Performance of the recommended ESC/EASD cardiovascular risk stratication model and SCORE in comparison to NT-proBNP as a single biomarker for risk prediction in type 2 diabetes mellitus

Background. Recently, the European Society of Cardiology (ESC) and European Association for the Society of Diabetes (EASD) introduced a new cardiovascular disease (CVD) risk stratication model to aid further treatment decisions in individuals with diabetes. Our study aimed to investigate the prognostic performance of the ESC/EASD risk model and the Systematic COronary Risk Evaluation (SCORE) in comparison to NT-proBNP in an unselected cohort of type 2 diabetes mellitus (T2DM). Methods & Results. A total of 1690 T2DM patients with a 10-year follow up for fatal CVD and all-cause death and a 5-year follow up for CVD and all-cause hospitalizations were analyzed. According to ESC/EASD risk criteria 25 (1.5%) patients were classied as moderate, 252 (14.9%) high, 1125 (66.6%) very high risk and 288 (17.0%) were not classiable. Both, NT-proBNP and SCORE risk model were associated with 10-year CVD and all-cause death and 5-year CVD and all-cause hospitalizations while the ESC/EASD model was only associated with 10-year all-cause death and 5-year all-cause hospitalizations. NT-proBNP showed signicantly higher C-indices than the ESC/EASD and SCORE risk model for CVD death [0.80 vs 0.53 vs 0.64, p<0.001] and all-cause death [0.73 vs 0.52 vs 0.66, p<0.001]. The performance of SCORE improved in a subgroup without CVD aged 40-64 years compared to the unselected cohort, while NT-proBNP performance was robust across all groups. Conclusion. The new introduced ESC/EASD risk stratication model performed limited compared to SCORE and single NT-proBNP assessment for predicting 10-year CVD and all-cause fatal events in individuals with T2DM. continuous overall NRI of 0.3223 (SE 0.0557, p<0.001) for all-cause hospitalizations and of 0.6961 (0.0631, p<0.001) for CVD hospitalizations. Similar results were obtained for NT-proBNP and the SCORE model with a continuous overall NRI of 0.3255 (SE 0.0502, p<0.001) for all-cause hospitalizations and of 0.694 (0.0590, p<0.001) for CVD hospitalizations.


P assessment
or predicting 10-year CVD and all-cause fatal events in individuals with T2DM.

Introduction

Diabetes is associated with a substantially increased risk to develop cardiovascular disease (CVD); 1 however, as individuals with diabetes represent a highly heterogeneous population incremental CVD risk is not equally distributed among diabetic patients. 2Therefore, the development of indiv dualized CVD risk assessment tools is essential to warrant a personalized therapy approach.

Recently, the European Society of Cardiology (ESC) in collaboration with the European Association for the Study of Diabetes (EASD) published new guidelines for the prevention and management of CVD in patients with prediabetes and diabetes. 3For the rst time, the use of a CVD risk strati cation model is recommended to aid treatment decisions in individuals with diabetes.The ESC/EASD risk model strati es diabetic patients into three different risk categories based on the 10-year risk estimate for fatal CVD adapted from the 2016 European Guidelines on CVD prevention in clinical practice. 4To the best of our knowledge, the predictive performance of the new y introduced risk strati cation model has not been veri ed in individuals with diabetes.

The Systematic COronary Risk Evaluation (SCORE) equation is commonly used for estimating the 10-year risk of fatal CVD in the general population. 5In the original publication of the SCORE project it was suggested that SCORE based risk assessment could be used for a rough assessment of CVD risk in diabetic patients.Yet, the predictive performance of SCORE has not been tested in patients with long-standing diabetes speci cally. 6As a result, its application for risk estimation in individuals with diabetes cannot be recommended. 4][9][10][11][12][13][14][15][16][17][18][19][20][21][22][23][24] Nonetheless, the new guidelines do not recommend routine assessment of circulating biomarkers for CVD risk estimation in diabetic patients. 3his study aimed to perform a head-to-head comparison of the predictive performance of the ESC/EASD model and SCORE against NT-proBNP for risk assessment of 10-year CVD death and all-cause death i) in an unselected T2DM cohort, ii) in selected patients with T2DM with characteristics similar to the SCORE derivation cohort and iii) to investigate outcome-speci c performance of the different risk estimates.Additionally, the prognostic utility of the risk assessments for 5-year CVD and all-cause hospitalization was evaluated.


