基于大型健康管理队列的慢性肾脏病预测模型

发布时间：2018-02-01 22:18

  本文关键词： 慢性肾脏病 健康管理队列 Cox回归 预测模型 弗罗明翰评分　出处：《山东大学》2017年硕士论文　论文类型：学位论文

【摘要】：研究背景慢性肾脏病(chronic kidney disease,CKD)是指任何原因引起的肾脏结构和功能障碍超过3个月,包括肾小球滤过率(glomerular filtration rate,GFR)正常和不正常的病理损伤、血液或尿液成分异常及影像学检查异常,或不明原因GFR[mL/(minx1.73m2)]超过3个月。近年来,慢性肾脏病患病率持续上升,全球CKD平均患病率约为13.4%,已成为全球重要的公共卫生问题。我国成人CKD患病率约为10.8%,现患病例近1.2亿人,且随着我国人口快速老龄化的趋势,高血压、糖尿病等患病率的增高,未来CKD患病人数势必会继续上升。然而,由于CKD在早期常无临床症状,CKD患者早期知晓率低,而发展到后期则预后差,通常会并发多种严重疾病,例如高血压、糖尿病、心血管疾病等,部分CKD患者可最终进展为终末期肾病(end-stage renal disease,ESRD),需要进行复杂且昂贵的肾脏替代治疗,带来严重的疾病负担。因此,明确CKD的危险因素及其效应,通过建立数学模型将各种危险因素组合起来,综合评估个体CKD的发病风险和预测发生概率,可以及早识别高风险个体并采取相应的健康管理措施,对早期预防、延缓甚至避免疾病发生具有重要意义。研究目的1、以"山东多中心健康管理纵向观察队列"为依托,采用多因素Cox比例风险回归模型,分性别构建CKD风险评估模型,并对模型进行合理的验证;2、将复杂的数学模型转化为CKD评分系统,为健康管理提供可直接应用于实践的工具。资料与方法利用"山东多中心健康管理纵向观察队列",建立CKD随访队列,采用多因素Cox比例风险回归建立CKD预测模型并采用ROC曲线、AUC、灵敏度、特异度等指标对其预测效果进行合理的评价,采用十折交叉验证法验证其预测效果,最终,使用弗罗明翰评分法构建风险评分矩阵。研究结果1、随访过程中发现,男性CKD的发病密度为30.96/1000人年,高于女性的13.92/1000人年,差异有统计学意义。2、在CKD发病组和CKD未发病组两组之间,多数体检指标存在统计学差异。使用单因素Cox回归分析初步筛选,并结合临床专业知识,确定男性预测模型的预测因子包括:年龄、体质量指数、对数转换的γ-谷胺酰转肽酶、血肌酐、甘油三酯、总胆固醇、血红蛋白、白细胞计数、血清白蛋白、血清球蛋白、糖尿病、高血压、肾囊肿、CVD;女性包括:年龄、体质量指数、对数转换的7-谷胺酰转肽酶、血肌酐、甘油三酯、总胆固醇、白细胞计数、血清白蛋白、血清球蛋白、血红蛋白、糖尿病、高血压、肾囊肿、CVD、睡眠状况。3、将以上变量进行多因素Cox回归分析,最终构建的预测模型包括的预测因子如下:男性:年龄、体质量指数、对数转换的γ-谷胺酰转肽酶、血肌酐、甘油三酯、白细胞计数、血清白蛋白、糖尿病、高血压、CVD;女性:年龄、血肌酐、白细胞计数、甘油三酯、血清白蛋白、高血压。4、使用受试者工作曲线(receiver operating characteristic curve,ROC)下面积AUC(area under curve)、灵敏度、特异度等指标评价上述模型的预测效果,男性CKD预测模型1-4年预测效果的AUC分别为0.669(95%可信区间(confidence interval,CI)为 0.661-0.676)、0.698(95%CI:0.690-0.707),0.687(95%CI:0.676-0.698),0.630(95%CI:0.615-0.644),灵敏度分别为 60.1%、55.9%、59.1%、58.6%,特异度分别为64.6%、77.0%、69.4%、63.6%;女性CKD预测模型的1-4年预测效果的 AUC 分别为 0.742(95%CI:0.732-0.752),0.793(95%CI:0.782-0.803),0.702(95%CI:0.687-0.717),0.621(95%CI:0.601-0.640),灵敏度分别为 61.1%、63.6%、57.7%、56.5%,特异度分别为 88.2%、89.6%、78.0%、62.9%。5、使用十折交叉验证对模型预测效果及稳定性进行验证,结果显示,经十折交叉验证,男性CKD预测模型1-4年预测效果的AUC分别为0.659(95%CI:0.651-0.666),0.692(95%CI:0.684-0.701),0.683(95%CI:0.672-0.694),0.620(95%CI:0.605-0.634);女性CKD预测模型1-4年预测效果的AUC分别为0.730(95%CI:0.719-0.740),0.789(95%CI:0.778-0.800),0.697C 95%CI:0.682-0.712),0.613 C95%CI:0.593-0.632)，，6、将预测模型转换为弗罗明翰评分模型后,男性总分范围为-2分至29分,-2分所对应的1-4年发病风险分别为0.08%,0.20%,0.33%,0.48%,29分所对应的1-4年发病风险分别为4.56%,10.78%,17.12%,23.74%;女性得分范围为-3至20分,-3分所对应的1-4年发病风险分别为0.03%,0.08%,0.12%,0.22%,20分所对应的1-4年发病风险分别为3.13%,7.24%,11.43%,20.00%。研究结论1、CKD发病密度存在性别差异,男性高于女性;2、本研究基于健康体检人群分性别构建了慢性肾脏病预测模型,男性模型预测因子为:年龄、体质量指数、对数转换的γ-谷胺酰转肽酶、血肌酐、甘油三酯、白细胞计数、血清白蛋白、糖尿病、高血压、CVD,女性模型预测因子为:年龄、血肌酐、白细胞计数、甘油三酯、血清白蛋白、高血压;3、模型用于预测1、2、3年的发病风险效果较好,且具有稳健性;4、本研究将预测模型转换为弗罗明翰风险评分,用于人群健康管理实践。本研究探索了针对健康管理人群进行疾病风险评估和建立预测模型的方法,建立了 CKD预测模型,并应用弗罗明翰风险评分法将预测模型进行转换,便于成果转化和实际应用。但因受资料的限制,建模时未能包含所有与CKD相关的指标,且现有健康管理队列可能存在一定的选择性偏倚,随访时间也较短,因此模型的稳定性及预测能力尚有待继续观察和进一步研究的验证。
