[995] Validation of Preoperative Predictive Models for Adverse Radical Prostatectomy Outcomes

K Trpkov, J Zhang, P Faris, A Yilmaz. Calgary Laboratory Services and University of Calgary, Calgary, Canada

Background: We previously developed preoperative predictive models for adverse radical prostatectomy (RP) outcomes, such as: stage pT3 prostate cancer, positive seminal vesicles (SV) or lymph nodes (LN), positive margins and tumor (TU) volume ≥10% gland volume (Lab Invest 2006;86:136A (Suppl 1).
Design: Predictive models were developed on a cohort comprising 1174 RP with matched 10-core biopsies (Cohort 1), performed in our institution from 07/2000 to 04/2005. We performed univariate and multivariate logistic regression analysis on Cohort 1 to establish the best predictors for adverse RP outcomes. The following biopsy variables were evaluated: total percent cancer (TPC), cancer length in mm, number of positive cores, Gleason Score (GS), perineural invasion, in addition to prostate specific antigen (PSA) and PSA density (PSAD). Validation of the predictive models was done on a separate cohort of 971 RP with matched biopsies (Cohort 2), accrued in our institution from 05/2005 to 12/2008.
Results: Multivariate models for the adverse RP outcome performed in the validation Cohort 2 as follows:

RP OutcomeOR (P)OR (P)OR (P)
Stage pT3GS≥812.9 (<0.001)TPC>30%5 (<0.001)PSAD>0.32.1 (<0.01)
Positive SV or LNGS≥817.1 (<0.001)TPC>30%6.5 (0.004)PSAD>0.32.4 (0.013)
Positive margins>3 positive cores2.6 (<0.001)Neural invasion1.5 (0.011)GS 71.5 (0.023)
TU vol>10% gl volTPC>30%10.8 (<0.001)PSAD >0.34.4 (<0.001)>3 positive cores4.2 (<0.001)
Odds Ratio (OR) calculated vs. reference

Comparison of the areas under the curves (AUC) for the initial Cohort 1 and the validation Cohort 2 are shown in the following table:

Stage pT3Positive SV or LNPositive marginsTU vol>10% gl vol
Cohort 1 AUC (CI)0.80 (0.76-0.84)0.82 (0.76-0.87)0.65 (0.61-0.68)0.77 (0.75-0.80)
Cohort 2 AUC (CI)0.78 (0.72-0.81)0.81 (0.76-0.86)0.67 (0.63-0.70)0.79 (0.76-0.82)
AUC, area under the curve; CI, confidence intervals

Conclusions: We validated the previously developed predictive models for adverse RP outcomes on an independent cohort from our institution. The models performed similarly in both cohorts. Validation of the models on cohorts from other institutions may further test their performance and substantiate their utility.
Category: Genitourinary (including renal tumors)

Monday, March 22, 2010 1:00 PM

Poster Session II # 125, Monday Afternoon


Close Window