摘要
目的探讨代谢组学分析构建风险预测模型对预测2型糖尿病肾病(type 2 diabetic nephropathy,T2DN)患者预后的应用价值。方法选取2014年9月~2017年9月于湘南学院附属医院接受治疗的282例T2DN患者,按照预后情况将患者分为预后良好组(n=199)和预后不良组(n=83)。采用超高效液相色谱系统分析测试样品,并运用Markieview软件进行判别分析,寻找出特征差异性代谢物。采用向前逐步法明确最终纳入预测模型的变量,根据各变量对应的偏回归系数(β)构建方程,建立T2DN患者预后预测模型。利用Hosmer-Lemeshow检验和受试者工作特征曲线(receiver operating characteristic curve,ROC)评价预测模型的拟合优度以及预测能力。结果与预后良好组比较,预后不良组患者血清空腹血糖(fasting blood sugar,FBG)(7.96±1.67mmol/L vs 5.03±0.69 mmol/L)、糖化血红蛋白(glycosylated hemoglobin,HbAlc)(8.01%±2.66%vs 5.96%±2.57%)、血尿素氮(blood urea nitrogen,BUN)(6.31±0.88 mol/L vs4.51±0.91 mol/L)、血尿酸(uric acid,UA)(411.25±79.61μmol/L vs 331.21±91.36μmol/L)、三酰甘油(triacylglycerol,TG)(2.18±0.33 mmol/L vs 1.75±0.63 mmol/L)水平升高,低密度脂蛋白-胆固醇(low density lipoprotein-cholesterol,LDL-C)(2.53±0.19 mmol/L vs 2.60±0.21 mmol/L)、高密度脂蛋白-胆固醇(high density lipoprotein-cholesterol,HDLC)(0.95±0.11 mmol/L vs 1.14±0.12 mmol/L)水平降低,差异均有统计学意义(t=10.998,6.042,9.644,4.796,4.101,1.676,7.916,均P<0.01)。代谢组学检验结果显示,预后良好组和预后不良组的代谢状态可明显区分,与预后良好组比较,预后不良组上调的代谢产物有溶血磷脂酰胆碱(t=1.362,P<0.01)、谷氨酰胺-精氨酸(t=2.302,P<0.01)、半乳糖羟赖氨酸(t=1.036,P<0.01)、鹅去氧胆酸(t=5.261,P<0.01);下调的代谢产物有溶血磷脂酰乙醇胺(t=2.321,P<0.01)、磷脂酰乙醇胺(t=5.261,P=0.001)、磷脂酰胆碱(t=2.528,P=0.001)、磷脂酰甘油(t=3.624,P<0.01)、鞘磷脂(t=2.591,P<0.01)、心磷脂(t=1.362,P<0.01)、二酰甘油(t=5.623,P<0.01)、三酰甘油(t=4.115,P<0.01)、苯丙胺酰-丙氨酸(t=2.361,P<0.01)、神经节苷脂(t=3.334,P<0.01)、透明质酸(t=2.924,P<0.01)和磷酸二氢丙酮(t=1.623,P=0.001),差异均有统计学意义(均P<0.05)。根据潜在标志物测定结果建立的预测模型如下:Logit(P)=-5.319+0.172(溶血磷脂酰乙醇胺)+0.669(磷脂酰乙醇胺)+0.624(溶血磷脂酰胆碱)+1.149(磷脂酰胆碱)+0.841(磷脂酰甘油)+0.271(鞘磷脂)+0.744(心磷脂)+0.102(二酰甘油)+0.667(三酰甘油)+0.676(谷氨酰胺-精氨酸)+1.067(半乳糖羟赖氨酸)+0.802(苯丙胺酰-丙氨酸)+0.203(鹅去氧胆酸)+0.711(神经节苷脂)+0.034(透明质酸)+0.494(磷酸二氢丙酮)。T2DN患者预后的Kaplan-Meier生存分析中位生存时间预后不良组为19个月,预后良好组为25个月。采用ROC曲线分析模型AUC为0.853(95%CI:0.759~0.909,均P<0.001),敏感度、特异度和约登指数分别为85.26%,82.84%和0.681,模型预测T2DN患者预后的水平较高,校准曲线和标准曲线结果无明显偏倚,一致性良好。结论基于代谢组学分析构建风险预测模型预测T2DN患者预后具有一定的预测价值。
Objective To explore the application value of metabolomics analysis and constructing a risk prediction model in predicting the prognosis of patients with type 2 diabetic nephropathy(T2DN).Methods The 282 patients with type 2 diabetic nephropathy who were treated in the Affiliated Hospital of Xiangnan University from September 2014 to September 2017 were divided into good prognosis group(n=199)and poor prognosis group(n=83)according to their prognosis.Test samples were analyzed by uHPLC system and analyzed by Markieview software to find the characteristic differential metabolites.The forward stepwise method was used to clarify the variables finally included in the prediction model,and the equation was constructed according to the corresponding partial regression coefficient(β)of each variable to establish the prognosis prediction model of patients with type 2 diabetic nephropathy.The goodness-of-fit and prediction ability of the prediction model were evaluated by Hosmer-Lemeshow test and subject operating characteristic curve.Results Compared with the good-prognosis group,patients in the poor prognosis group had serum fasting