Visual Abstract
Keywords: albuminuria, cardiovascular disease, cardiovascular events, CKD, coronary artery disease, cardiovascular-kidney-metabolic (CKM) syndrome
Abstract
Key Points
The Predicting Risk of Cardiovascular Disease EVENTs equations optimize risk prediction in the modern era of increasing cardiovascular kidney metabolic disease prevalence.
Risk prediction is most helpful to intensify therapy in those with less advanced kidney disease.
In those with advanced kidney disease, risk indices provide little utility beyond biomarkers alone such as eGFR and albuminuria.
Background
Cardiovascular kidney metabolic syndrome is a growing public health problem and leading cause of cardiovascular mortality. We evaluated model performance of the American Heart Association Predicting Risk of Cardiovascular Disease EVENTs (PREVENT) equations in a large single health system population in patients with CKD and across strata of kidney function and compared findings where applicable to those of the pooled cohort equations (PCE).
Methods
Veterans with eGFR measurements were divided into three groups: eGFR ≥60, eGFR 30–59, and eGFR 15–29. We assessed the base PREVENT equations for total cardiovascular disease (CVD), atherosclerotic CVD (ASCVD), and heart failure as well as add-on equations using albuminuria. Accuracy of equations was assessed with c-index as a measure of discrimination and calibration curve slopes as a measure of calibration.
Results
Overall c-index was 0.651 for PREVENT-CVD, 0.633 for PREVENT-ASCVD, and 0.673 for PREVENT-heart failure. Discrimination of PREVENT-ASCVD was modestly greater than that of PCE (0.633 versus 0.629, respectively). For all three base PREVENT equations, discrimination declined with advancing kidney disease. Overall calibration slopes for the PREVENT equations ranged from 0.78 to 1.27. PREVENT-ASCVD had superior calibration compared to PCE with a calibration slope of 1.27 and 0.53, respectively. In those with urine albumin to creatinine ratio (UACR) measurements, discrimination of UACR add-on equations improved from base equations in those with less advanced kidney disease (c-index 0.624 for PREVENT-CVD UACR versus 0.607 for PREVENT-CVD base equation in those with eGFR ≥60, and c-index 0.586 for PREVENT-CVD UACR versus 0.563 for PREVENT-CVD base equation in those with eGFR 30–59). This trend was similar for the PREVENT-ASCVD and PREVENT-heart failure UACR add-on equations.
Conclusions
In a CKD population, the PREVENT equations perform best in those less advanced stages of CKD and cardiovascular kidney metabolic, presenting an opportunity to apply disease-preventing and disease-modifying therapeutics in a timely manner.
Introduction
Cardiovascular disease (CVD) is the leading cause of mortality in the United States.1 In recent years, there has been an increased recognition of the interconnectedness of cardiovascular, kidney, and metabolic disease with the definition of the cardiovascular–kidney–metabolic (CKM) syndrome as an important public health problem.2 The concept of CKM syndrome expands upon the previously recognized cardiometabolic syndrome by incorporating kidney disease as an additional risk factor for CVD.3 Kidney disease is characterized by the Kidney Disease Improving Global Outcomes (KDIGO) heat map framework and includes moderate-high risk CKD (CKD; KDIGO G3a and G3b) in stage 2 CKM and very high-risk CKD (KDIGO G4 and G5) in stage 3 CKM.4,5 The importance of kidney disease as a risk modifier within this spectrum has been amplified by novel therapeutics leading to marked improvements in both cardiovascular and kidney outcomes.
