Establishment And Validation Of Nomogram Prediction Model For Cardiovascular Events Based On Surface Electrocardiogram

Objectives:Atherosclerotic cardiovascular disease(ASCVD)is the main disease burden and cause of mortality in China and the world.The accurate assessment of individual risk is of great value to guide and promote the prevention of ASCVD.Currently,most commonly used disease prediction models are established based on Western population.Recently,an ASCVD prediction model is also developed in China(China-PAR model),but its applicability in the population has not been fully evaluated,especially the high-risk population of cardiovascular disease in rural areas in Northeast China.Previous studies in Western general populations have indicated that certain electrocardiogram(ECG)indicators be used to assess the risk of future cardiovascular events independently of traditional risk factors,and have the potential to become emerging risk factors.Considering racial and regional differences,the prospective relationship between common ECG indicators and adverse outcomes in Asian population,especially Chinese population,has not been fully explored.In addition,there is insufficient evidence that a new model established by adding common ECG indicators to conventional risk factors can better predict ASCVD.This study is based on the follow-up data from the general population in the northeastern rural areas,and aims to investigate the China-PAR model performance in this region;evaluate the correlation between ECG indicators and adverse outcomes and develop and verify a new ASCVD prediction equation,combining traditional risk factors and common ECG indicators,suitable for the general population in the region.Methods:Part 1:The Northeast China Rural Cardiovascular Health Study(NCRCHS)was implemented from January to August 2013.Using a multistage,stratified,cluster randomized sampling strategy,11,956 participants aged≥35 years in rural areas of Liaoning Province were recruited.In 2015 to 2018,participants were invited to attend follow-up visits.The ECG indicators that previous researches focused on were selected for evaluation,including PR interval,QTc interval,QRS-T angle and ECG left ventricular hypertrophy(LVH).The prevalence of these abnormal ECG indicators in the general population was identified,and the multivariate Cox proportional hazard model was used to assesses the independent relationship between these ECG pattern and adverse outcomes.Part 2:According to the four-year Kaplan-Meier ASCVD rate of NCRCHS cohort,the China-PAR model was recalibrated to obtain a four-year ASCVD prediction equation.The cumulative incidence and actual number of ASCVD events within 4 years were obtained through Kaplan-Meier analysis,and the number of events predicted within 4years was estimated through the corrected China-PAR equation.The C statistic was used to evaluate the discrimination of the corrected China-PAR equation,and the Hosmer-Lemeshow test was used to evaluate the calibration of the corrected prediction model.We also calculate the expected-observed ratio and draw a calibration chart to evaluate the predictive performance of the corrected equation.In the sensitivity analysis,the four-year Kaplan-Meier ASCVD rate of China-PAR derivation cohorts(the Inter ASIA and China MUCA 1998)was used to recalibrate the equation.Part 3:Based on the NCRCHS cohort,2/3 of the participants are randomly selected to form the derivation cohort,and the remaining 1/3 of the participants constituted the validation cohort.Multiple imputation is used to fill in missing variables.Cox proportional hazards models was used to evaluate the relationship between traditional risk and ASCVD.The variables with P value<0.05 in the Cox regression analysis or variables with clinical significance were selected into the model.Univariate,multivariate Cox regression and predictive incremental evaluation analyses were used to determine ECG indicators.The determined traditional risk factors and ECG indicators was drawn into a nomogram model based on their regression coefficients.The values of different variables are projected by vertical lines to correspond to the points at the top of the nomogram(total score is 100 points),and then the points corresponding to each variable are accumulated to obtain the total score.the corresponding ASCVD incidence is found on the prediction line at the bottom of the nomogram according to the total score.In order to reduce the overfitting of the nomogram model,The Bootstrap self-sampling method was used for internal verification(500 sampling times).The external verification of the model is performed on the validation cohort.The discrimination was evaluated by using C-index and receiver operating characteristic curve(ROC),and the calibration was displayed by using calibrationχ2 and charts,and decision analysis curve was used to evaluate the applicability of this equation.Results:Part 1:Of the 9,633 NCRCHS participants included in the analysis,the crude prevalence rates of QTc prolongation,ECG-LVH,planar QRS-T angle abnormality,and first-degree AVB were respectively 13.0%(95%CI,12.2%-13.7%),8.4%(95%CI,7.8%-9.0%),2.6%(95%CI,2.3%-2.9%),and 1.3%(95%CI,1.1%-1.5%).Multivariate Cox regression using CVP,QTc interval and QRS-T angle as continuous variables indicated the risk of CVD