VALIDATING THE PARTIN TABLES USING THE NCI SEER DATABASE

Abstract

INTRODUCTION AND OBJECTIVE: The Partin Tables are a widely used nomogram predicting the pathologic stage of prostate cancer (CaP) given common preoperative clinical characteristics. The model is based on patients undergoing radical prostatectomy (RP) at the Johns Hopkins Medical Institutions, a tertiary care center. We sought to validate the Partin tables using a large national cohort to see whether the model could be applied to all patients undergoing RP for CaP in the United States. METHODS: The published 2007 Partin Tables were used to estimate the risk of positive lymph nodes (LN+), seminal vesicle invasion (SV+), and extraprostatic extension (EPE+) in men with prostate cancer from the NCI-SEER Database using clinical stage, preoperative PSA, and Gleason score. The discriminative and predictive ability of the tables were explored by constructing receiver operating characteristic (ROC) and calibration curves. An overall Brier score was calculated for each endpoint. The utility of the nomogram was determined based on clinically relevant endpoints. RESULTS: We identified 11,185 men who underwent RP for CaP in the years 2004-2005. The Partin tables discriminated between patient groups at risk for lymph node involvement and seminal vesicle involvement well, with an AUC of 0.77 and 0.74, and Brier score 0.026 and 0.042, respectively. Extraprostatic extension discrimination and prediction was more limited (AUC 0.62, Brier score 0.171). CONCLUSIONS: The Partin tables showed excellent discrimination for seminal vesicle and lymph node involvement. Prediction of lymph node involvement was reasonable even within lower risk groups. Prediction of extraprostatic extension and seminal vesicle involvement was somewhat more limited.