Incorporating tumor pathological subtype into a multiparameter model: Improved accuracy in predicting prostate cancer extraprostatic extension.

e17117 Background: Extraprostatic extension (EPE) at radical prostatectomy (RP) is associated with worse biochemical recurrence-free survival and informs nerve-sparing decisions in localized prostate cancer. Existing nomograms and MRI-based scoring systems provide moderate accuracy, and no prior model has incorporated tumor pathological subtype. This study evaluated whether adding pathological subtype improves EPE prediction beyond MRI-based parameters. Methods: We retrospectively analyzed 343 patients who underwent preoperative multiparametric MRI and RP (August 2023-August 2025). EPE was present in 138 patients. Cases treated after January 2025 formed a temporal validation cohort (n=82), with earlier cases assigned to training (n=183) and test (n=78) cohorts. Candidate predictors included MRI-based parameters, clinical features, and pathological subtype. Logistic regression–based models and machine-learning algorithms were developed; model selection was based on multivariable performance. Results: The final model included pathological subtype (acinar adenocarcinoma, PAA), ESUR score, PSA density (PSAD), capsular contact length (CCL), and positive core rate. The model demonstrated superior discrimination compared with MRI-based systems, achieving an AUC of 0.829 in the training cohort, 0.795 in the test cohort, and 0.849 in the temporal validation cohort. Decision-curve analysis showed greater net clinical benefit across threshold probabilities. SHAP analysis identified pathological subtype as a major contributor to prediction. Conclusions: A pathology-centric multiparameter model incorporating PAA improves the accuracy and generalizability of EPE prediction. Pathological subtype contributes independent predictive value beyond MRI-based measures. This model may assist preoperative risk stratification and guide nerve-sparing decisions in patients undergoing radical prostatectomy. Comparison of AUC values among different models. Model Training Set AUC Test Set AUC Temporal Validation Set AUC EPE Grade 0.758 0.698 0.731 ESUR Score 0.774 0.745 0.765 Our Model 0.829 0.795 0.849 EPE Grade=MRI-based EPE grade. ESUR Score=The scoring scheme proposed by the European Society of Urogenital Radiology. AUC=Area Under the Curve.