Validation of Partin tables for predicting pathological stage of clinically localized prostate cancer

Purpose: The accurate prediction of pathological stage of prostate cancer using preoperative factors is a critical aspect of treatment. In 1997 Partin et al published tables predicting pathological stage using clinical stage, Gleason score and prostate specific antigen (PSA). We tested the validity of the Partin tables. Materials and Methods: From 1990 to 1996 inclusively 5,780 patients underwent bilateral pelvic lymphadenectomy and radical prostatectomy for prostate cancer at the Mayo Clinic. However, only 2,475 of these patients met all inclusion criteria of no preoperative treatment, known biopsy Gleason score, available preoperative PSA done either before biopsy or more than 28 days after biopsy and clinical stage T1, T2 or T3a. Among the 2,475 patients 15 had positive lymph nodes and planned prostatectomy was abandoned. The receiver operating characteristics (ROC) curve area, observed and predicted Partin rates of each pathological stage, and positive and negative predictive values were used to compare the Mayo study to the Partin tables. Results: The distribution of pathological stage was organ confined in 67% of Mayo cases versus 48% in the Partin study, extracapsular without seminal vesicle or node involvement in 18% versus 40%, seminal vesicle involvement without nodes in 9% versus 7% and were positive nodes in 6% versus 5%. Using the predicted probabilities of Partin et al the ROC curve area for predicted node positive disease was 0.84 for Mayo cases compared to an estimated 0.82 in the Partin series. The ROC curve area for predicting organ confined cancer was 0.76 for the Mayo Clinic compared to an estimated 0.73 for the Partin series. The observed rates of node positive disease were similar to those predicted (Partin) based on clinical stage, PSA and Gleason score. For organ confined disease Mayo rates were consistently higher than those predicted from the Partin series using a cut point of 0.50 or greater. Positive and negative predictive values were 0.83 and 0.49 versus 0.63 and 0.70 for the Mayo Clinic and Partin series. Conclusions: Our study provides strong evidence that sensitivity and specificity of the Partin tables for external clinical sites are similar to what was reported.