Methods


Study population

From December 2005 through January 2010 a total of 2186 patients with T2D

betes outpatient
linics were included in a prospective registry.Medical history including comorbidities, diabetes duration, medical therapy and assessment of risk factors was recorded at enrolment.Patients were followed up as clinically appropriate.All patients gave written informed consent.The study was approved by the local Ethics Committee of the Medical University of Vienna and complies with the principles of the Declaration of Helsinki.


Laboratory analysis

Blood samples were collected under fasting conditions and immediately sent to the lo

l laboratory.Estimat
d glomerular ltration rate (eGFR) was assessed by the Modi ed Diet for Renal Disease Study equation.NT-proBNP determination was performed directly using the cobas h 232 point-of-care analyzer by Roche (Roche Diagnostics, Basel, Switzerland) with a lower detection limit of 59 pg/ml.The median coe cients of variation were previously determined as described by Bertsch et al. 25 Urine albumin creatinine ratio was assessed quantitively in fresh spot urine samples according to the local laboratory standards.


Calculating risk estimates a. ESC/EASD risk strati cation model

Patients were categorized as moderate, high and very high risk based on the predicted 10-year risk estimates for CVD death <5%, 5-10 and >10%, respectively, according to the ESC/EASD cardiovascular risk categories as indicated in Additional le 1: Table S1. 3 Except for renal impairment no precise de nition on the rating of the respective risk factors used within the ESC/EASD strati cation model was given.Thus, we de ned age, obesity and proteinuria as being at risk >50years, ≥30kg/m 2 and with a urinary albumin/creatinine ratio >30mg/mmol, respectively.High blood pressure and dyslipidemia were de ned according to the criteria of the respective current European guidelines, 26,27 as documented in medical charts or on speci c therapy.Smoking status was assessed based on hospital charts and by self-report.CVD diagnosis was de ned with a corresponding main diagnosis according to at least one of the following International Statistical Classi cation of Diseases and Related Health Problems 10th Revision (ICD-10) codes: I21-I25 (ischemic heart diseases), I63-I66 (cerebral artery disease), I70-I74 (peripheral artery disease), I44.7 (left bundle-branch block), I50 (heart failure) I48 (atrial brillation and utter), I11 (hypertensive heart disease) and I34-I36 (valve disorders).As retinopathy and left ventricular hypertrophy were not systematically assessed, these variables were not included in the analysis.Additional le 1: Table S2 provides an overview about the speci c cutoffs and de nitions used for risk strati cation of the ESC/EASD model.b.SCORE risk model SCORE risk estimation is recommended for individuals without CVD and aged 40-64 years in accordance to the selection criteria of SCORE. 4,5The 10-year fatal CVD risk was calculated using the low SCORE risk chart based on the risk variables age, sex, smoking status, total cholesterol and systolic blood pressure. 5As indicated in the reference publication, we multiplied the SCORE risk estimates by 2 for men and 4 for women to account for the increased CV (cardiovascular) risk in individuals with diabetes. 5Risk estimation for individuals aged <40 and >65 years was performed referring to the risk estimates provided for individuals aged 40 and 65 years, respectively.


Endpoints

The primary outcome measure was CVD death at 10 years.Additional secondary outcome measures were all-cause

eath at 10
years and unplanned CVD as well as all-cause hospitalization at 5 years.Time at risk was calculated as time between enrolment and event or end of follow-up period whichever came rst.Data on death diagnosis was obtained from the Austrian Death Registry which includes information on cause of death based on ICD-10 codes.

CVD events resulting in death or hospitalization were de ned as atherosclerotic CVD, valvular heart disease, heart failur , malignant arrhythmia, peripheral artery disease and cerebrovascular disease.In case of unclear ICDclassi cations, the hospital charts were further examined to give a de nitive diagnosis of cause of death.The adjudication of CVD deaths was carried out by an experienced clinician who was blinded for the various risk assessments.


Statistical analysis

Continuous data are presented as median and interquartile range (IQR) and discrete data as frequency

nd percentages.Contin
ous variables were compared by the Kruskal-Wallis and Mann-Whitney-U-test, counts by the Fisher's exact test.