respectively increased 10%(HR,1.10;95%CI,1.04–1.16),18%(1.18;1.09–1.28)and 12%(1.12;1.04–1.20),and the risk of all-cause mortality respectively increased 13%(1.13;1.05–1.21),25%(1.25;1.14–1.36)and 19%(1.19;1.11–1.29)per SD increase after adjusting the traditional risk factors.According to the clinical definition,ECG indicators as binary variables were brought into multivariate Cox proportional hazard model.ECG-LVH,frontal QRS-T angle abnormality and QTc prolongation were still independent predictors of CVD and all-cause death.Even if the Echo(LVMI)was further adjusted in regression,the correlation between ECG-LVH and adverse outcomes still existed.Only when PR interval was used as a categorical variable(First-degree AVB),it was statistically significant with CVD(1.90;1.15-3.13).In the sensitivity analysis,regression analyses were independently conducted after exclusion of baseline CVD history,and the above conclusions remained unchanged.Part 2:During the total follow-up of 39736.46 person-years(average follow-up4.32±0.93 years),359 participants developed ASCVD.The China-PAR model calibrated by NCRCHS had acceptable discrimination,and the C statistic in both males and females were greater than 0.7[male:0.789,95%CI(0.762-0.816);female:0.771,95%CI(0.738-0.803)].However,the calibration of the corrected model was not ideal,and the four-year risk of ASCVD was overestimated in both men and women.The predicted number of gender-specific ASCVD within 4 years by corrected China-PAR model(male:233.04;female:231.06)was significantly higher than the actual number of events calculated by the Kaplan-Meier method(male:184.45;female:154.26).The corrected equation overestimated by 32.3%in men(calibrationχ~2=29.646,P<0.05)and by 49.8%in women(calibrationχ~2=27.176,P<0.05).The calibration charts showed that the predicted probability was inconsistent with the actual incidence rate.In the sensitivity analysis,the four-year Kaplan-Meier ASCVD rate of China-PAR derivation cohorts(the Inter ASIA and China MUCA 1998)was used to recalibrate the equation,and the above conclusions remained unchanged.Part 3:The baseline characteristics of participants are similar between derivation cohort(n=5,983)with validation cohort(n=3,096).The variables with P value<0.05 in the Cox regression analysis or variables with clinical significance were selected into the model,including age,sex,waist circumference,education status,smoking status,hypertension and diabetes.ECG indicators with statistical significance or marginal statistical significance for NRI and IDI,and after taking full consideration of collinearity,QTc interval,Sv1+Rv5/Rv6 and QRS-T angle as categorical variables were included in our ASCVD prediction equation.The nomogram model showed the score increased with age,from 0 to 100,and the score was 50 at the age of 65.Similarly,the score increased with waist circumference,from 0 to 35.81.Although gender is a common conventional ASCVD risk factor,in the current model,men only scored 2.66 points and smoking scored 3.97 points.The score of middle school and primary school or below were 7.63 and 12.92 respectively.The hypertension scored 11.46.The diabetes scored 6.28.The scores of QRS-T angle were respectively17.87(0-45 degrees),17.84(46-90 degrees)and 39.22(136-180 degrees)points.The score of voltage(Sv1+Rv5/Rv6)in the 1.81-2.20,2.21-2.60,2.61-3.0and>3.0 mv groups were 4.14,10.48,11.20 and 19.55 respectively.Besides,the score of QTc interval in the 400-424,425-449 and≥450 ms groups were respectively 10.68,15.91,and 17.89 points.As the total score of the model increasing,the risk of ASCVD increased in 1,2,3,and 4 years.The C-index was 0.795(95%CI,0.769-0.821)in derivation cohort and 0.767(0.732-0.802)in validation cohort.In derivation cohort,the1-year,2-year,3-year risk and 4-year cut-off value were respectively 0.009270561,0.02013694,0.02214208,and 0.02800091;The values of corresponding AUC were0.829,0.827,0.803 and 0.797,respectively.In the validation cohort,the 1-year,2-year,3-year,and 4-year risk cutoff values were 0.007401116,0.01560465,0.02331485,and0.0306784,respectively.The values of corresponding AUC were respectively 0.769,0.741,0.755,and 0.782.The calibration curves of the derivation cohort and the validation cohort showed that the model fairly predicted ASCVD risk at different follow-up times.The decision curve analysis indicated that when the 4-year ASCVD incidence was less than 13%,the model had excellent judgment in both derivation cohort and the validation cohort.The intervention of high-risk individuals selected by the prediction model can achieve better results.Conclusion:This study reported the prevalence of abnormal ECG indicators in the Chinese general population,and prospectively confirmed the independent correlation between common ECG indicators and adverse outcomes.It was the first time to confirm the applicability of the China-PAR Project in northeast rural areas.As a result,we combined ECG indicators and traditional risk factors to develop and validate the first prediction model of cardiovascular events based on surface ECG indicators in China,and it was also the first ASCVD equation in Northeast China.In addition,we implemented graphical visual display of the prediction model with nomograms,which provides evidence-based medical proof for the prevention of ASCVD.