摘要
目的 建立评估慢性肾脏疾病(chronic kidney disease,CKD)患者疾病进展的预测模型.方法 数据来自日本东京医科齿科大学附属医院的一项前瞻性、观察性队列研究,数据收集了2010年10月至2011年11月来自日本东京地区日本东京医科齿科大学附属16家医院共1 138例CKD 2～5期患者,根据纳入与排除标准,共纳入975例CKD患者进行分析.将符合标准的患者分为无进展组、进展组及未达到感兴趣事件终点发生死亡的死亡组,并按7∶3随机分为模型训练组和内部验证组.分别采用单因素及多因素Fine-Gray竞争风险模型及Cox回归分析CKD进展的危险因素,并绘制独立危险因素的累计CKD疾病进展特异性死亡率和累积竞争风险事件发生率的生存曲线,采用R语言软件构建Fine-Gray竞争风险模型列线图预测模型,并采用受试者工作特征(receiver operating characteristic,ROC)曲线及校准曲线对模型的区分度和校准度进行验证.结果 975例CKD患者中,男684例,占71.2%,平均年龄(67.5±13.5)岁.CKD无进展组658例,进展组254例,死亡组63例.单因素Cox回归及单因素Fine-Gray 竞争风险分析结果显示,年龄小、贫血、低蛋白血症、肌酐水平高、CKD分期高、蛋白尿、尿潜血、高血压、糖尿病、使用药物等是CKD患者疾病进展的危险因素(P<0.05).多因素Fine-Gray竞争风险模型显示,年龄小、男性、贫血、CKD分期高、蛋白尿、糖尿病6个指标是CKD患者疾病进展的独立危险因素(P<0.05),其中CKD 5期(SHR=9.6,95%CI 5.7～16.1)和蛋白尿(SHR=8.8,95%CI 3.8～20.5)对CKD进展影响较大.ROC曲线结果显示,模型训练组1年及3年无进展C-index分别为0.895(95%CI 0.853～0.937)和0.890(95%CI 0.856～0.924),1年及3年无进展校准曲线显示良好的校准度,在内部验证组也有较高的区分度及校准度,总体模型性能良好.结论 年龄小、男性、贫血、CKD分期、蛋白尿、糖尿病在CKD患者中是较为特异性的独立危险因素,而基于此建立的列线图可较好地预测1年和3年疾病进展的风险,可为临床提供较好的预测工具.
Abstract
Objective To establish a predictive model to evaluate disease progression in patients with chronic kidney disease(CKD).Methods The data were obtained from a prospective,observational cohort study The Chronic Kidney Disease Research of Outcomes in Treatment and Epidemiology conducted at the Tokyo Medical and Dental University Hospital(Tokyo,Japan).Data were collected from 1 138 CKD stage 2-5 patients at 16 hospitals in the Tokyo area between October 2010 and November 2011.After applying inclusion and exclu-sion criteria,975 CKD patients were analyzed and categorized into non-progression,progression,and death groups,in which the event of interest was not reached due to death.The patients were randomly divided into model training and internal validation groups in a 7∶3 ratio.Single and multiple factor Fine-Gray compet-ing risk models and Cox regression were used to analyze the risk factors for CKD progression.Cumulative CKD-specific mortality and cumulative competing risk event occurrence rates for independent risk factors were plotted as survival curves.The Fine-Gray competing risk model line chart prediction model was con-structed using R software,and the model discrimination and calibration were validated using receiver operat-ing characteristic(ROC)and calibration curves.Results Among the 975 CKD patients,684(71.2%)were male,with a mean age of(67.5±13.5)years.There were 658,254,and 63 patients in the non-pro-gression,progression,and death groups,respectively,who did not reach the event endpoint due to competing risks.The results of single-factor Cox regression and single-factor Fine-Gray competing risk analyses showed that younger age,anemia,hypoalbuminemia,higher creatinine levels,CKD stage,proteinuria,hematuria,hypertension,diabetes,and medication use were risk factors for CKD progression(P<0.05).The multi-factor Fine-Gray competing risk model showed that younger age,male,anemia,CKD stage,proteinuria,and diabe-tes were independent risk factors for CKD progression(P<0.05),with CKD stage 5(SHR=9.6,95%CI:5.7-16.1)and proteinuria(SHR=8.8,95%CI:3.8-20.5)having the greatest impact on CKD progression.The ROC curve showed that the C-index of the model training group for no progression at 1 and 3 years were 0.895(95%CI:0.853-0.937)and 0.890(95%CI:0.856-0.924),respectively,with good calibra-tion for no progression at 1 and 3 years.The internal validation group also showed good discrimination and calibration,indicating good overall model performance.Conclusions Young age,male,anemia,CKD stage,proteinuria,and diabetes are shown to be independent risk factors with relatively high specificity for disease progression in patients with CKD.A column chart based on these factors predicts the risk of disease progression at 1 and 3 years,providing a useful predictive tool for clinical practice.
维普
万方数据