Prospective validation of the Kaiser Permanente prostate cancer risk calculator in a contemporary, racially diverse, referral population.

PURPOSE: To prospectively validate a new prostate cancer risk calculator in a racially diverse population.

MATERIALS AND

METHODS

We recently developed, internally validated and published the Kaiser Permanente Prostate Cancer Risk Calculator. This study is a prospective validation of the calculator in a separate, referral population over a 21-month period. All patients were tested with a uniform PSA assay and a standardized systematic, ultrasound-guided biopsy scheme. We report on 3 calculator models: Model 1 included age, race, PSA, prior biopsy status, body mass index, and family history of prostate cancer; Model 2 added digital rectal exam to Model 1 variables; Model 3 added prostate volume to Model 2 variables. We considered three outcomes: high-grade disease (Gleason score ≥7), low-grade disease (Gleason score=6), and no cancer. Predictive discrimination and calibration were calculated. How each model might alter biopsy frequency and outcomes at various thresholds of risk was assessed. We compared the performance of our calculator with two other calculators.

RESULTS

In 4178 patients (16.2% Asian, 11.3% African American, 13.5% Hispanic), cancer was found in 53%; 62% were Gleason score ≥7. Using a high-grade risk threshold for biopsy of ≥10%, Model 2 predictions would result in 9% of men avoiding a biopsy, while only missing 2% of high-grade cancers. At the same threshold, Model 3 predictions would result in 26% of men avoiding a biopsy, while only missing 5% of high-grade cancers. The c-statistics for Models 1, 2, and 3 to predict high-grade disease vs. low-grade or no cancer were 0.76, 0.79 and 0.85, respectively. The c-statistics for Models 1, 2, and 3 to predict any prostate cancer vs. no cancer were 0.70, 0.72 and 0.80, respectively. All models were well calibrated for all outcomes. Our Model 3 calculator had superior discrimination for high grade disease (c-statistic=0.85, 0.84-0.86) and any cancer (0.80, 0.79-0.82) compared to the PBCG calculator [(0.79, 0.78-0.80); 0.72 (0.70-0.73)] and the PCPT calculator [(0.75, 0.74-0.77); 0.69 (0.67-0.70)], respectively. In the high-grade cancer predicted risk range of 0-30%, our Model 2 was better calibrated than the PCPT and PBCG calculators.

CONCLUSIONS

This validation of our calculator showed excellent performance characteristics.