Predictive accuracy of PSA density in the UCSF-CAPRA score.

Abstract

233 Background: The UCSF-CAPRA score is a validated risk tool that uses patient factors at diagnosis to predict prostate cancer outcomes after radical prostatectomy. This study aims to evaluate whether the substitution of prostate-specific antigen (PSA) density for serum PSA improves the predictive performance of the clinical CAPRA model. Methods: Participants were diagnosed in 2000-2019 with stage T1/T2 cancer, underwent radical prostatectomy (RP) for prostate cancer, and had at least 6 months of post-surgical follow up. We computed the standard CAPRA clinical risk score using diagnostic age, Gleason grade, percentage of positive cores, clinical T stage, and serum PSA, as well as an alternate score based on the same variables but substituting PSA density (incremental categories) for serum PSA. We reported validated CAPRA categories of low (0-2), intermediate (3-5) and high (6-10) risk. Recurrence was defined as 2 consecutive PSA > = 0.2 ng/ml or receipt of salvage treatment. Life table and Kaplan-Meier analysis was used to evaluate recurrence-free survival after prostatectomy. Two Cox proportional hazards regression models were used to test associations of standard or alternate CAPRA component variables with risk of recurrence. Two additional models tested associations between standard CAPRA score or alternate CAPRA score with risk of recurrence. The Cox log-likelihood ratio test (-2 LOG L) was used to assess model accuracy. Results: A total of 2,557 patients had a median age of 61 years (IQR 56-66), GG1 34% and GG2 30%, median PSA 6.3 (IQR 4.8-9.3), and median PSA density 0.19 (IQR 0.13-0.29). Post-surgical follow up was 49 months (IQR 25-86). The alternate CAPRA model was associated with significant shifts in risk scores, with 16% of patients increasing and 6% decreasing (p < 0.01). Recurrence-free survival after RP was 75% at 5 years and 63% at 10 years. Both CAPRA component models were associated with risk of recurrence after RP on Cox regression. Covariate fit statistics showed significantly better fit (low -2 LOG L) for the standard CAPRA model compared to the alternate model (-p < 0.01). Both standard (HR 1.55; 95% CI 1.49-1.61) and alternate (HR 1.52; 95%CI 1.46-1.58) CAPRA scores were associated with risk of recurrence, with better fit for the standard model p < 0.01). Conclusions: In a cohort of 2,557 men followed for a median of 49 months after RP, a alternate CAPRA model using PSA was associated with higher risk of BCR, but did not perform better than the standard CAPRA model at predicting BCR. While PSA density is well established as an independent prognostic variable in the pre-diagnostic setting and for sub-stratifying otherwise low-risk disease, it does not appear to improve predictive accuracy of BCR models when applied across the full range of cancer risk.