[Preoperative predictive model for biochemical recurrence in patients with localized prostate cancer treated with radical prostatectomy].

OBJECTIVES

To identify pre-prostatectomy clinical prognostic factors for biochemical recurrence (BR) and to create a predictive model for BR based or predictive clinical variables prior to radical prostatectomy (RP).

METHODS

a retrospective case-records study of patients with clinically localized prostate cancer treated with RPas monotherapy pN0-pNx and monitored at least for 12 months between 1996 and 2007. We considered BR the PSA persistence or elevation after RP greater than 0.4 ng/ml. The clinical variables analyzed were PSA, clinical stage and Gleason score from the biopsy (GS). Univariate and multivariate analysis were carried out using the chi squared test and logistic regression to determine the variables associated with BR. In order to estimate BR based on the variables identified we developed a mathematical model and designed an Excel spreadsheet to apply it. Calibration and discrimination were performed using the Hosmer-Lemeshow test and an ROC curve determining the area under the curve.

RESULTS

We included 627 patients. The mean age was 64 years with a mean follow- up of 87 months. The mean PSA was 8 ng/ml. 68.6% of patients had a PSA ≤ 10 ng/ml, 53,1% had a GS ≤ 6 and 61,7% had a clinical stage of cT1a-c. BR was observed in 204 (32,5%) patients, 39 due to biochemical persistence. The mean time to BR was 28 months with 89,7% of instances occurring in the first 8 years. On the multivariate analysis, PSA and GS were independent predictors of BR ( p=0.001), while the cT2c stage had a tendency towards statistical significance ( p=0.06). The three variables were included in the equation for the model with different specific weight. Specificity was 93.6%, sensitivity was 36.8% and an overall precision of 75.1%. The model had a predictive capacity of 73% and a p-value < 0.001.

CONCLUSIONS

PSA and GS are independent prognostic clinical variables associated with BR-free survival. The predictive model developed allows the risk of BR to be estimated with 73% reliability.