构建和验证慢性胰腺炎发生胰管结石的预测模型:一种机器学习方法

Construction and validation of prediction models for pancreatic stone in chronic pancreatitis:A machine learning approach

目的探讨机器学习模型在慢性胰腺炎(CP)患者发生胰管结石风险预测中的效能。方法回顾性分析2016年1月-2021年1月因CP就诊于兰州大学第一医院患者的临床资料。通过基于支持向量机的递归特征消除法(RFE)筛选预测因子,使用4种机器学习模型拟合,受试者操作特征(ROC)曲线评估模型准确性。结果纳入患者128例,其中发生胰管结石患者44例,对照组84例。5倍交叉验证的RFE显示吸烟、糖尿病、胰腺假性囊肿、胰管引流、胰腺钙化、胰管直径是最相关的6个预测因子。建立4种机器学习模型的ROC曲线,随机森林模型在验证集的曲线下面积最大。结论随机森林模型为预测CP发生胰管结石风险概率的有效机器学习模型。当胰管直径达到4.10 mm后,发生胰管结石的风险概率均大于50%。

Objective To investigate the efficacy of machine learning model in predicting the risk of pancreatic stone in patients with chronic pancreatitis(CP).Methods The clinical data of CP patients admitted to The First Hospital of Lanzhou University from January 2016 to January 2021 were retrospectively analyzed.The predictive factors were screened by recursive feature elimination(RFE)method based on support vector machine.Four machine learning models were used for fitting,and the accuracy of the model was evaluated by receiver operating characteristic(ROC)curve.Results A total of 128 patients were included,including 44 patients with pancreatic stone and 84 patients in the control group.5-fold cross-validated RFE showed that smoking,diabetes mellitus,pancreatic pseudocyst,pancreatic duct drainage,pancreatic calcification,and pancreatic duct diameter were the six most relevant predictors.The ROC curve of the four machine learning models was established.The area under the curve of the random forest model was the largest in the validation.Conclusion Random forest model is the best machine learning model to predict the risk probability of pan-creatic duct stones in patients with CP.When the diameter of pancreatic duct reached 4.10 mm,the risk probability of pancreatic duct stones was greater than 50%.