blood glucose(FBG)(7.96±1.67mmol/L vs 5.03±0.69 mmol/L),glycosylated hemoglobin(HbAlc)(8.01%±2.66%vs 5.96%±2.57%),blood urea nitrogen(BUN)(6.31±0.88 mol/L vs 4.51±0.91 mol/L),Blood uric acid(UA)(411.25±79.61μmol/L vs 331.21±91.36μmol/L),Triacylglycerol(2.18±0.33 mmol/L vs 1.75±0.63 mmol/L)increased,low density lipoprotein-cholesterol(LDL-C)(2.53±0.19 mmol/L vs 2.60±0.21 mmol/L),high density lipoprotein-cholesterol(HDL-C)(0.95±0.11 mmol/L vs 1.14±0.12 mmol/L)decreased,and the differences were statistically significant(t=10.998,6.042,9.644,4.796,4.101,1.676,7.916,all P<0.01).The results of the metabolomics test showed that the metabolic status was clearly distinguished between the good and poor prognosis groups.In comparison with the well-prognosis group,the upregulated metabolites in the poor prognosis group were lysophosphatidylcholine(t=1.362,P=0.000),glutamine-arginine(t=2.302,P=0.000),galactose hydroxylysine(t=1.036,P=0.000)and goose deoxycholic acid(t=5.261,P=0.006).The downregulated metabolites were lysophosphatidylethanolamine(t=2.321,P=0.000),phosphatidylethanolamine(t=5.261,P=0.001),phosphatidylcholine(t=2.528,P=0.001),phosphatidyl glycerol(t=3.624,P=0.000),sphingingipid(t=2.591,P=0.000),cardiolipin(t=1.362,P=0.000),glycerol difat(t=5.623,P=0.000),triglycerides(t=4.115,P=0.000),phenylalanide-alanine(t=2.361,P=0.000),ganglioside(t=3.334,P=0.000),hyaluronic acid(t=2.924,P=0.000)and dihydroacetone phosphate(t=1.623,P=0.001),and the differences were statistically significant(all P<0.05).Based on the determination of potential markers,Logit(P)=-5.319+0.172(lysophosphatidylethanolamine)+0.669(phosphatidylet hanolamine)+0.624(lyso-phosphatidylcholine)+1.149(phosphatidylcholine)+0.841(phosphatidylglycerol)+0.27(sphingomyelin)+0.744(cardiolipin)+0.102(glycer)+0.667(triglyceride)+0.676(glutamine-arginine)+1.067(galactose hydroxylysine)+0.802(phenylalanide-alanine)+0.203(goose deoxycholic acid)+0.711(ganglioside)+0.034(hyaluronic acid)+0.494(dihydroacetone phosphate).The median survival time of Kaplan-Meier survival analysis of patients with T2DN was 19 months in the group with poor survival time and 25 months in the group with good prognosis.The ROC curve analysis model AUC was 0.853(95%CI:0.759~0.909,P<0.001),sensitivity,specificity and yoden index were 85.26%,82.84%and 0.681,respectively.The model predicted high levels of prognosis of T2DN patients,and the calibration curve and standard curve results showed no obvious bias,and good consistency.Conclusion Building a risk prediction model based on metabolomic analysis to predict the prognosis of T2DN patients has certain predictive value.
作者
陈筱涛
刘韵
龙碧莹
郭皖北
CHEN Xiao-tao;LIU Yun;LONG Bi-ying;GUO Wan-bei(Department of Endocrinology,Affiliated Hospital of Xiangnan University,Hunan Chenzhou 423000,China)
出处
《现代检验医学杂志》
CAS
2023年第3期97-102,164,共7页
Journal of Modern Laboratory Medicine
基金
湖南省教育厅科学研究项目(项目编号:20A466):糖尿病肾病发病的相关危险因素及代谢组学分析研究。
关键词
代谢组学
2型糖尿病肾病
预测模型
metabolomics
type 2 diabetic nephropathy
predictive model