For the comparison between the risk estimators, i.e.NT-proBNP, ESC/EASD risk estimate and SCORE, both NT-proBNP and SCORE we e entered as continuous as well as categorical variables (NT-proBNP: tertiles and two groups with cut-off at 125 pg/ml; SCORE: 3 risk groups with cut-off: <5%, 5-10%, >10%).All comparisons were made for the total cohort and two subgroups similar to the characteristics of the original derivation cohort of SCORE.For the rst subgroup individuals with established CVD were excluded and the second subgroup consisted of patients without established CVD and an age ≥40 and <65 years.For outcome analysis we used a cause-speci c hazard model.Cox regression was performed to evaluate the association of the risk assessments with 10-year fatal CVD events and secondary outcome measures.Hazard ratios (HRs) of continuous NT-proBNP refer to ln-transformed NT-proBNP per 1-IQR increase.In addition to the univariate analysis, adjustments for potential confounders were conducted to demonstrate the robustness of NT-proBNP.Albumin/creatinine ratio, eGFR and age were added as continuous variables to the Cox multivariate regression model, while smoking status, hypertension and sex were added as dichotomous variables (yes/no).

Proportional hazard assumption was assessed and satis ed for all variables based on time interaction tests.

Cumulative incidence pl ts of the events of interest are shown for the various risk assessments.

Predictive performance was express d as discrimination (receiver operating characteristic [ROC] curve, C-index) and calibration usin Hosmer-Lemeshow goodness-of-t test (H-L).Observed 10-year risk for fatal CVD is presented using the Kaplan-Meier estimates.Differences in outcomes were assessed by non-overlapping con dence intervals (CI 95%) between C-indices corresponding to p-values of ≤ 0.05.

An improvement in individual risk prediction for the risk assessments was examined by the overall continuous net reclassi cation improvement (NRI) and presented as NRI (standard error [SE], p-value) as described by Pencina et al. 28 Furthermore risk classi cation tables for all reported outcomes (i.e.all-cause mortality, CVD mortality, all-cause hospitalization, CVD hospitalization) were presented comparing SCORE categories and the ESC/EASD risk model with tertiles of NT-proBNP.

A two-tailed p-value lower than 0.05 was considered statistically signi cant.Statistical analysis was performed using SPSS software (I M SPSS, Chicago, Illinois, USA) version 24 and STATA software (StataCorp, College Station, Texas, USA) version 13.


Results


Study population

A total of 2,186 T2DM patients were enrolled in the study, 496 patients were excluded from the analysis as

atus (n=460) and
iabetes duration (n=36) was not available, thus a total of 1,690 T2DM patients were analyzed.

Detailed description of the baseline characteristics is displayed in Table 1.Distribution of patients in the ESC/EASD and SCORE risk st ata as well as proportion of patients with normal (n=871) and elevated (n=819) NT-proBNP levels at a cut-off 125 pg/ml within these groups are illustrated in Figure 1 (p<0.001for both models).

Association of NT-proBNP, the ESC/EASD risk strata and SCORE with the primary endpoint CVD death and the secondary outcome all-cause deat at 10 years During 10 years of follow-up, 448 (26.5%) patients died, CVD death accounted to 44.9% (n=201) of all deaths.The cumulative incidences for all risk models strati ed into three groups with regards to both endpoints are shown in Figure 2. SCORE and NT-proBNP were both signi cantly associated with the primary outcome CVD death and the secondary o tcome all-cause death (p<0.001 for both) while the ESC/EASD risk model was only associated with allcause death (moderate risk vs. high risk: p=0.046 and vs. very high risk: p=0.031).Table 2 shows the results of the univariate Cox regression analysis.


Discussion

This is the rst study evaluating the predictive performance of the recently published ESC/EASD risk strati cation model and the SCORE risk estimation in a reasonably large real-world cohort of patients with T2DM and directly comparing these risk models with the biomarker NT-proBNP.Our results demonstrate that i) the ESC/EASD risk strati cation model performs limited compared to SCORE and NT-proBNP in terms of risk prediction and discriminatory accuracy ii) application of SCORE in a s

ected subgroup of T2DM patients resulted in a similar discriminative ability as achieved in no
-diabetics, iii) in both unselected and selected T2DM patients NT-proBNP remains a robust predictor for outcome, iv) in contrast to NT-proBNP the ESC/EASD nd SCORE risk model showed no outcome speci ty for future CVD events in T2DM individuals.