Gaps remain in the clinical implementation of these effective therapies, especially in the CKD population. Utilization of readily available cardiovascular risk prediction tools have become the basis for aligning preventive therapies based on absolute risk in the general population. Previously validated risk models have maintained a focus on risk assessment for atherosclerotic CVD (ASCVD) and do not address the growing burden of CVD phenotypes seen in overlapping CKM conditions especially heart failure.4,6 Challenges that have arisen with recalibrating existing CVD risk equations include having multiple equations for each unique populations and not using the readily available kidney disease markers, such as eGFR and urine albumin to creatinine ratio (UACR) that are highly predictive of CVD.7–11
The American Heart Association's (AHA) Predicting Risk of CVD EVENTs (PREVENT) equations were developed to address key gaps in current risk prediction models and to guide primary prevention across diverse populations.12 The PREVENT equations are sex-specific, race-free models that include metrics of kidney function to provide both 10- and 30-year risk estimates of CVD.6 In addition, they attempt to address the socioeconomic implications that lead to increased mortality in patients with CKM through the inclusion of social deprivation index.13 Importantly, the use of kidney function maximizes the predictive capability of these equations by adding serum creatinine and, when available, UACR. Whether this would also allow for improved prediction across the spectrum of kidney function has not been formally tested.14 In this study, we evaluate and compare the model performance of the PREVENT equations with the pooled cohort equations (PCE) in their ability to predict CVD events in a nationwide Veteran population with CKM risk factors.
Methods
Data were obtained from the Veterans Healthcare Administration Corporate Data Warehouse through the Veterans Integrated Networking and Computing Infrastructure. Approval was granted through expedited review by the Loma Linda Veterans Affairs (VA) Health care System Institutional Review Board and Research and Development committees.
Population
The creation of the cohort has been described previously.14 Briefly, all patients aged 30–79 years without a history of CVD (myocardial infarction [MI], stroke, or heart failure) and with available total cholesterol and HDL cholesterol levels between January 1, 2007, and December 31, 2009, were eligible for inclusion (4,087,496 individual records). Cholesterol and other laboratory values were restricted to noninpatient or nonemergency settings.
Exclusion criteria included age younger than 30 or older than 80 years, missing values for the base equation (age, sex, cholesterol values, systolic BP [SBP], or creatinine), history of CVD (including MI, stroke or heart failure), extreme values (SBP <90 or >200 mm Hg, total cholesterol <130 or>320 mg/dl, high-density lipoprotein cholesterol <20 or >100 mg/dl), and eGFR <15 ml/min per 1.73 m2.6 To be consistent with the PREVENT derivation cohort, missing values were not imputed.
eGFR was calculated using the 2021 race-free modification of the CKD-Epidemiology Collaboration formula.15 Patients were segregated into groups based on their eGFR only because UACR was only available in approximately 10% of patients, according to the KDIGO eGFR (in ml/min per 1.73 m2) based categories: eGFR ≥60 group; eGFR 30–59 (CKD G3a and 3b) group; and eGFR 15–29 (CKD G4) group. Patients with eGFR <15 ml/min per 1.73 m2 or on dialysis were excluded from this analysis, consistent with the PREVENT development population. UACR was also estimated from urine protein-to-creatinine ratio when that was available, using established estimating equations.16
Variables
Baseline covariates included age and sex, race and ethnicity, body mass index (in kg/m2), and diagnoses including hypertension, diabetes, ASCVD, and heart failure. The 10-year PREVENT score was calculated for total CVD (ASCVD [MI, and ischemic or nontraumatic hemorrhagic stroke], and heart failure) with the base model for the primary analysis.
