基于XGBoost算法的肝移植术后患者服药依从性预测模型研究

A predictive modeling study of medication adherence among patients after liver transplantation based on XGBoost algorithm

目的:分析肝移植术后患者服药依从性的影响因素,并基于XGBoost算法构建预测模型。方法:分析2020年6月至2022年6月于郑州大学第一附属医院肝移植中心接受肝移植的99例患者临床及随访资料,采用免疫抑制药物依从性Basel评估量表评价患者移植术后服药依从性。运用Logistic回归分析服药依从性影响因素,运用XGBoost算法建立各变量与依从性风险模型,利用灵敏度、特异度、准确率、Kappa值、曲线下面积(AUC)值对模型进行评估。结果:在本次调查中,依从性存在问题者(Basel评分≥4)38例(38.38%),Logistic回归分析显示,平均月收入{比值比(OR)[95%可信区间(CI)]:0.21(0.15,0.36),P<0.01}、对药物的了解情况[OR(95%CI):2.64(1.56,4.03),P<0.01]及定期复查[OR(95%CI):8.67(3.67,15.86),P<0.01]是肝移植术后患者服药依从性影响因素(P<0.05);XGboost模型中肝移植术后患者服药依从性影响因素的重要性排序从大到小依次为定期复查、平均月收入、对药物的了解情况、每天药物服用次数、合并抑郁及教育背景;与Logistic回归模型比较,XGBoost预测模型准确率(92.4%比88.0%)、敏感度(95.1%比90.0%)、特异度(85.1%比82.0%)、Kappa值(0.84比0.71)及AUC值(0.86比0.71)更高(Z=2.313,P<0.05)。结论:基于XGBoost算法建立的肝移植术后服药依从性预测模型可用于患者服药依从性预测。

Objective:To analyze the influencing factors of medication compliance of patients after liver transplantation and construction of prediction model based on XGBoost algorithm.Methods:Analyze the clinical and follow-up data of 99 patients who underwent liver transplantation at the Liver Transplantation Center of the First Affiliated Hospital of Zhengzhou University from June 2020 to June 2022.Evaluate patients’post-transplant medication adherence using the Immunosuppressive Drug Adherence Basel Assessment Scale.Use Logistic regression to analyze factors influencing medication adherence,and employ the XGBoost algorithm to establish a risk model for adherence with various variables.Evaluate the model using sensitivity,specificity,accuracy,Kappa value,and area under curve(AUC)value.Results:In this survey,38 cases(38.38%)had problems with compliance.Logistic regression analysis found that average monthly income[OR(95%CI):0.21(0.15,0.36),P<0.01],knowledge of drugs[OR(95%CI):2.64(1.56,4.03),P<0.01]and regular review[OR(95%CI):8.67(3.67,15.86),P<0.01]were the factors affecting medication compliance of patients after liver transplantation(P<0.05).In the XGboost model,the importance of influencing factors on medication compliance of patients after liver transplantation was ranked in descending order as regular review,average monthly income,knowledge of drugs,number of times per day of drug taking,combined depression,and educational background.Compared with Logistic regression model,XGBoost model had higher accuracy(92.4%vs.88.0%),sensitivity(95.1%vs.90.0%),specificity(85.1%vs.82.0%),Kappa value(0.84 vs.0.71)and AUC value(0.86 vs.0.71,Z=2.313,P<0.05).Conclusion:The prediction model of medication compliance after liver transplantation based on XGBoost algorithm can be used to predict patients’medication compliance.