The 2016 ESC guidelines on CVD prevention classify most patients with T2DM at high or very high risk. 4Although diabetes has been long considered as a "cardiovascular risk equivalent", 29 more recent data indicate that incremental CV risk does not uniformly affect all patients with T2DM. 30,31Therefore, tools advocating a more individualized risk assessment are mandatory.The recently updated guideline

for the man
gement and prevention of CVD risk in individuals with diabetes integrated the aforementioned approach and introduced a new risk model accounting also for individuals at moderate risk. 3The new guidelines recommend for the rst time the use of a risk strati cation model based on three risk categories (moderate/high/very high) to aid treatment decisions in diabetes.However, the predictive performance of this model has neither been derived nor tested in patients with diabetes.

When applying the ESC/EASD risk criteria to our cohort, most patients with 67% were strati ed to the very high risk category, 15% met the ESC/EASD criteria for the high risk category whereas the moderate risk category was poorly represented with 1.5%; not to mention that a signi cant proportion of patients (17%) could not be categorized into either of the ESC/EASD CVD risk categories based on the model's strati cation criteria.Apparently, the new de ned moderate risk category as determined by short diabetes duration and no risk factors at all is poorly represented in a typical cohort of patients with T2DM.The high prevalence of risk factors such as obesity, hypertension and dyslipidemia even in individuals with short T2DM duration (as part of the metabolic syndrome) may explain the small number of patients strati ed into this group. 32ORE based risk estimation has been developed in the general population, but notably the original derivation cohort included diabetic patients. 5Since data on diabetes has not been collected uniformly in the SCORE project, the presence of diabetes has not been included as a predictor variable in the risk algorithm.However, it has been suggested that SCORE could be used for a rough risk assessment in patien s with diabetes.In the current report, SCORE was signi cantly associated with 10-year risk of fatal CVD but underestimated the risk for fatal CVD events in patients with T2DM.It is noteworthy that agreement between observed and predicted risk improved in the subgroup closest to the derivation cohort of SCORE, i.e. patients without CVD and aged between 40 and 64 years.

Similarly, the discriminatory ability for 10-years CVD death improved in this subgroup with a C-index of 0.69, which is similar to SCORE in non-diabetics; 5 however, resulted in the exclusion of more than half of the study population.

Since T2DM is particularly a disease of the elderly, age restrictions as given by SCORE would limit its utility in clinical practice.

Previous studies reported that CVD risk scores developed in the general population underestimate risk in individuals with T2DM. 33Conversely, a brief report by Coleman et al. investigating SCORE in 3,898 individuals with newly diagnosed T2DM from the UKPDS cohort reported that SCORE risk equation overestimates the 10-year risk for fatal CVD in individuals with T2DM by 18%, but provides good discriminatory accuracy with a C-index of 0.77 for fatal CVD events. 6However, the direct comparison of these results with our data may be limited, as the UKPDS included only individuals with newly diagnosed T2DM and treatment has been signi cantly changed within the study period.

Although much of CV risk can be attributed to traditional risk factors incorporated in classical risk prediction models, 34 they do not explain the full spectrum of CVD risk in diabetes. 35,36Potential limitations of risk scores calculated at a single point in time may be their inability to account for variability in measured risk factors (e.g.blood pressure), biological variation, exposure duration or untreated risk factor severity.It is not surpr sing, then, that major CV risk factors are also highly prevalent among individuals who will never experience a CVD event. 36This parad xon points to the need for risk assessment tools that allow a more integrated approach at the individual patient level.Intimately tied to this effort will be the requirement to individualize risk beyond the presence of established risk factors.

The current report demonstrates that single NT-proBNP measurement provides a more accurate risk estimate than the newly introduced ESC/EASD risk model and SCORE based risk assessment.7 In daily clinical practice, the assessment of a single marker that allows to identify those at highest risk seems attractive.In this study, as in others, [18][19][20][21][22] NT-proBNP levels above 125 pg/ml were strongly associated with adverse o tcomes in T2DM patients.Given the high negative predictive value of NT-proBNP at a cutoff level of 125mg/dl, 21,22 initial assessment of NT-proBNP could serve as a rst-line screening tool that allows to safely and effectively rule increased CV risk, while higher values would require further evaluation.In a second approach, additional cardiac inve tigations may be applied to further re ne individual risk.A recent published study conducted in asymptomatic individuals with T2DM reported an additive predictive value of NT-proBNP combined with coronary artery calcium scoring. 24 Siilarly, several other studies reported incremental prognostic information of NT-proBNP and troponin T when used in combination.[13][14][15]23 The association of NT-proBNP with (CV) hospital admissions observed in this report indicates that NT-proBNPguided isk strati cation may also have the potential for overall cost reductions as already exempli ed by previous natriuretic-guided trials in heart failure.37 The use of NT-proBNP would omit the need for calculation of scores as well as the problem of non-or misclassi cation or over tting as observed in global risk estimation models.