Outcomes
Outcomes were assessed using inpatient and new outpatient diagnoses following index date (date of lipid panel); the last date of follow-up assessed was December 31, 2019. These events included total CVD: fatal and nonfatal ASCVD and heart failure; for nonfatal MI, only diagnostic codes (International Classification of Disease 10 Clinical Modification; and not the presence of revascularization) was used to identify cases. Cause of death was provided by Department of VA, Office of Mental Health and Suicide Prevention, Joint Department of VA, and Department of Defense Mortality Data Repository, data compiled from the National Death Index.17
Statistical Analysis
Baseline characteristics were evaluated based on distribution characteristics and type of variable. Normally distributed continuous variables were evaluated using parametric tests (one-way ANOVA); non-normally distributed continuous variables were evaluated using nonparametric tests (Kruskal). Categorical/factor variables were evaluated using chi-squared tests. A two-sided P value of <0.05 was the significance threshold for these descriptive comparisons. Outcomes were evaluated using a competing risk framework. Noncardiovascular deaths were considered competing events to cardiovascular deaths. Cumulative incidence functions were estimated for expected event rates over time accounting for competing events. The performance of PREVENT was assessed within the VA cohort using methodology to assess risk prediction models in a framework of competing risk analysis as described by van Geloven et al.18 The test comparison metrics used were calibration using the slope of the calibration curve and discrimination using the c-index or time specific area under the receiver operating characteristic curve. The calibration curves were directly comparable across equations as the intercept was set to zero for all equations. The calibration intercept was fixed at zero to exclude recalibration-in-the-large (systematic over- or underprediction of overall event risk) and to independently assess calibration slope. Calibration slope of 1 indicates perfect calibration, while <1 indicates overprediction of risk and >1 indicates underprediction of risk. Calibration slope was drawn using the ggplot2 package after estimation of the cumulative incidence function at each decile of predicted risk using cuminc function in cmprsk package. Calibration slope was assessed using generalized linear regression using observed values within deciles of predicted risk, with intercept set at zero for strict slope interpretation (using glm). C-index and 95% confidence interval (CI) were calculated using the score function in RiskRegression package of R. For the comparison between PREVENT-ASCVD and PCE, only the 40–79-year-old population was used to reflect the narrower age range for PCE. To compare discrimination of the PREVENT UACR add-on equations compared with PREVENT base, separate evaluations were conducted in patients with UACR measurements. Analyses were stratified by UACR values <30 mg/g and UACR ≥30 mg/g as well as by eGFR grouping. Difference in c-indices between PREVENT base and PREVENT UACR and PREVENT-ASCVD and PCE were conducted using bootstrapping to estimate difference and 95% CI of the differences. All analyses were performed using complete case analysis using R software for statistical computing.19 Specific packages used included TableOne, ggplot2, data.table, PredictABEL, rms, pROC, survival.
Results
Baseline Characteristics
A total of 2,498,611 individuals were included and stratified by eGFR groups: eGFR ≥60 (n=2,213,517), eGFR 30–59 (CKD G3; n=272,852), and eGFR 15–29 (CKD G4; n=12,242). Patients in the eGFR 30–59 and eGFR 15–29 groups were older with median age 71 years (interquartile range [IQR], 63–76) and 70 (IQR, 61–76), respectively. Patients in the eGFR ≥60 group were comparatively younger with median age 60 years (IQR, 53–66). The study population consisted predominantly of non-Hispanic White men. No considerable differences in race were observed between eGFR groups. Participants within the eGFR 15–29 group were more likely to have diabetes (51% versus 37% in the eGFR 30–59 group and 22% in the eGFR ≥60 group) and to be prescribed antihypertensive medications (66% versus 62% in the eGFR 30–59 group and 40% in the eGFR ≥60 group). Mean SBP, total cholesterol, HDL cholesterol, body mass index, and smoking prevalence were similar between participants in all three groups (Table 1).
Table 1.
