Establishment And Validation Of Clinical Prediction Models For Predicting Renal Function Progression In Patients With Type 2 Diabetes Mellitus

Objective:Diabetes mellitus(DM)has become a major chronic disease that seriously affects human health.Type 2 diabetes mellitus(T2DM)is the most common type of DM,accounting for more than 90%.At present,T2DM has become the leading cause of end-stage renal disease(ESRD)in developed countries.In China,T2DM has become the leading cause of chronic kidney disease(CKD)and the second cause of ESRD.Risk assessment based on clinical prediction model can provide the prognosis information of patients,which is helpful to the risk stratification of patients and the rational allocation of medical resources.This study mainly focused on:(1)prevalence and clinical characteristics of kidney disease in Chinese adult T2DM patients;(2)the establishment and validation of clinical prediction models for incident CKD in Chinese adult T2DM inpatients;(3)the establishment and validation of clinical prediction models for the progression to ESRD in Chinese adult T2DM patients.The main purpose of this study is to develop clinical prediction models for predicting the progression of renal function in Chinese adult T2DM patients based on readily available clinical parameters,so as to guide the risk stratification of patients and initiate individualized prevention and treatment measures,hoping to reducing the burden of T2DM related renal complications.This study includes three parts:PartⅠ:Prevalence and clinical characteristics of kidney disease in Chinese adult inpatients with T2DMMethods:We retrospectively collected the clinical data of T2DM patients hospitalized from January 2014 to December 2018 from the Chinese Renal disease Data System(CRDS).A total of 345,387 adult T2DM inpatients from 17 tertiary hospitals were included in this study.The definitions of T2DM,ESRD and related complications and comorbidities were defined based on the International Classification of Diseases and Related Health Problems,the 10th Revision.The overall prevalence,annual prevalence and regional prevalence of ESRD in T2DM patients were calculated.The distribution characteristics of renal function and urinary protein in T2DM patients were investigated.The prevalence of ESRD in patients of different genders and ages was analyzed and compared.The characteristics of the distribution of inpatient departments,length of stay,total cost of hospitalization,complications and combidities of T2DM patients with ESRD were also investigated.Results:Among 345,387 adult T2DM inpatients,17,448 were complicated with ESRD.The overall prevalence of ESRD was 5.05%.From 2014 to 2018,the prevalence of ESRD in T2DM patients increased over the years,from a prevalence of5.20%in 2014 to 6.61%in 2018.Among 260,957 T2DM inpatients who underwent a serum cratinine(Scr)examination,18.96%had renal function impairment,defined as an estimated glomerular filtration rate(e GFR)<60 ml/min/1.73 m~2).Among 250,569T2DM inpatients who underwent a routine urine test,19.90%had proteinuria(defined as urine protein≥±).Among 345,387 T2DM inpatients,the prevalence of ESRD in the age group of 70-79 is the highest,which is 5.30%,and the prevalence of ESRD in the age group of 18-29 is the lowest,which is 3.08%.Among the 17,448 T2DM patients with ESRD,patients aged 60 to 69 years old accounted for the highest proportion(31.33%),and the average age of male patients was lower than that of female patients(t=15.308,P<0.001).Compared with T2DM patients without ESRD,patients with ESRD had longer hospitalization stay(t=-12.758,P<0.001),higher total hospitalization costs(t=-5.3329,P<0.001),higher proportion of diabetic retinopathy(DR)(X~2=1163.4,P<0.001),diabetic peripheral vascular disease(X~2=15.589,P<0.001),hypertension(X~2=2611.8,P<0.001)and autoimmune diseases(X~2=5.4022,P<0.020).Conclusion:From 2014 to 2018,there was a high prevalence of renal impairment or positive urinary protein in Chinese adult patients with T2DM.The number of T2DM patients with ESRD increased over the years,and the prevalence of ESRD showed an increased trend.Patients with ESRD have longer hospitalization stay,higher hospitalization cost,and a larger proportion of DR,diabetes peripheral vascular disease,hypertension and autoimmune diseases.Prevention and treatment of ESRD are of utmost relevance in T2DM patients.PartⅡ:Establishment and validation of clinical prediction models for incident CKD in Chinese adult patients with T2DMMethods:In a multi-center retrospective cohort study,we collected the baseline and follow-up clinical data of T2DM patients hospitalized from January 2013 to December 2020 from CRDS.A total of 42,683 T2DM patients aged from 18 to 80years old were included in our study,which were then divided into the derivation cohort(27,460 cases)and the external validation cohort(15,223 cases)according to different medical centers.The outcome event was defined as incident of e GFR less than 60 m L/min/1.73 m~2.Gender-specific clinical prediction models were established by univariate and multivariate Cox proportional hazards regression.The discrimination of these models is evaluated by the area under the curve(AUC)of the receiver operating characteristic(ROC)curve,and calibration