基于集成学习的2型糖尿病患者降糖药用药方案智能分类探讨

宋亚男; 金昕晔; 张颖; 陈康; 应俊; 薛万国; 母义明

doi:10.3969/j.issn.2095-5227.2019.08.004

基于集成学习的2型糖尿病患者降糖药用药方案智能分类探讨

Intelligent classification of hypoglycemia treatment plan for patients with type 2 diabetes based on ensemble learning

摘要

摘要:
目的探讨集成学习中的Adaboost算法在2型糖尿病患者降糖药用药模式分析中的应用。
方法收集解放军总医院第一医学中心2013 - 2017年的2型糖尿病住院患者病例资料3 005例，随机选择1 697例为训练集，1 308例为测试集，根据医嘱用药、生化检验、基本体征、人口统计学等资料，应用Adaboost算法建立学习模型，对患者用药模式进行分类，并计算模型的准确性和Kappa系数。
结果 Adaboost模型预测的用药分类准确率为64.2%，Kappa系数为0.36。通过Adaboost模型分析，发现与降糖药用药相关的重要变量有尿肌酐、糖化血红蛋白、肌酸激酶同工酶、空腹血糖等。
结论 Adaboost算法在降糖药用药方案的预测方面具有较好的效果，集成学习方法在患者用药决策方面具有一定可行性。

Abstract:
Objective To apply Adaboost in the determination of hypoglycemia treatment plan in patients with type 2 diabetes.
Methods Clinical data about 3 005 patients with type 2 diabetes hospitalized in the first medical center of Chinese PLA General Hospital from 2013 to 2017 were collected, including medical prescriptions, biochemical testing results, clinical manifestations, demographic characteristics, etc. Adaboost algorithm was used to establish the machine learning model and classify the treatment plan of the patients, with 1 697 cases as training set and 1 308 cases as testing set randomly, and then accuracy and Kappa coefficient of the model were computed.
Results The prediction accuracy of the model by Adaboost was 64.2% and the Kappa coefficient was 0.36. After analyzing the model established by Adaboost, we found that UCr, HbA1c, CK-MB, FBG, etc. were significantly related to the treatment plan selecting.
Conclusion To some extent, Adaboost algorithm is feasible and accurate in predicting hypoglycemia treatment plan.

HTML全文

参考文献(20)

施引文献

资源附件(0)