Using support vector machine analysis to assess PartinMR: A new prediction model for organ-confined prostate cancer

Background: Partin tables represent the most widely used predictive tool for prostate cancer stage at prostatectomy but with potential limitations.

Purpose: To develop a new PartinMR model for organ-confined prostate cancer (OCPCA) by incorporating Partin table and mp-MRI with a support vector machine (SVM) analysis.

Study type: Retrospective.

Population: In all, 541 patients with biopsy-confirmed prostate cancer underwent mp-MRI.

Field strength: T₂ -weighted, diffusion-weighted imaging with a 3.0T MR scanner.

Assessment: Candidate predictors included age, prostate-specific antigen, clinical stage, biopsy Gleason score (GS), and mp-MRI findings, ie, tumor location, Prostate Imaging and Reporting and Data System (PI-RADS) score, diameter (D-max), and 6-point MR stage. The PartinMR model with combination of a Partin table and mp-MRI findings was developed using SVM and 5-fold crossvalidation analysis.

Statistical tests: The predicted ability of the PartinMR model was compared with a standard Partin and a modified Partin table (mPartin) which used for mp-MRI staging. Statistical tests were made by area under receiver operating characteristic curve (AUC), adjusted proportional hazard ratio (HR), and a cost-effective benefit analysis.

Results: The rate of OCPCA at prostatectomy was 46.4% (251/541). Using MR staging, mPartin table (AUC, 0.814, 95% confidence interval [CI]: 0.779-0.846, P = 0.001) is appreciably better than the Partin table (AUC, 0.730, 95% CI: 0.690-0.767). Contrarily, adding all MR variables, the PartinMR model (AUC, 0.891, 95% CI: 0.884-0.899, P < 0.001) outperformed any other scheme, with 79.3% sensitivity, 75.7% specificity, 79% positive predictive value, and 76.0% negative predictive value for OCPCA. MR stage represented the most influential predictor of extracapsular extension (HR, 2.77, 95% CI: 1.54-3.33), followed by D-max (2.01, 95% CI: 1.31-2.68), biopsy GS (1.64, 95% CI: 1.35-2.12), and PI-RADS score (1.21, 95% CI: 1.01-1.98).

Data conclusion: The new PartinMR model is superior to the conventional Partin table for OCPCA. Clinical implications of mp-MRI for prostate cancer stage must be confirmed in further trials.

Level of evidence: 3 Technical Efficacy: Stage 2 J. MAGN. RESON. IMAGING 2018;48:499-506.