is evaluated by the calibration curve.Results:The average age of T2DM patients in the derivation cohort was 61±11 years old,with a median follow-up time of 16(8,31)months,of which 2,792 patients(10.17%)developed CKD.Ten predictors were selected to establish the final model after multivariate Cox regression:age,sex,hemoglobin(HGB),plasma albumin(ALB),Scr,serum uric acid,serum cystatin C(Cys C),fibrinogen,low density lipoprotein cholesterol(LDL-C),the usage of insulin use and the usage of renin-angiotensin-aldosterone system(RAAS)inhibitors.Gender-specific CKD predicition model were further developed using cox regression analysis.The common predictors of male and female included age,HGB,plasma ALB,Scr,blood uric acid and fibrinogen.The male specific predictor was serum Cys C,and the female specific predictor was glycosylated hemoglobin and the usage of RAAS inhibitors.The AUC of these models in the derivation cohort was between 0.728 and 0.748,and the AUC in the validation cohort was between 0.719 and 0.742.The calibration curves showed good calibration both in the derivation cohort and the validation cohort.In addition,the prediction model was further presented as nomogram,which could be more convenient for application in clinical practice.Conclusion:Based on readily available clinical parameters,we established and external validated gender-specific clinical prediction models for incident CKD in Chinese T2DM patients.It is suitable for T2DM patients aged from 18 to 80 years old without renal complications.These models achieved satisfactory discrimination and calibration in the derivation cohort and the validation cohort,which could be applied to early identification of high-risk groups of incident CKD.PartⅢ:Establishment and validation of clinical prediction models for the progression to ESRD in Chinese adult patients with T2DMMethods:In a multi-center retrospective cohort study,we collected the baseline and follow-up clinical data of T2DM patients hospitalized from January 2013 to December 2020 from CRDS.A total of 81,569 T2DM patients aged from 18 to 80years old were included in our study,which were then divided into the derivation cohort(55,824 cases)and the external validation cohort(25,745 cases)according to different medical centers.The outcome event was defined as the occurrence of ESRD,including e GFR<15 m L/min/1.73 m~2or the initiation of renal replacement therapy(maintenance hemodialysis,peritoneal dialysis or renal transplantation).Clinical prediction models were established by univariate and multivariate Cox proportional hazards regression and were presented in the form of risk score,nomogram and web calculator.The discrimination of these models is evaluated by the AUC of the ROC curve,and calibration is evaluated by the calibration curve.Results:The average age of T2DM patients in the derivation cohort was 62±11 years old,with a median follow-up time of 384(123,900)days,of which 1,527 patients(2.74%)developed ESRD.Ten predictors were selected to establish the final model after multivariate Cox regression:age,HGB,plasma ALB,serum uric acid,LDL-C,fibrinogen,Scr,urinary protein levels,hypertension and diabetes retinopathy.Three clinical prediction models were established,including the full model,which included all the above ten predictors;laboratory model,only including laboratory examination indicators(HGB,plasma ALB,serum uric acid,LDL-C,fibrinogen,Scr and urinary protein levels);simplified model,only including five readily accessible indicators(age,HGB,Scr,urinary protein levels and hypertension).The AUC of these models in the derivation cohort was between 0.916 and 0.926,and the AUC in the validation cohort was between 0.868 and 0.882.The calibration curves showed good calibration both in the derivation cohort and the validation cohort.Combining the minimum P-value method and restricted cubic spline,the continuous variables in the full model were transformed into categorical variables,and the risk score model with a highest total score of 24 is established.Based on the risk score model,T2DM patients were divided into four different risk groups,including the low-risk group(total score<8),moderate-risk(8≤total score<15),high-risk(15≤total score<20)and very high-risk group(20≤total score<24).In the derivation cohort,compared with patients in the low-risk group,the hazard ratios of outcome events in the moderate-risk,high-risk and very high-risk group were 18.64(95%CI:15.97,21.76),80.62(95%CI:69.91,94.33)and 178.16(95%CI:148.66,213.52)(P<0.001),respectively.In addition,the simplified model was further presented in the form of a nomogram and a web calculator,which could be more convenient for application in clinical practice.Conclusion:Based on common clinical indicators,we established and external validated clinical prediction models for predicting the occurrence of ESRD in Chinese T2DM patients,which is suitable for T2DM patients aged from 18 to 80 years old with or without renal complications.These prediction model were further presented in the form of risk score,nomogram and web calculator,which could serve as helpful bedside tools to identify high-risk groups of ESRD and guide clinical decision making,so as to delay or even reduce the occurrence and development of ESRD.