Background Pseudomyxoma peritonei(PMP)is a rare clinical syndrome characterized by the accumulation of mucinous ascites in the abdominal cavity.The standard treatment for PMP involves cytoreductive surgery with hypert...Background Pseudomyxoma peritonei(PMP)is a rare clinical syndrome characterized by the accumulation of mucinous ascites in the abdominal cavity.The standard treatment for PMP involves cytoreductive surgery with hyperthermic intraperitoneal chemotherapy(CRS+HIPEC).However,there is significant variability in patient outcomes following surgical intervention,highlighting the need for an effective prognostic tool to predict individual survival rates.Recent advancements have shown that machine learning(ML)algorithms hold great promise in developing predictive models within the medical field.Methods This study involved a cohort of 577 patients diagnosed with PMP who underwent CRS+HIPEC.Data analysis was conducted using three distinct ML algorithms:Random Forest(RF),Support Vector Machine(SVM),and Artificial Neural Network(ANN).The effectiveness of these models in predicting 3-year and 5-year survival rates was systematically assessed,leading to the identification of the most efficient algorithm.Results The area under the ROC curve(AUC)for the three machine learning methods was evaluated.The AUC of RF for predicting 3-year survival emerged as the most effective model among the three methodologies(0.797 vs.0.773 vs.0.735).Additionally,RF also demonstrated superior performance in predicting 5-year survival(0.792 vs.0.759 vs.0.736),thereby establishing RF as the optimal machine learning approach for survival prediction.Furthermore,the nomogram validation set generated by RF indicated that the AUC for the 3-year validation cohort was 0.745,while that for the 5-year validation cohort was recorded at 0.722,underscoring this model's enhanced accuracy and robustness in forecasting survival rates.Conclusions This study successfully developed and validated a nomogram model using the RF algorithm to predict 3-year and 5-year survival rates in patients with PMP,providing a practical and efficient tool for clinical decision-making.展开更多
文摘Background Pseudomyxoma peritonei(PMP)is a rare clinical syndrome characterized by the accumulation of mucinous ascites in the abdominal cavity.The standard treatment for PMP involves cytoreductive surgery with hyperthermic intraperitoneal chemotherapy(CRS+HIPEC).However,there is significant variability in patient outcomes following surgical intervention,highlighting the need for an effective prognostic tool to predict individual survival rates.Recent advancements have shown that machine learning(ML)algorithms hold great promise in developing predictive models within the medical field.Methods This study involved a cohort of 577 patients diagnosed with PMP who underwent CRS+HIPEC.Data analysis was conducted using three distinct ML algorithms:Random Forest(RF),Support Vector Machine(SVM),and Artificial Neural Network(ANN).The effectiveness of these models in predicting 3-year and 5-year survival rates was systematically assessed,leading to the identification of the most efficient algorithm.Results The area under the ROC curve(AUC)for the three machine learning methods was evaluated.The AUC of RF for predicting 3-year survival emerged as the most effective model among the three methodologies(0.797 vs.0.773 vs.0.735).Additionally,RF also demonstrated superior performance in predicting 5-year survival(0.792 vs.0.759 vs.0.736),thereby establishing RF as the optimal machine learning approach for survival prediction.Furthermore,the nomogram validation set generated by RF indicated that the AUC for the 3-year validation cohort was 0.745,while that for the 5-year validation cohort was recorded at 0.722,underscoring this model's enhanced accuracy and robustness in forecasting survival rates.Conclusions This study successfully developed and validated a nomogram model using the RF algorithm to predict 3-year and 5-year survival rates in patients with PMP,providing a practical and efficient tool for clinical decision-making.