[Abstract]:On the background of chronic kidney disease (chronic kidney, disease, CKD) refers to the renal structure and dysfunction of any cause for more than 3 months, including glomerular filtration rate (glomerular, filtration rate, GFR) of normal and abnormal pathological damage, blood or urine and abnormal imaging abnormalities, GFR[mL/ or unexplained (minx1.73m2)] for more than 3 months. In recent years, the prevalence of chronic kidney disease continues to rise, the global average CKD prevalence rate is about 13.4%, has become an important public health problem in the world. The prevalence rate is about 10.8% of China's adult CKD, the case of nearly 120 million people, with China's rapid population aging trend. Increased prevalence of hypertension, diabetes and so on, the future CKD the number of patients will continue to rise. However, due to CKD in early CKD patients are usually asymptomatic, early low awareness, and the development of late prognosis, usually associated with A variety of serious diseases, such as hypertension, diabetes, cardiovascular disease, CKD patients may eventually progress to end-stage renal disease (end-stage renal, disease, ESRD), the need for complex and expensive renal replacement therapy, brings the serious burden of disease. Therefore, the risk factors of CKD and its effect, through the establishment of mathematical models of various risk the risk factors in combination, and comprehensive evaluation of individual CKD prediction probability, early identification of high risk individuals and take corresponding measures for health management, early prevention, delay or even avoid is important diseases. Objective: 1, to "Shandong health management center longitudinal observational cohort" based on the multi factors Cox proportional hazards regression model, CKD risk assessment model of gender construction, and make reasonable verification of the model; 2, the complex mathematical model into CKD score The system can be directly applied to practice, to provide tools for health management. Materials and methods using the "Shandong Center for health management, establish the CKD longitudinal study cohort follow-up cohort, regression to establish CKD prediction model and the ROC curve, using multivariate Cox proportional hazard AUC, sensitivity, specificity and other indexes were used to evaluate the prediction results using ten fold cross validation method to verify the prediction effect, finally, using Freund mingsh scoring method to construct risk rating matrix. The results of the 1, was found during the follow-up, the incidence density of male CKD was 30.96/1000 person years, higher than the female 13.92/1000 years, there was significant difference in the incidence of CKD group.2 and CKD onset group between the two groups, there were significant differences in most physical indicators. Using single factor Cox regression analysis screening, combined with clinical expertise to identify predictors of male package prediction model Including: age, body mass index, log transformed gamma glutamyl transpeptidase, serum creatinine, triglyceride, total cholesterol, hemoglobin, white blood cell count, serum albumin, serum globulin, diabetes, hypertension, renal cyst, CVD; women included: age, body mass index, the log transformed 7- Valley enzyme, creatinine, triglyceride, total cholesterol, white blood cell count, serum albumin, serum globulin, hemoglobin, diabetes, hypertension, renal cyst, CVD, sleep.3, the variable Cox regression