Development And Validation Of A Risk Prediction Model For The New Incident Chronic Kidney Disease In The Routine Physical Examination Cohort

ObjectivesChronic kidney disease(CKD)is progressively emerging as a substantial health care problem worldwide.Early recognition of individuals at high risk for developing CKD can help to alleviate the suffering of the affected individuals and reduce the socioeconomic burden.However,there is a paucity of reliable instruments to predict risk of CKD in the general population.The objective of this study was to develop and validate a visualized model for predicting the risk of CKD incidence within 4 years in the routine physical examination cohort.MethodsThe present study enrolled 6515 participants who underwent the annual physical examination at the Healthy Management Center of Zhejiang Provincial People’s Hospital between January 2017 and December 2021.A ratio of 7:3 was used to divide the aforementioned cohort into the training and internal validation cohorts.In the same period,another 3152 participants received the annual physical examination at the Healthy Management Center of Zhejiang Provincial Hospital of Traditional Chinese Medicine were included for external validation purposes.Demographic data,medical histories,annual physical examinations and laboratory investigations were carefully recorded for each participants.The primary outcome was new incident CKD during the period of 4 years following the study start date,defined as an estimated glomerular filtration rate(e GFR)falling below 60 ml/min/1.73m~2,computed by the2009 Chronic Kidney Disease Epidemiology Collaboration(CKD-EPI)equation.In the training cohort,univariate analysis were preformed to assess the potential predictors of CKD onset in a preliminary manner and statistically significant variables from univariate analysis were selected in a logistic regression model for multivariate analysis.In our final model,a backward stepwise procedure was used to reduce the number of variables using the Akaike’s information criterion(AIC).A conventional nomogram and a web-based dynamic nomogram were developed based on the results of the multivariate analysis.We used the area under the receiver operating characteristic(ROC)curve(AUC),calibration plot and decision curve analysis(DCA)to evaluate the model’s discrimination,calibration and clinical utility.In addition,the model is evaluated for clinical extensibility in an external validation cohort to ensure its practical application potential.Results1.A total of 9667 participants,comprising 4563 cases in the training dataset,1952 cases in the internal validation dataset and 3152 cases in the external validation dataset,were ultimately enrolled in the current research.Over the course of follow-up,207(2.14%)individuals developed CKD in the total population,of whom 201(97.10%)had CKD stage 3a,4(1.93%)CKD stage 3b,and 2(0.97%)CKD stage 5.2.Univariate analysis results showed that patients with CKD were more likely to be older,have higher proportions of history of hypertension and diabetes mellitus,have higher body mass index,systolic blood pressure,diastolic blood pressure,serum creatinine,uric acid,triglyceride and glycated haemoglobin A1c levels,and have lower serum albumin levels,compared with healthy controls(p<0.05).No statistically significant differences were found between the two groups in terms of sex,history of stroke,haemoglobin,high density lipoprotein cholesterol and low density lipoprotein cholesterol levels(p>0.05).3.According to the results of multivariate logistic regression analysis,age(odds ratio[OR]1.118;95%confidence interval[CI]1.090-1.145;p<0.001),history of diabetes mellitus(OR 3.683;95%CI 1.791-7.575;p<0.001),serum creatinine(OR1.068;95%CI 1.048-1.088;p<0.001)and triglyceride levels(OR 1.201;95%CI1.083-1.332;p=0.001<0.05)were independent predictors of incident CKD within 4years in the routine physical examination cohort.The final predictive model was determined by stepwise backward selection based on the AIC.A total of six predictors,including age,history of diabetes mellitus,systolic blood pressure,serum albumin,creatinine and triglyceride levels,were used to built the nomogram.4.The AUC of the generated model was 0.8806(95%CI 0.8472-0.9141)in the training dataset,0.8506(95%CI 0.7856-0.9156)in the internal validation dataset,and0.9183(95%CI 0.8698-0.9669)in the external validation dataset.The calibration plots and DCA showed satisfactory calibration and clinical practicability of this model.5.To better facilitate the use of the nomogram by clinicians,we also created a web-based dynamic nomogram which was freely available.Conclusion1.In this study,the overall 4-year incidence rate of CKD was 2.14%(207/9667)in the routine physical examination cohort,with a predominance of CKD stage 3a.2.Age,history of diabetes mellitus,serum creatinine and triglyceride levels were independent predictors of incident CKD within 4 years in the routine physical examination cohort,as uncovered by both univariate and multivariate logistic regression analyses.3.The prediction model was constructed based on age,history of diabetes mellitus,systolic blood pressure,serum albumin,creatinine and triglyceride levels to assess the risk of CKD incidence within 4 years in the routine physical examination cohort,with excellent prediction accuracy and good generalization ability.4.We developed a web-based dynamic nomogram for the risk prediction of developing CKD within 4 years in the routine physical examination cohort.