[关键词]
[摘要]
目的 基于机器学习算法构建原发性开角型青光眼(POAG)伴视野缺损患者生活质量预测模型。方法 回顾性选取2020年5月~2025年5月期间在医院就诊的526例POAG伴视野缺损患者,采用国家眼科研究所视觉功能问卷25(NEI VFQ-25)评估患者生活质量,据此分为正常组与异常组,收集两组患者一般资料,应用最小绝对收缩和选择算子(LASSO)回归及Logistic回归分析筛选变量,纳入有统计学意义的因素(P<0.05)构建预测模型,并采用受试者工作特征曲线(ROC)的曲线下面积(AUC)、校准曲线及决策曲线对模型的区分度、校准度及临床实用性进行评价。结果 调查发现,526例POAG伴视野缺损患者中,NEI VFQ-25<50分共72例患者,纳入异常组(占比13.69%);余下454例患者NEI VFQ-25≥50分,纳入正常组(占比86.31%)。异常组的合并偏头痛占比、晚期视野缺损占比、高度近视占比及焦虑自评量表(SAS)评分、疾病不确定感量表成人版(MUIS-A)评分、匹兹堡睡眠质量指数量表(PSQI)评分均高于正常组。经LASSO回归筛选出的5个指标:合并偏头痛、视野缺损、SAS评分、MUIS-A评分、PSQI评分,将其进一步进行Logistic回归分析,结果表明,合并偏头痛(OR=5.215)、SAS评分(OR=1.174)、MUIS-A评分(OR=1.124)、PSQI评分(OR=1.352)是POAG视野缺损患者生活质量的危险因素(P<0.05),早期视野缺损(OR=0.121)为保护因素(P<0.05)。基于上述因素建立风险预测模型,绘制ROC曲线可知,该模型预测POAG视野缺损患者生活质量的AUC值为0.927(95%CI:0.895~0.960),且通过Bootstrap法重复1000次采样对模型进行验证,验证AUC值为0.916,表明该列线图模型具有较好的区分度。绘制校准曲线显示,校准曲线和参考曲线相近,Cox-Snell R2为0.587,Brier Score为0.057,且Hosmer-Lemeshow拟合度检验显示,模型拟合值和观测值的吻合程度一致,χ2= 9.770,P = 0.282>0.05,说明本次模型通过HL检验,模型拟合优度较好。决策曲线显示,阈值范围中预测模型净获益率较高,表明该模型在预测生活质量方面具有良好临床效用。结论 POAG伴视野缺损患者生活质量受合并偏头痛、视野缺损、SAS评分、MUIS-A评分、PSQI评分影响,基于上述因素构建的预测模型具有良好预测效能。
[Key word]
[Abstract]
Objective To construct a predictive model for the quality of life in patients with primary open-angle glaucoma(POAG)accompanied by visual field defects based on machine learning algorithms.Methods A total of 526 POAG patients with visual field defects who were treated in the hospital from May 2020 to May 2025 were retrospectively selected.The National Institute of Ophthalmology Visual Function Questionnaire-25(NEI VFQ-25)was used to evaluate the quality of life of patients.According to this,they were divided into normal group and abnormal group.The general data of the two groups were collected.The variables were screened by the least absolute shrinkage and selection operator(LASSO)regression and Logistic regression analysis,and the factors with statistical significance(P<0.05)were included to construct the prediction model.The area under the curve(AUC),calibration curve and decision curve of the receiver operating characteristic curve(ROC)were used to evaluate the discrimination,calibration and clinical practicability of the model.Results The investigation revealed that among the 526 POAG patients with visual field defects,72 patients had an NEI VFQ-25 score<50,and were included in the abnormal group(accounting for 13.69%);the remaining 454 patients had an NEI VFQ-25 score≥50,and were included in the normal group(accounting for 86.31%).The proportions of comorbid migraine,advanced visual field defects,and high myopia,as well as the scores on the Self-Rating Anxiety Scale(SAS),the Adult Version of the Mishel Uncertainty in Illness Scale(MUIS-A),and the Pittsburgh Sleep Quality Index(PSQI)in the abnormal group were all higher than those in the normal group.Five indicators were selected via LASSO regression:comorbid migraine,visual field defect,SAS score,MUIS-A score,and PSQI score.Further Logistic regression analysis showed that comorbid migraine(OR=5.215),SAS score(OR=1.174),MUIS-A score(OR=1.124),and PSQI score(OR=1.352)were risk factors for the quality of life in POAG patients with visual field defects(P<0.05),while early visual field defect(OR=0.121)was a protective factor(P<0.05).A risk prediction model was established based on the above factors.The ROC curve showed that the AUC of this model for predicting the quality of life in POAG patients with visual field defects was 0.927(95%CI:0.895-0.960).The model was validated using the Bootstrap method with 1000 repetitions,yielding a validation AUC of 0.916,indicating good discrimination of the nomogram model.The calibration curve plotted was close to the reference curve,with a Cox-Snell R²of 0.587 and a Brier Score of 0.057.The Hosmer-Lemeshow goodness-of-fit test showed a consistent between the model''s fitted values and the observed values(χ²=9.770,P=0.282>0.05),indicating that the model passed the HL test and had good goodness-of-fit.The decision curve showed that the predictive model had a high net benefit rate within the threshold range,suggesting good clinical utility of the model in predicting quality of life.Conclusion The quality of life in POAG patients with visual field defects is influenced by comorbid migraine,visual field defect status,SAS score,MUIS-A score,and PSQI score.The predictive model constructed based on these factors demonstrates good predictive performance.
[中图分类号]
[基金项目]
2023年度河北省医学科学研究课题计划(20231825)