Table 3. Low-Risk Threshold Summary.
| Risk threshold (percentile)a | TPR (95% CIb) | FPR (95% CIb) | NPV (95% CIb) | PPV (95% CIb) | Follow-ups avoided per 1000 menc,d | Reclass missed, No. (95% CI) per 1000 mene |
|---|---|---|---|---|---|---|
| PASS | ||||||
| 0.08 (10th) | 0.98 (0.93-1.00) | 0.87 (0.75-1.00) | 0.95 (0.89-1.00) | 0.26 (0.23-0.30) | 100 | 5 (0-11) |
| 0.10 (15th) | 0.93 (0.86-1.00) | 0.80 (0.65-0.95) | 0.90 (0.84-0.97) | 0.27 (0.23-0.31) | 150 | 14 (5-24) |
| 0.11 (20th) | 0.90 (0.81-0.98) | 0.73 (0.58-0.89) | 0.88 (0.82-0.94) | 0.28 (0.24-0.32) | 200 | 23 (12-35) |
| 0.13 (25th) | 0.87 (0.78-0.95) | 0.67 (0.51-0.83) | 0.88 (0.83-0.94) | 0.29 (0.24-0.34) | 250 | 29 (16-42) |
| UCSF | ||||||
| 0.08 (3rd) | 0.99 (0.96-1.00) | 0.96 (0.93-0.99) | 0.87 (0.54-1.00) | 0.35 (0.30-0.41) | 34 (21-49) | 4 (0-15) |
| 0.10 (7th) | 0.98 (0.95-1.00) | 0.89 (0.84-0.94) | 0.92 (0.77-1.00) | 0.37 (0.32-0.43) | 66 (47-85) | 5 (0-15) |
| 0.11 (9th) | 0.96 (0.93-0.99) | 0.86 (0.81-0.92) | 0.88 (0.71-0.98) | 0.37 (0.32-0.43) | 94 (71-116) | 12 (2-26) |
| 0.13 (13th) | 0.94 (0.89-0.98) | 0.82 (0.77-0.89) | 0.84 (0.70-0.94) | 0.38 (0.32-0.43) | 126 (99-152) | 20 (7-38) |
Abbreviations: FPR, false-positive rate; NPV, negative predictive value; PASS, Canary Prostate Active Surveillance Study; PPV, positive predictive value; PSA, prostate-specific antigen; reclass, reclassification; TPR, true-positive rate; UCSF, University of California, San Francisco.
Risk thresholds calculated based on PASS model fit, predicting out 4 years from confirmatory biopsy. Risk threshold is set at PASS risk percentiles; corresponding UCSF percentiles are listed.
95% CI based on 500 bootstrap samples.
Includes all surveillance biopsies and PSA follow-ups for 4 years.
The PASS data are defined exactly by the row quantiles: By definition, a 10th-percentile threshold will avoid 10% of biopsies, a 15th-percentile threshold will avoid 15%, etc. The UCSF data are validating the Canary thresholds, with ranges in parentheses.
In PASS, the estimated number of reclassifications at 4 years from confirmatory biopsy is 245 per 1000 men; in UCSF, the No. is 346. Canary model predictions are based on 4-year predictions from confirmatory biopsy. PASS estimates have been corrected for overfitting using 250 repeats of 2-fold cross-validation.