Table 2.
Model performance in the full-cohort internal validation analysis.
| Model | AUROC (95% CI) |
Sensitivity (95% CI) |
Specificity (95% CI) |
Brier score |
Calibration intercept |
Calibration slope |
|---|---|---|---|---|---|---|
| Logistic regression | 0.840 (0.754-0.916) |
0.512 (0.365-0.657) |
0.967 (0.925-0.986) |
0.105 | -0.122 | 0.917 |
| Random forest | 0.777 (0.675-0.876) |
0.488 (0.343-0.635) |
0.947 (0.900-0.973) |
0.121 | -0.683 | 0.321 |
| XGBoost | 0.768 (0.667-0.854) |
0.463 (0.321-0.613) |
0.947 (0.900-0.973) |
0.128 | -0.394 | 0.520 |
| Support vector machine | 0.757 (0.648-0.855) |
0.537 (0.387-0.679) |
0.947 (0.900-0.973) |
0.115 | -0.280 | 0.852 |
Performance estimates were derived from 5-fold cross-validated out-of-fold predictions in the full cohort, with feature selection embedded within each training fold by LASSO logistic regression. The final displayed predictor set for the logistic model was age, PHI and PHID. Classification metrics were calculated using a probability threshold of 0.50. AUROC 95% confidence intervals were estimated using 1,000 patient-level bootstrap resamples. AUROC, area under the receiver operating characteristic curve; CI, confidence interval; CSPCa, clinically significant prostate cancer; PHI, Prostate Health Index; PHID, PHI density; XGBoost, extreme gradient boosting.