Baseline characteristics by eGFR grouping: eGFR ≥60 ml/min per 1.73 m2 correlates to Kidney Disease Improving Global Outcomes CKD stages 0, G1, and G2; eGFR 30–59 ml/min per 1.73 m2 correlates to Kidney Disease Improving Global Outcomes CKD stages G3a and G3b; and eGFR 15–29 ml/min per 1.73 m2 correlates to Kidney Disease Improving Global Outcomes CKD stage G4
| Characteristic | Overall | EGFR ≥60 | eGFR 30–59 (CKD G3) | eGFR 15–29 (CKD G4) |
|---|---|---|---|---|
| N | 2,498,611 | 2,213,517 | 272,852 | 12,242 |
| Age, median (IQR) | 61 (54–69) | 60 (53–66) | 71.30 (63,76) | 70 (61,76) |
| Male sex, n (%) | 2,346,579 (94) | 2,071,498 (94) | 263,163 (96) | 11,918 (97) |
| Race and ethnicity, n (%) | ||||
| Asian/Native Hawaiian/Pacific Islander | 30,200 (1) | 27,198 (1) | 2845 (1) | 157 (1) |
| Hispanic | 129,633 (5) | 120,040 (5) | 8997 (3) | 596 (5) |
| Non-Hispanic Black | 381,651 (15) | 329,768 (15) | 48,952 (18) | 2931 (24) |
| Non-Hispanic White | 1,740,884 (70) | 1,553,078 (70) | 180,749 (66) | 7057 (58) |
| Other/unknown | 216,243 (9) | 183,433 (8) | 31,309 (12) | 1501 (12) |
| Cardiovascular risk factors or risk factors in PREVENT base model | ||||
| SBP, mm Hg, mean (SD) | 132 (15) | 132 (14) | 134 (16) | 139 (17) |
| Total cholesterol, mg/dl, mean (SD) | 189 (36) | 190 (36) | 180 (35) | 181 (38) |
| HDL, mg/dl, mean (SD) | 45 (13) | 45 (13) | 43 (12) | 41 (13) |
| BMI, kg/m2, mean (SD) | 29 (4) | 29 (5) | 29 (4) | 29 (5) |
| Diabetes, n (%) | 590,220 (24) | 483,538 (22) | 100,393 (37) | 6289 (51) |
| Current smoking, n (%) | 320,140 (13) | 284,369 (13) | 33,955 (12) | 1816 (15) |
| Antihypertensive treatment, n (%) | 1,066,225 (43) | 889,634 (40) | 168,467 (62) | 8124 (66) |
| Statin treatment, n (%) | 503,951 (20) | 415,943 (19) | 84,359 (31) | 3649 (30) |
| eGFR, median (IQR) | 82 (70–94) | 85 (75–96) | 53 (47–57) | 25 (21–28) |
| Add-on risk factors/predictors in optional models | ||||
| UACR, median (IQR) | 11 (5–38) | 10 (5–32) | 22 (7–86) | 104 (14–629) |
| Hemoglobin A1c among those with diabetes, median (%) | 7.0 (6.4–8.0) | 7.0 (6.4–8.1) | 6.9 (6.4–7.9) | 7.0 (6.3–8.0) |
| Hemoglobin A1c among those without diabetes, median (%) | 5.7 (5.4–5.9) | 5.7 (5.4–5.9) | 5.8 (5.5–6.0) | 5.7 (5.4–6.0) |
BMI, body mass index; IQR, interquartile range; PREVENT, Predicting Risk of Cardiovascular Disease EVENTs; SBP, systolic BP; UACR, urine albumin-to-creatinine ratio.
Incident Cardiovascular Events
Over a mean follow-up time of 8.7 years (SD 2.6 years), there were a total of 412,618 CVD events. There was a general trend toward higher total CVD events with advanced stages of CKD (38% in the eGFR 15–29 group, 27% in the eGFR 30–59 group, and 15% in the eGFR ≥60 group); a similar trend was observed with ASCVD events and heart failure events (Table 2). Cumulative incidence graphs of observed events over time for each eGFR group for the outcomes of total CVD, ASCVD, and heart failure are depicted in Supplemental Figure 1.
Table 2.
