Novel Clinical Tools to Guide Management of Prostate Cancer: Preoperative Predictive Models to Identify Patients at Risk for Needing Adjuvant Radiation (T3 or +margin) or Having Detectable Postoperative PSA

Abstract

Several valuable models use preoperative data to predict organ- confined disease at the time of radical prostatectomy. However, there are no similar tools to predict which patients will qualify for adjuvant radiation therapy (ART), or have a detectable postoperative prostate specific antigen (PSA). We developed logistic regression models to predict risk for extracapsular extension (ECE) and seminal vesicle invasion (SVI) and compared them to others (Roach, CAPRA, Memorial Sloan Kettering, and Partin models). We then created two novel formulas to predict risk for ART (either T3 or +margin) or detectable postoperative PSA. We conducted a retrospective multi-institutional review of patients with clinically localized prostate adenocarcinoma who had radical prostatectomy from 2001-2015 (N=1,676) at Kaiser Permanente in Southern California. Patients treated with preoperative leuprolide or missing data were excluded. Three multivariate models were developed to predict ECE, SVI, and either ECE or SVI. A fourth model was developed to predict ECE/SVI/+margin. A fifth model was developed to predict detectable postoperative PSA (≥0.1). The area under the receiver curve (AUC) was calculated and corrected using bootstrap resampling with 2000 samples. Statistical analysis was performed with R version 3.3.1. Statistically significant factors on multivariate analysis included percent positive biopsy cores, biopsy Gleason score, PSA, and clinical T-stage. Age and year of surgery were not statistically significant. The ECE, SVI, and ECE/SVI models (AUC 0.69-0.81) performed favorably compared to other tools (AUC 0.72-0.79) [Table 1]. Our T3/+margin model performs at corrected AUC 0.75. Our detectable postoperative PSA model performs at corrected AUC 0.81 [Table 1]. We developed three models predicting risk of extraprostatic disease, which compared favorably over other widely used models. We created two novel clinical formulas to predict preoperatively whether patients may require ART or have a detectable postoperative PSA. These tools may help guide the selection of surgery or radiation upfront, and could potentially reduce dual modality treatment. While our novel clinical prediction models performed well in our data set, no known tools exist for comparison, and external validation is needed.Abstract 2630; Table 1RoachCAPRAMSKPartinKaiserECE (25% risk) AUC0.740.740.760.720.69 Sen/Spec1.00/00.63/0.701.00/00.87/0.460.37/0.91 PPV/NPV0.22/ NA0.39/0.860.22/0.830.31/0.920.54/0.84SVI (25% risk) AUC0.760.780.790.760.81 Sen/Spec0.47/0.850.22/0.960.28/0.960.03/0.990.39/0.94 PPV/NPV0.27/0.930.39/0.910.47/0.920.40/0.900.42/0.93ECE/SVI (40% risk) AUC---0.760.79 Sen/Spec---0.64/0.720.52/0.86 PPV/NPV---0.48/0.830.61/0.82ECE, SVI, or +margin (20% risk) AUC----0.75 Sen/Spec----0.91/0.35 PPV/NPV----0.44/0.87Detectable 1st Postop PSA (20% risk) AUC----0.81 Sen/Spec----0.36/0.95 PPV/NPV----0.34/0.96 Open table in a new tab