摘要
目的分析突发性感音神经性聋(SSNHL)伴高血压患者发生不良预后的影响因素,基于4种机器学习算法构建预测模型,比较不同模型的预测效果。方法选取2023年2月1日—2024年5月31日就诊于郑州大学第二附属医院耳鼻咽喉科SSNHL合并高血压的患者作为研究对象。通过医院电子病历系统收集患者临床资料,根据治疗后听力恢复情况分为有效组和无效组,采用单因素分析、最小绝对收缩和选择算子(LASSO回归)和Boruta算法筛选预测变量。采用逻辑回归(LR)、随机森林(RF)、极端梯度提升树(XGBoost)和支持向量机(SVM)4种机器学习算法构建预测模型并进行验证,Delong检验各模型在验证集中的曲线下面积(AUC),同时比较各模型在验证集中的评价指标以确定最佳模型,并使用Shapley加性解释(SHAP)算法对模型进行解释性分析。结果232例患者纳入研究,共筛选出7个与无效预后密切相关的变量,包括听力损失程度、听力图类型、糖尿病、高血压病程等。对4种模型的预测性能进行验证,XGBoost模型整体预测性能最好,AUC(0.787)、准确度(78.56%)、精确度(78.95%)、召回率(71.43%)和F1分数(75%),且4种模型之间AUC的差异无统计学意义(P>0.05)。结论合并高血压的SSNHL患者不良预后的发生风险受患者初始听力水平、听力图类型、糖尿病、高血压病程、吸烟及高尿酸血症等因素影响。4种机器学习模型均具有良好的预测性能,其中XGBoost模型预测性能最优。
Objective To analyze the influencing factors for the poor prognosis in patients with sudden sensorineural hearing loss(SSNHL)accompanied with hypertension,construct prediction models based on four machine learning algorithms,and compare the prediction effects of different models.Methods Patients with SSNHL combined with hypertension admitted to the Department of Otorhinolaryngology of the Second Affiliated Hospital of Zhengzhou University from February 1,2023 to May 31,2024 were selected as the research subjects.The clinical data of patients were collected through the hospital's electronic medical record system.According to the hearing recovery after treatment,the patients were divided into the effective group and the ineffective group.Univariate analysis,the least absolute shrinkage and selection operator(LASSO regression),and the Boruta algorithm were used to screen the predictive variables.Four machine learning algorithms,namely logistic regression(LR),random forest(RF),extreme gradient boosting(XGBoost),and support vector machine(SVM),were adopted to construct prediction models and conduct verification.The predictive performances of the four models were compared.The Delong test was employed to compare the area under the curve(AUC)of the models in the validation set,and the evaluation indicators of each model in the validation set were compared to determine the optimal model.Moreover,the Shapley additive explanation(SHAP)algorithm was utilized to conduct explanatory analyses on the models.Results A total of 232 patients were included in this study,and 7 variables closely related to poor prognosis were screened out,including the degree of hearing loss,audiogram type,diabetes mellitus,and the course of hypertension.The predictive performance of the four models was verified.Among them,the XGBoost model demonstrated the best overall predictive performance,with an AUC of 0.787,an accuracy of 78.56%,a precision of 78.95%,a recall of 71.43%,and an F1 score of 75%.Moreover,there were no statistically significant differences in the AUC among the receiver operator characteristic curves of the four models(P>0.05).Conclusions The risk of poor prognosis in SSNHL patients with hypertension is affected by multiple factors,including initial hearing level,type of audiogram,diabetes mellitus,duration of hypertension,smoking and hyperuricemia.All the four machine learning models have good predictive performance,with the XGboost model being the optimal.
作者
杨瑾
米彦芳
YANG Jin;MI Yanfang(Department of Otorhinolaryngology,the Second Affiliated Hospital of Zhengzhou University,Zhengzhou 450003,China)
出处
《中国耳鼻咽喉颅底外科杂志》
2025年第2期30-36,共7页
Chinese Journal of Otorhinolaryngology-skull Base Surgery
关键词
突发性感音神经性聋
高血压
机器学习
预后模型
Sudden sensorineural hearing loss
Hypertension
Machine learning
Prognostic model