Cardiovascular outcomes by eGFR grouping: eGFR ≥60 ml/min per 1.73 m2 correlates to Kidney Disease Improving Global Outcomes CKD stages 0, G1, and G2; eGFR 30–59 ml/min per 1.73 m2 correlates to Kidney Disease Improving Global Outcomes CKD stages G3a and G3b; and eGFR 15–29 ml/min per 1.73 m2 correlates to Kidney Disease Improving Global Outcomes CKD stage G4
| Cardiovascular Outcomes, n (%) | ||||
|---|---|---|---|---|
| Cardiovascular Outcome | Overall | eGFR ≥60 | eGFR 30–59 (CKD G3) | eGFR 15–29 (CKD G4) |
| N | 2,498,611 | 2,213,517 | 272,852 | 12,242 |
| Total CVD events | 412,618 (17) | 334,212 (15) | 73,821 (27) | 4585 (38) |
| ASCVD events | 268,264 (11) | 219,716 (10) | 45,821 (17) | 2727 (22) |
| Fatal and nonfatal heart failure | 194,332 (8) | 152,106 (7) | 39,421 (14) | 2805 (23) |
All values across rows were statistically significant to a P value < 0.05 using ANOVA. ASCVD, atherosclerotic cardiovascular disease; CVD, cardiovascular disease.
Supplemental Figure 2 demonstrates the predicted and observed events for ASCVD by PCE and PREVENT and for total CVD and heart failure. Overall, PCE demonstrated higher predicted risk as compared with observed, PREVENT-ASCVD and PREVENT-CVD demonstrated underprediction, with a slight overprediction for PREVENT-heart failure.
Model Performance across Characteristics
Median predicted risk by the base 10-year PREVENT equation is displayed by eGFR group and by occurrence of CVD event, mortality, or none (Figure 1). Predicted risk was higher across eGFR groups, as well as for those with CVD events or mortality versus those without in each eGFR group. However, there was greater overlap in predicted risk between those without and with events or mortality in the lower eGFR ranges (Figure 2).
Figure 1.
AHA PREVENT total CVD risk estimates across CKD stages by occurrence of event. 0=no event; 1=total CVD (fatal or nonfatal MI, stroke, or heart failure); 2=noncardiovascular disease death). AHA, American Heart Association; CVD, cardiovascular disease; MI, myocardial infarction; PREVENT, Predicting Risk of Cardiovascular Disease EVENTs.
Figure 2.
Calibration curves for PREVENT base equations and PCE by eGFR grouping: eGFR ≥60 ml/min per 1.73 m2 correlates to KDIGO CKD stages 0, G1, and G2; eGFR 30–59 ml/min per 1.73 m2 correlates to KDIGO CKD stages G3a and G3b; and eGFR 15–29 ml/min per 1.73 m2 correlates to KDIGO CKD stage G4. (A) Ten-year total CVD (fatal and nonfatal MI, stroke, or heart failure); (B) 10-year ASCVD (fatal and nonfatal MI or stroke); (C) 10-year heart failure; (D) 10-year ASCVD for PCE. ASCVD, atherosclerotic cardiovascular disease; CIF, cumulative index function; HF, heart failure; KDIGO, Kidney Disease Improving Global Outcomes; PCE, pooled cohort equation.
Model performance, including discrimination and calibration for prediction of total CVD, ASCVD, and heart failure, are depicted in Tables 3 and 4. The discrimination of each risk prediction model was assessed using the c-index (which includes time to event). The c-indices were the highest for the eGFR ≥60 group, and the highest in the overall cohort, as compared with the lower eGFR groups. Calibration demonstrated underprediction for PREVENT-CVD and PREVENT-ASCVD (calibration slope >1) versus overprediction of risk for PREVENT-heart failure and PCE (calibration slope <1).
Table 3.
