Constructing and Validating Models for Predicting Gleason Grade Group Upgrading following Radical Prostatectomy in...

The occurrence of Gleason grade group upgrading (GGU) significantly impacts treatment strategy developments. We aimed to develop an optimal predictive model to assess the risk of GGU in patients with localized prostate cancer (PCa), by comparing traditional logistic regression (LR) with seven machine learning algorithms.

A retrospective collection of clinical data was conducted on patients who underwent radical prostatectomy at Wuhan Central Hospital (January 2017 to December 2023, n = 177) and Jiangxi Cancer Hospital (July 2019 to February 2024, n = 87). The least absolute shrinkage and selection operator regression was employed to filter the clinical characteristics of patients. Subsequently, models were conducted using multivariate LR, along with seven diverse machine learning algorithms: extreme gradient boosting, decision tree, multilayer perceptron, naive Bayes, K-nearest neighbors, random forest, and support vector machine. By employing the receiver operating characteristic curves, accuracy, brier score, recall, calibration curves, and decision curve analysis (DCA), we compared the predictive capabilities and clinical utility of eight models to identify the optimal one.

In the evaluation of eight models, the LR model demonstrated superior performance. In the modeling set, it achieved an area under curve (AUC) of 0.826 (95% CI: 0.808-0.845), accuracy of 0.765, and a brier score of 0.167. In the validation set, it kept good results with an AUC of 0.819 (95% CI: 0.758-0.880), accuracy of 0.725, and a brier score of 0.180. The calibration curves, brier score, and DCA also demonstrated the excellent calibration and net benefit of the LR model.

After conducting a comprehensive multi-model comparison, we concluded that the LR model was optimal for predicting GGU, which was confirmed by external validation. Our study also revealed percent free prostate-specific antigen density as a predictive factor for GGU, offering a novel approach for managing localized PCa patients.