RESUMO
OBJECTIVE: To establish a predictive model for prostate cancer bone metastasis utilizing multiple machine learning algorithms. METHODS: Retrospective analysis of the clinical data of prostate cancer initially diagnosed in the Department of Urology of Gansu Provincial People's Hospital from June 2017 to June 2022. Logistic regression (LR) and least absolute shrinkage and selection operator (LASSO) are used to jointly screen the model features. The filtered features are incorporated into algorithms including LR, random forest (RF), extreme gradient boosting (XGBoost), naive Bayes (NB), k-nearest neighbor (KNN), and decision tree (DT), to develop prostate cancer bone metastasis models. RESULTS: A total of 404 patients were finally screened. Gleason score, T stage, N stage, PSA, and ALP were used as features for modeling. The average AUC of the 5-fold cross-validation for each machine learning model in the training set is as follows: LR (AUC=0.9054), RF (AUC=0.9032), NB (AUC=0.8961), KNN (AUC=0.8704), DT (AUC=0.8526), XGBoost (AUC=0.8066). The AUC of each machine learning model in the test set is KNN (AUC=0.9390, 95%CI: 0.8760-1), RF (AUC=0.9290, 95%CI: 0.8718-0.9861), NB (AUC=0.9268, 95%CI: 0.8615-0.9920), LR (AUC=0.9212, 95%CI: 0.8506-0.9917), XGBoost (AUC=0.8292, 95%CI: 0.7442-0.9141), DT (AUC=0.8057, 95%CI: 0.7100-0.9014). A comprehensive evaluation showed that LR performed well in interpretability and clinical applications. CONCLUSION: A bone metastasis model of prostate cancer was established, and it was observed that indicators such as inflammation and nutrition had a weak correlation with bone metastasis.