C-index performance by eGFR grouping (confidence interval) for Predicting Risk of Cardiovascular Disease EVENTs-cardiovascular disease, Predicting Risk of Cardiovascular Disease EVENTs-atherosclerotic cardiovascular disease, Predicting Risk of Cardiovascular Disease EVENTs-heart failure, and pooled cohort equation: eGFR ≥60 ml/min per 1.73 m2 correlates to Kidney Disease Improving Global Outcomes CKD stages 0, G1, and G2; eGFR 30–59 ml/min per 1.73 m2 correlates to Kidney Disease Improving Global Outcomes CKD stages G3a and G3b; and eGFR 15–29 ml/min per 1.73 m2 correlates to Kidney Disease Improving Global Outcomes CKD stage G4
| Prediction Tool | Overall | EGFR ≥60 | eGFR 30–59 (CKD G3) | eGFR 15–29 (CKD G4) |
|---|---|---|---|---|
| PREVENT-CVD | 0.651 (0.651 to 0.651) | 0.640 (0.639 to 0.622) | 0.590 (0.590 to 0.590) | 0.562 (0.551 to 0.572) |
| PREVENT-ASCVD | 0.633 (0.633 to 0.633) | 0.627 (0.627 to 0.627) | 0.574 (0.574 to 0.574) | 0.552 (0.539 to 0.564) |
| PREVENT-heart failure | 0.673 (0.673 to 0.673) | 0.659 (0.659 to 0.659) | 0.608 (0.608 to 0.608) | 0.571 (0.559 to 0.583) |
| PCE | 0.629 (0.629 to 0.629) | 0.625 (0.625 to 0.625) | 0.564 (0.564 to 0.564) | 0.542 (0.530 to 0.554) |
ASCVD, atherosclerotic cardiovascular disease; CVD, cardiovascular disease; PCE, pooled cohort equation; PREVENT, Predicting Risk of Cardiovascular Disease EVENTs.
Table 4.
Calibration by eGFR grouping for Predicting Risk of Cardiovascular Disease EVENTs-cardiovascular disease, Predicting Risk of Cardiovascular Disease EVENTs-atherosclerotic cardiovascular disease, Predicting Risk of Cardiovascular Disease EVENTs-heart failure, and pooled cohort equation: eGFR ≥60 ml/min per 1.73 m2 correlates to Kidney Disease Improving Global Outcomes CKD stages 0, G1, and G2; eGFR 30–59 ml/min per 1.73 m2 correlates to Kidney Disease Improving Global Outcomes CKD stages G3a and G3b; and eGFR 15–29 ml/min per 1.73 m2 correlates to Kidney Disease Improving Global Outcomes CKD stage G4
| Prediction Tool | Overall | eGFR ≥60 | eGFR 30–59 (CKD G3) | eGFR 15–29 (CKD G4) |
|---|---|---|---|---|
| PREVENT-CVD | ||||
| Events, n (%) | 412,618 (16.5) | 334,212 (15.1) | 73,821 (27.1) | 4585 (37.5) |
| Calibration slope (95% CI) | 1.21 (1.08 to 1.34) | 1.09 (0.92 to 1.27) | 1.26 (1.03 to 1.48) | 1.25 (0.98 to 1.52) |
| PREVENT-ASCVD | ||||
| Events, n (%) | 268,264 (10.7) | 219,716 (9.9) | 45,821 (16.8) | 2727 (22.3) |
| Calibration slope | 1.27 (1.13 to 1.41) | 1.15 (0.96 to 1.34) | 1.30 (1.06 to 1.55) | 1.34 (1.05 to 1.63) |
| PREVENT-heart failure | ||||
| Events, n (%) | 194,332 (7.8) | 152,106 (6.9) | 39,421 (14.4) | 2805 (22.9) |
| Calibration slope | 0.78 (0.73 to 0.83) | 0.90 (0.74 to 1.05) | 0.86 (0.77 to 0.95) | 0.75 (0.67 to 0.82) |
| PCE | ||||
| Events, n (%) | 268,264 (10.7) | 219,716 (9.9) | 45,821 (16.8) | 2727 (22.3) |
| Calibration slope | 0.53 (0.45 to 0.60) | 0.50 (0.38 to 0.61) | 0.49 (0.39 to 0.59) | 0.57 (0.42 to 0.72) |
ASCVD, atherosclerotic cardiovascular disease; CI, confidence interval; CVD, cardiovascular disease; PCE, pooled cohort equation; PREVENT, Predicting Risk of Cardiovascular Disease EVENTs.
