ASCVD患者eGDR与长期死亡风险的关系:一项回顾性队列研究与机器学习预测模型构建OA
Association between eGDR and long-term mortality risk in elderly patients with ASCVD:a retrospective cohort study and construction of machine-learning predictive models
目的 探讨估算葡萄糖处置率(eGDR)与老年动脉粥样硬化性心血管疾病(ASCVD)患者全因死亡及心血管死亡的关联,构建全因死亡风险预测模型并进行验证.方法 收集2008年1月-2022年12月解放军总医院第二医学中心≥60岁、确诊ASCVD的患者2479例行回顾性分析.根据eGDR四分位数将所有患者分为4组:Q1 组(n=617)、Q2 组(n=620)、Q3组(n=622)、Q4组(n=620),以评估eGDR与死亡风险的剂量-反应关系.采用Cox比例风险模型和限制性立方样条(RCS)分析eGDR与死亡风险的关联.比较7种机器学习算法[K最近邻、支持向量机、随机森林(RF)、最小绝对收缩和选择算子回归(LASSO)回归、朴素贝叶斯、极限梯度提升和决策树]对全因死亡的预测效能.最优模型采用夏普利加性解释(SHAP)对变量重要性进行排序.结果 在中位随访8.1年期间,发生全因死亡1101例(44.4%)和心血管死亡429例(17.3%).与eGDR最低四分位数(Q1)患者相比,最高四分位数(Q4)患者的全因死亡风险降低28%(HR=0.72,95%CI 0.57～0.90),心血管死亡风险降低42%(HR=0.58,95%CI 0.44～0.78).RCS分析显示,eGDR与全因和心血管死亡呈明显负相关(P<0.05).7种算法模型中,RF的曲线下面积(AUC)最高(AUC=0.748,95%CI 0.740～0.756).SHAP分析显示,年龄、低密度脂蛋白胆固醇、肾小球滤过率、eGDR是预测全因死亡的4个最重要特征.结论 eGDR降低与≥60岁ASCVD患者全因死亡及心血管死亡风险增加独立相关.纳入eGDR的RF模型对全因死亡具有较好的预测能力,eGDR可作为该人群死亡风险分层的潜在的生物标志物.
Objective To investigate the association between estimated glucose disposal rate(eGDR)and all-cause and cardiovascular mortality in elderly patients with atherosclerotic cardiovascular disease(ASCVD),and to develop and validate a predictive model for all-cause mortality risk.Methods A total of 2479 patients aged≥60 years with established ASCVD treated at the Second Medical Center of Chinese PLA General Hospital from January 2008 to December 2022 were included in this retrospective analysis.Patients were divided into four groups according to the eGDR quartiles:Q1 group(n=617),Q2 group(n=620),Q3 group(n=622),and Q4 group(n=620),to evaluate the dose-response relationship between eGDR and mortality risk.Cox proportional hazards models and restricted cubic spline(RCS)analyses were employed to assess the association between eGDR and mortality risk.Seven machine learning algorithms(K-nearest neighbours,support vector machines,random forests,least absolute shrinkage and selection operator(LASSO)regression,naive Bayes,extreme gradient boosting,and decision trees)were compared in terms of their predictive performances for all-cause mortality.The optimal model was further interpreted using SHapley Additive exPlanations(SHAP)to rank variable importance.Results During a median follow-up of 8.1 years,1101 all-cause deaths(44.4%)and 429 cardiovascular deaths(17.3%)occurred.Compared with patients in the lowest eGDR quartile,those in the highest quartile had a 28%reduction in all-cause mortality risk(HR=0.72,95%CI 0.57-0.90)and a 42%reduction in cardiovascular mortality risk(HR=0.58,95%CI 0.44-0.78).RCS analysis demonstrated a significant negative correlation between eGDR and both all-cause and cardiovascular mortality(P<0.05).Among the seven algorithms,random forest achieved the highest area under the curve(AUC=0.748,95%CI 0.740-0.756).SHAP analysis identified age,low-density lipoprotein cholesterol(LDL-C),estimated glomerular filtration rate(eGFR),and eGDR as the four most important predictors of all-cause mortality.Conclusions Lower eGDR levels are independently associated with increased risks of both all-cause and cardiovascular mortality in ASCVD patients aged≥60 years.The random forest model incorporating eGDR achieves favorable predictive efficacy for all-cause mortality,supporting eGDR as a promising biomarker for mortality risk stratification in this population.
安莉;张帷;付治卿;吕月;李珊
解放军总医院第二医学中心呼吸与危重症医学科/国家老年疾病临床研究中心,北京 100853解放军总医院第二医学中心心血管内科/国家老年疾病临床研究中心,北京 100853解放军总医院第二医学中心心血管内科/国家老年疾病临床研究中心,北京 100853解放军总医院第二医学中心保健三科/国家老年疾病临床研究中心,北京 100853解放军总医院第二医学中心保健三科/国家老年疾病临床研究中心,北京 100853
医药卫生
胰岛素抵抗估算葡萄糖处置率动脉粥样硬化性心血管疾病死亡机器学习风险预测
insulin resistanceestimated glucose disposal rateatherosclerotic cardiovascular diseasemortalitymachine learningrisk prediction
《解放军医学杂志》 2026 (7)
1022-1030,9
This work was supported by the Open Project of National Clinical Research Centre for Geriatric Diseases(NCRCG-PLAGH-2024016) 国家老年疾病临床医学研究中心开放课题(NCRCG-PLAGH-2024016)
评论