Yet, two trials provided initial evidence on the effectiveness of natriuretic peptides in guiding preventive efforts in patients at high risk for developing CVD events. 19,38In the prospective randomized controlled PONTIAC trial (NT-proBNP selected prevention of cardiac events in a population of diabetic patients without a history of cardiac disease) measurement of NT-proBNP (cut-off 125 pg/ml) was used to identify T2DM patients at high risk for developing CVD. 19These patients were then strati ed to either standard of care treatment or titration for reninangiotensin inhibitors and beta-blockers.A signi cant reduction of CVD events was reported in the treatment arm providing initial evidence for a NT-proBNP measurement-based selection of high-risk individuals with T2DM.Similarly, the STOP-HF study (St.Vincent's Screening to Prevent Heart Failure) demonstrated the effectiveness of BNP in guiding preventive efforts in patients with various CV risk factors. 38A recent observation from the CANVAS (Canagli ozin Cardiovascular Assessment Study) study has shown that canagli ozin treatment in T2DM patients with NT-proBNP levels above 125 pg/ml achieved greater absolute risk reductions in event rates compared to those with lower concentrations. 18 general, screening for high-risk individuals may only be appropriate when effective treatments are available.The high prevalence of modi able risk factors among individuals with diabetes but also the recent emerge of new therapies with favorable effects on CVD outcome such as sodium-glucose cotransporter-2 inhibitors underscore the impor ance of identifying those who would most probably bene t from initiation or optimizing treatment.


LIMITATIONS

We are aware of the following limitations of our study.First, retinopathy was not generally documented in this registry and could therefore not be implemented in the risk score which could have led to misclassi cation.In the current report only 16% of the patients were categorized as moderate or high risk.Hypothetically, the eventual identi cation of individuals with retinopathy would have led to even more patients being strati ed into the very highrisk group, resulting in an even greater weighting of the very high-risk category.A study by Klein et al. demonstrated   that retinopathy occurs more frequently in patients with long-term diabetes, CVD and proteinuria. 39Given the very high risk criteria of the ESC/EASD risk model it seems conceivable that patients with retinopathy may also have been captured in the very high-risk category.Second, as this registry included mainly outpatients followed in hospital, T2DM individuals at lower risk who are more often treated by general practitioners might be underrepresented.In this context, the ESC/EASD model might have performed different with an altered cohort including these patients.Third, survival status was not available in 460 patients after 10 years follow-up.Last, an inverse association between circulating levels of natriuretic peptides and body mass index has been reported earlier, 40 which could be a potential limitation in a population of T2DM patients.However, despite these observations, natriuretic peptides have been shown to retain prognostic performance in obese patients. 41


Conclusion

Overall, the current re

rt shows tha
the recently introduced ESC/EASD risk strati cation model provides only limited prognostic information in direct comparison to SCORE and most notably to single NT-proBNP assessment.NT-proBNP measurement is a simple and independent screening tool to identify individuals at increased risk for speci cally adverse CVD outcome applicable in a broad spectrum of T2DM patients.Future studies need to investigate the cost-effectiveness and feasibility of NT-proBNP-based screening and probably further add on marker.CONSENT PUBLICATION Not applicable.


List Of Abbreviations


AVAILABILITY OF DATA AND MATERIALS



CONSENT TO PARTICIPATEAll patients gave written informed consent and the study protocol was approved by the Ethics Committee of the Medical University of Vienna.


Table 1 .
1
Baseline characteristics of the overall study cohort.onthestated ESC/EASD criteria, as 280 patients had diabetes duration less than 10 years with 1 or 2 (but < 3) established CV risk factors and 8 patients presented with a diabetes duration longer than 10 years without any risk factors.Detailed characteristics for the ESC/EASD risk strata are presented in Additional le 1: TableS3.In the overall cohort, 654 patients (39%) had a calculated SCORE risk estimate below 5%, 525 patients (31%) between 5 and 10% and 511 patients (30%) above 10%.The calculated SCORE risk estimates for 10-year fatal increased with ESC/EASD risk category (0% [IQR 0-0] vs 6% [IQR 2-10] vs 8% [IQR 4-12], p<0.001 for the moderate, high and very-high risk category).
CharacteristicsOverall cohort (n=16