不同肾功能的糖尿病合并慢性肾脏病患者血糖管理指标估算模型的构建与验证

An estimation model for glucose management indicator in diabetes mellitus with chronic kidney disease stratified by renal function

目的构建并验证适合不同肾功能的糖尿病合并慢性肾脏病(CKD)患者的血糖管理指标(GMI)估算模型。方法本研究为回顾性研究。选择2018年8月至2022年10月于南方医科大学南方医院内分泌科住院期间佩戴持续葡萄糖监测设备的163例糖尿病合并CKD患者作为研究对象。收集糖化血红蛋白(HbA1c)、血白蛋白(ALB)、血红蛋白(Hb)、GMI等资料。根据估算的肾小球滤过率(eGFR)水平,将患者分为G1~3a期[eGFR≥45 ml·min^(-1)·(1.73 m^(2))^(-1),101例]和G3b~5期[eGFR<45 ml·min^(-1)·(1.73 m^(2))^(-1),62例]两组。采用Pearson相关性分析法分析HbA1c和GMI的相关性。在G1~3a期和G3b~5期两组患者中,通过简单随机抽样,将研究对象按6∶4的比例划分为训练集和验证集,利用患者的实验室生化指标数据,分别构建与验证G1~3a期和G3b~5期两组患者的GMI估算模型。在训练集中,将各生化指标进行单因素线性回归分析,将具有临床意义和统计学意义(P<0.05)的变量纳入多因素逐步回归分析,作为GMI估算模型的自变量。在验证集中,使用构建的模型估算GMI,用配对t检验及Bland-Altman分析评估估算GMI和实测GMI的一致性。结果G1~3a期糖尿病合并CKD患者HbA1c和GMI之间呈正相关(r=0.547,P<0.001),而G3b~5期患者HbA1c和GMI之间不存在相关性(r=0.197,P=0.125)。在G1~3a期患者中构建的GMI估算模型为:GMI(%)=0.245×HbA1c(%)-0.043×ALB(g/L)+6.380;验证集GMI估算值为7.04%±0.76%,实测值为6.81%±0.81%,两者差异无统计学意义(P=0.059)。在G3b~5期患者中构建的GMI估算模型为:GMI(%)=0.108×HbA1c(%)-0.03×ALB(g/L)+贫血(不合并贫血,+0;合并贫血,-0.840)+7.549;验证集GMI估算值为6.55%±0.32%,实测值为6.82%±0.89%,两者差异无统计学意义(P=0.165);Bland-Altman分析结果提示,在G1~3a和G3b~5期患者中,分别有8.33%(3/36)和4.35%(1/23)的点在95%一致性界限之外。结论HbA1c评估G3b~5期糖尿病合并慢性肾脏病患者血糖控制情况的准确性较弱。使用适合不同肾功能的糖尿病合并慢性肾脏病患者GMI的估算模型,可较为合理地估算GMI值。

Objective To establish and validate a model for estimating the glucose management indicator(GMI)in diabetes mellitus with chronic kidney disease(CKD)stratified by estimated glomerular filtration rate(eGFR)level.Methods A total of 163 diabetes mellitus with CKD who wore continuous glucose monitoring during hospitalization in the Department of Endocrinology,Nanfang Hospital of Southern Medical University from August 2018 to October 2022 were retrospectively enrolled and divided into G1-3a[eGFR≥45 ml·min^(-1)·(1.73 m^(2))^(-1),101 cases]and G3b-5[eGFR<45 ml·min^(-1)·(1.73 m^(2))^(-1),62 cases]groups.General,biochemical and CGM data were collected.Pearson correlation analysis was used to investigate the correlation between glycated hemoglobin A1c(HbA1c)and GMI.Patients in G1-3a and G3b-5 groups were divided into training set and validation sets by simple random sampling with a ratio of 6∶4.Based on biochemical data,models to calculate GMI were constructed in G1-3a and G3b-5 groups.In the training set,univariate linear regression analysis was performed for each clinical variable,and univariate significant variables(P<0.05)were selected for multivariate stepwise regression analysis.For model validation,the estimated GMI was compared with real GMI using Pair′s t test and Bland-Altman analysis.Results GMI correlated positively with HbA1c in G1-3a group(r=0.547,P<0.001)but not in G3b-5 group(r=0.197,P=0.125).Based on the clinical data,a linear regression model was developed to estimate the GMI in G1-3a group:GMI(%)=0.245×HbA1c(%)-0.043×ALB(g/L)+6.380.In the validation set,the estimated GMI was 7.04%±0.76%and real GMI was 6.81%±0.81%,no significant difference was observed(P=0.059).In G3b-5 group,the model was:GMI(%)=0.108×HbA1c(%)-0.03×ALB(g/L)+anemia(non-anemia,+0;anemia,-0.840)+7.549.In the validation set,the estimated GMI was 6.55%±0.32%and real GMI was 6.82%±0.89%,with no significant difference was observed(P=0.165).Bland-Altman analysis showed that there was 8.33%(3/36)and 4.35%(1/23)pair of estimated GMI and real GMI falling outside the 95%agreement line in the G1-3a and G3b-5 group respectively.Conclusions The accuracy of HbA1c to assess glycemic status in diabetes mellitus with stage G3b-5 CKD decreased.The developed linear model can estimate the GMI level for diabetes mellitus with CKD with reasonable accuracy.