[A preoperative prediction model for pelvic lymph node metastasis in prostate cancer: Integrating clinical characteristics and multiparametric MRI].

OBJECTIVE

To analyze the clinical features associated with pelvic lymph node metastasis (PLNM) in prostate cancer and to construct a preoperative prediction model for PLNM, thereby reducing unnecessary extended pelvic lymph node dissection (ePLND).

METHODS

Based on predefined inclusion and exclusion criteria, 344 patients who underwent radical prostatectomy and ePLND at the First Affiliated Hospital of Zhengzhou University between 2014 and 2024 were retrospectively enrolled, among whom, 77 patients (22.4%) were pathologically confirmed to have lymph node-positive disease. The clinical characteristics, MRI reports, and pathological results were collected. The data were then randomly divi-ded into a training cohort (241 cases, 70%) and a validation cohort (103 cases, 30%). Univariate and multivariate Logistic regression analysis were employed to construct a preoperative prediction model for PLNM.

RESULTS

Univariate Logistic regression analysis revealed that total prostate specific antigen (tPSA) (=0.021), free prostate specific antigen (fPSA) (=0.002), fPSA to tPSA ratio (fPSA/tPSA) (=0.011), percentage of positive biopsy cores ( < 0.001), prostate imaging reporting and data system (PI-RADS) score (=0.004), biopsy Gleason score ≥8 (=0.005), clinical T stage ( < 0.001), and MRI-indicated lymph node involvement (MRI-LNI) ( < 0.001) were significant predictors of PLNM. Multivariate Logistic regression analysis demonstrated that the percentage of positive biopsy cores (=91.24, 95%: 13.34-968.68), PI-RADS score (=7.64, 95%: 1.78-138.06), and MRI-LNI (=4.67, 95%: 1.74-13.24) were independent risk factors for PLNM. And a novel nomogram for predicting PLNM was developed by integrating all these three variables. Compared with the individual predictors: percentage of positive biopsy cores [area under curve (AUC)=0.806], PI-RADS score (AUC=0.679), and MRI-LNI (AUC=0.768), the multivariate model incorporating all three variables demonstrated significantly superior predictive performance (AUC=0.883). Consistently, calibration curves and decision curve analyses confirmed that the multivariable model had high predictive accuracy and provided significant net clinical benefit relative to single-variable models. And using a cutoff of 6%, the multiparameter model missed only approximately 5.2% of PLNM cases (4/77), while reducing approximately 53% of ePLND procedures (139/267), demonstrating favorable predictive efficacy.

CONCLUSION

Percentage of positive biopsy cores, PI-RADS score and MRI-LNI are independent risk factors for PLNM. The constructed multivariate model significantly improves predictive efficacy, offering a valuable tool to guide clinical decisions on ePLND.