The Prediction Model Of Distant Metastasis Of Pancreatic Cancer Established By Machine Learning

Background:Combine machine learning algorithms with the SEER database(a database which affiliated with the US national cancer institute)to assess predictors of distant metastasis in pancreatic cancer and to construct a predicted model,which in order to help the clinical diagnosis and treatment of pancreatic cancer patients.Method:Based on our screening criteria,13,606 pancreatic cancer cases were identified from the SEER database between 2010 and 2015.Two models for predicting distant metastasis of pancreatic cancer were established using traditional logistic regression statistical method and machine learning algorithm based on gradient elevation.The AUC,calibration curve and decision curve(DCA)were compared to verify the identification ability and clinical practicability of the two models.At the same time,the stepwise COX regression analysis was carried out and the survival curve of related factors was made by kaplan-meier method.Results:Among the 13,606 patients with pancreatic cancer,the median age was 65 years old,and the average age was 64.98 years old.Patients of pancreatic cancer with distant metastasis accounted for 25.02% of the total population.Patients with distant metastases to pancreatic cancer have a significantly poorer prognosis.The AUC of the XGBoost model was 0.77,which was better than the 0.75 of the logistic regression model.Moreover,the prediction results of XGBosst model were in good agreement with the correction curve for predicting distant metastasis in patients with pancreatic cancer,and the results of decision curve(DCA)also showed good clinical net benefits.The XGBoost prediction model has better practicability,reliability and clinical net benefit in evaluating distant metastasis of pancreatic cancer than traditional prediction model.Conclusion:The machine learning algorithm provides a better prediction technology than the traditional method?Prediction model of distant metastasis of pancreatic cancer can be established by using the machine learning algorithm,which can be easily applied in the clinic and help improve the clinical diagnosis and treatment of pancreatic cancer.