基于极端梯度提升算法的高血压识别模型建立

Establishment of Hypertension Recognition Model based on Extreme Gradient Boosting Algorithm

目的探索基于极端梯度提升(extreme gradient boosting,XGBoost)算法构建的高血压识别模型性能。方法本研究收集了2020年1月至12月南京大学附属鼓楼医院健康管理中心健康体检人群中1577位高血压确诊患者和3754位同期健康对照的相关数据,采用单因素分析对高血压影响因素进行筛选,基于XGBoost算法和自适应增强(AdaBoost)算法构建高血压识别模型,采用留出法验证模型泛化性能,灵敏度、特异度、阳性预测值、准确度、G-mean、F-measure、马修斯相关系数(MCC)和受试者特征曲线下面积综合评价和比较模型性能。结果XGBoost模型灵敏度(90.3%)、特异度(86.8%)、阳性预测值(87.3%)、准确度(88.6%)、G-mean(0.886)、F-measure(0.888)、MCC(0.772)和受试者工作特征曲线下面积(0.954)表明其具有更好的识别高血压患者的能力。结论XGBoost算法对识别高血压患者具有较强的实用性和可行性,为未来类似研究提供一定的模型选择参考。

Objective Explore the performance of Extreme Gradient Boosting(XGBoost)algorithm in identifying patients with hypertension.Methods The data were collected from 1577 patients with hypertension and 3754 healthy control participants at the Health Management Centre of Drum Tower Hospital from January 2020 to December 2020.Univariate analysis was conducted to select the factors.XGBoost and AdaBoost algorithms were used to construct the recognition models,which will be demonstrated by the holdout cross-validation.The performance of models was evaluated and compared by sensitivity,specificity,accuracy,positive predictive value,G-mean,F-measure,Matthews correlation coefficient(MCC)and the area under the receiver operating characteristic curve(AUC).Results The sensitivity(90.3%),specificity(86.8%),positive predictive value(87.3%),accuracy(88.6%),G-mean(0.886),F-measure(0.888),MCC(0.772)and AUC(0.954)indicated the superior ability of XGBoost model in identifying patients with hypertension.Conclusion The XGBoost algorithm has strong practicability and feasibility for identifying patients with hypertension,which provides a model selection reference for similar studies in the future.