Development and explainability of a machine learning prediction model for histological prostatic inflammation in surgically treated patients with benign prostatic hyperplasia: a single-center internal validation study.

Background: Histological prostatitis is highly prevalent among patients with benign prostatic hyperplasia (BPH) and has been shown to be closely associated with disease progression and postoperative outcomes. However, its diagnosis still mainly relies on postoperative or biopsy pathology, and there is a lack of non-invasive tools to estimate the risk of histological inflammation before surgery or biopsy. Methods: This was a single-center retrospective machine-learning prediction model study with internal validation, involving 723 patients admitted to the urology department of a tertiary medical center for BPH between June 2020 and June 2023. The cohort was randomly divided into a training set and a validation set in a 7:3 ratio. Least absolute shrinkage and selection operator (LASSO) regression and the Boruta algorithm were used for feature selection, on the basis of which six machine learning models were constructed. Model performance and clinical net benefit were evaluated using the area under the receiver operating characteristic curve, calibration, and decision curve analysis. Shapley additive explanations (SHAP) were applied to provide interpretability at both the global and individual levels, and the best-performing model was further deployed as an online prediction tool. Results: Among the 723 patients, 387 (53.5%) had histological prostatitis. Of the six machine learning models, the artificial neural network (ANN) model showed the best discriminative ability, with an AUC of 0.852 [95% confidence interval (CI): 0.804-0.899] in the validation set, overall performance superior to the other models, and the lowest Brier score (0.160). It also provided the greatest net benefit across a wide range of threshold probabilities. SHAP analysis indicated that prostate volume (PV), neutrophil-to-lymphocyte ratio (NLR), International Prostate Symptom Score (IPSS), age, systemic immune-inflammation index (SII), and acute urinary retention (AUR) were the key predictors driving model performance; higher values of these variables (or the presence of AUR) were associated with an increased predicted risk of histological prostatitis. Conclusions: Machine learning models, particularly the ANN model, showed good discriminative ability and reasonable calibration for predicting the risk of histological prostatic inflammation. Age, prostate volume, IPSS score, AUR, NLR, and SII were identified as the most important predictors. This prediction model enables accurate and interpretable risk assessment of histological prostatitis in patients with BPH, thereby facilitating early identification of high-risk individuals, supporting refined risk stratification, and optimizing perioperative decision-making.