from NHIP, GHS, and TCM-HS.
| Models | ROC-AUC | P-value | PR-AUC | P-value |
|---|---|---|---|---|
| NHIPa | ||||
| LR | 0.51 (±0.076) | .000 | 0.30 (±0.074) | .000 |
| RF | 0.52 (±0.061) | .000 | 0.28 (±0.058) | .000 |
| XGB | 0.53 (±0.065) | .000 | 0.26 (±0.039) | .000 |
| DAP | 0.75 (±0.045) | ref | 0.47 (±0.081) | ref |
| GHSL | ||||
| LR | 0.50 (±0.055) | .000 | 0.28 (±0.060) | .009 |
| RF | 0.50 (±0.061) | .000 | 0.27 (±0.037) | .000 |
| XGB | 0.61 (±0.051) | .476 | 0.37 (±0.053) | .804 |
| DAP | 0.62 (±0.059) | ref | 0.36 (±0.055) | ref |
| GHSM | ||||
| LR | 0.40 (±0.069) | .005 | 0.23 (±0.042) | .281 |
| RF | 0.50 (±0.059) | .737 | 0.27 (±0.050) | .453 |
| XGB | 0.49 (±0.063) | .848 | 0.27 (±0.037) | .473 |
| DAP | 0.49 (±0.060) | ref | 0.26 (±0.053) | ref |
| TCM-HS | ||||
| LR | 0.56 (±0.058) | .000 | 0.33 (±0.053) | .000 |
| RF | 0.55 (±0.056) | .000 | 0.30 (±0.036) | .000 |
| XGB | 0.56 (±0.066) | .000 | 0.31 (±0.066) | .000 |
| DAP | 0.74 (±0.035) | ref | 0.46 (±0.073) | ref |
The records from PHS were removed.
The numbers in parentheses are the standard deviation. The bolded part indicates the best performance of the corresponding data under the respective metric. Evaluation metrics included ROC-AUC and PR-AUC. We conducted a t-test to compare the differences in results between the two groups generated from the 10-fold data.
Abbreviations: DAP, depression and anxiety prediction; LR, logistic regression; NHIP, Nanjing Health Information Platform; PHS, primary healthcare services; RF, random forest; TCM-HS, Traditional Chinese Medicine healthcare service; XGB, extreme gradient boosting.