analysis, the following factors including prediction prediction model is constructed finally: male age:, body mass index, log transformed gamma glutamyl transpeptidase, serum creatinine, triglyceride, leukocyte count, serum albumin, diabetes, hypertension, CVD; female: age, serum creatinine, white blood cell count, serum albumin, triglyceride, hypertension,.4, use The receiver operating curve (receiver operating characteristic curve ROC AUC (area) and area under the under curve), the sensitivity, specificity and other indexes to evaluate the prediction effect of the model, the male CKD prediction model prediction effect of 1-4 years of AUC were 0.669 (95% confidence interval (confidence, interval, CI), 0.698 (0.661-0.676) 95%CI:0.690-0.707), 0.687 (95%CI:0.676-0.698), 0.630 (95%CI:0.615-0.644), the sensitivity was 60.1%, 55.9%, 59.1%, 58.6%, the specificity was 64.6%, respectively, 77%, 69.4%, 63.6%; female CKD prediction model of the prediction effect of 1-4 years of AUC were 0.742 (95%CI:0.732-0.752), 0.793 (95%CI:0.782-0.803), 0.702 (95%CI:0.687-0.717). 0.621 (95%CI:0.601-0.640), the sensitivity was 61.1%, 63.6%, 57.7%, 56.5%, the specificity was 88.2%, 89.6%, 78%, 62.9%.5, using ten fold cross validation test of predictive effect and stability of the model C, results show that by ten fold cross validation, male CKD prediction model prediction effect of 1-4 years of AUC were 0.659 (95%CI:0.651-0.666), 0.692 (95%CI:0.684-0.701), 0.683 (95%CI:0.672-0.694), 0.620 (95%CI:0.605-0.634); female CKD prediction model for 1-4 years the results of prediction AUC were 0.730 (95%CI:0.719-0.740), 0.789 (95%CI:0.778-0.800 0.697C), 95%CI:0.682-0.712), 0.613 C95%CI:0.593-0.632, 6), the prediction model is converted to galantine mingsh scoring model, male scores range from -2 to 29, -2 branch of the corresponding risk 1-4 years were 0.08%, 0.20%, 0.33%, 0.48%, 29 branch of the corresponding risk 1-4 years were 4.56%, 10.78%, 17.12%, 23.74%; female scores ranged from -3 to 20, -3 branch of the corresponding risk 1-4 years were 0.03%, 0.08%, 0.12%, 0.22%, 20 branch of the corresponding risk 1-4 years were 3.13%, 7.24%, 11.43%, 20.00%. The conclusion of the study 1, there are gender differences in the incidence of CKD density was higher in male than in female; 2, based on the healthy population gender construct prediction model of chronic kidney disease, male model predictive factors: age, body mass index, log transformed gamma glutamyl transpeptidase, serum creatinine, triglyceride, white blood cell count, serum albumin, diabetes, hypertension, CVD, female model predictive factors: age, serum creatinine, white blood cell count, serum albumin, triglyceride, hypertension; 3, risk prediction model for the effect of 1,2,3 is better, and has robustness; 4, the prediction model is converted to galantine mingsh risk score for the practice of population health management. This study explores the health management of disease risk assessment and prediction model method, we establish CKD prediction model and application of Freund mingsh risk score method forecast 妯″瀷杩涜�岃浆鎹�,渚夸簬鎴愭灉杞�鍖栧拰瀹為檯搴旂�

本文编号：1482999