PREVENT-CVD
Across all eGFR groups, the PREVENT-CVD equation had an overall c-index 0.651. Discrimination declined with advancing kidney disease: c-index was 0.640 in the eGFR ≥60 group, 0.590 in the eGFR 30–59 group, and 0.562 in the eGFR 15–29 group (Table 3).
Calibration slope analysis (Table 4) demonstrated that the PREVENT-CVD was best calibrated at less advanced disease where predicted risks aligned more closely with observed events and had an overall tendency to underpredict risk. Calibration slope was 1.09 in the eGFR≥ 60 group, 1.26 in the eGFR 30–59 group, and 1.25 in the eGFR 15–29 group. Overall calibration slope for this equation was 1.21.
PREVENT-ASCVD
Across all eGFR groups, the PREVENT-ASCVD equations had an overall c-index of 0.633. Discrimination declined with advancing disease: c-index was 0.627 in the eGFR ≥60 group, 0.574 in the eGFR 30–59 group, and 0.552 in the eGFR 15–29 group. PREVENT-ASCVD had better discrimination across all eGFR groups when compared with PCE (Table 3).
Calibration analysis demonstrated that PREVENT-ASCVD performed best at less advanced disease and had an overall tendency to underpredict risk: Calibration slope was 1.15 in the eGFR ≥60 group, 1.30 in the eGFR 30–59 group, and 1.34 in the eGFR 15–29 group. Comparison of calibration slope between PREVENT-ASCVD and PCE showed a deviation toward underprediction with advancing disease for PREVENT and a deviation toward overprediction with advancing disease for PCE. Overall calibration was 1.27 (95% CI, 1.13 to 1.41) for PREVENT-ASCVD and 0.53 (95% CI, 0.45 to 0.60) for PCE.
PREVENT-Heart Failure
Across all eGFR groups, the PREVENT-heart failure equation had the best discrimination out of all PREVENT equations with an overall c-index 0.673. Discrimination declined with advancing disease: c-index was 0.659 in the eGFR ≥60 group, 0.608 in the eGFR 30–59 group, and 0.571 in the eGFR 15–29 group (Table 3).
Calibration analysis for PREVENT-heart failure showed the best calibration in less advanced disease and an overall tendency to overpredict risk. There was more overprediction of events with more advanced kidney disease (calibration slope was 0.90 in the eGFR ≥60 group, 0.86 in the eGFR 30–59 group, and 0.75 in the eGFR 15–29 group). Overall calibration for this equation was 0.78.
UACR Add-On Equations
In the analysis of PREVENT base equations and PREVENT UACR add-on equations stratified by eGFR grouping in the overall population, discrimination of UACR add-on was not superior to PREVENT base; of note, the majority of individuals would have used the UACR missing variable; thus, the overall β estimates in the UACR only PREVENT equation would likely be inferior to the base equation (Supplemental Table 1). Discrimination of PREVENT-ASCVD and PREVENT-ASCVD UACR was superior to PCE and had a positive delta c-index at all eGFR groups (Supplemental Table 1).
In the subset with UACR (N=228,898) measurements, discrimination and calibration analysis was performed on all PREVENT base and PREVENT UACR add-on equations as well as PCE, one stratified by eGFR grouping and another by UACR <30 mg/g and UACR ≥30 mg/g (Supplemental Tables 2 and 3).
In the entire population with UACR measurements (Supplemental Table 2), PREVENT UACR improved discrimination (higher) for the eGFR ≥60 and eGFR 30–59 groups, but not in the eGFR 15–29 group. In the analysis of PREVENT base and PREVENT UACR add-on equations stratified by UACR <30 mg/g and UACR ≥30 mg/g, discrimination of PREVENT UACR add-on equations were superior to PREVENT base equations in those with UACR ≥30. In those with UACR <30 mg/g, base and UACR add-on equations had similar discrimination (Supplemental Table 3).
Analysis across Regions of the United States
Regional analysis was done using PREVENT-CVD base equation across eGFR grouping (Supplemental Table 4). Results showed that the Pacific region had the highest overall c-index (0.664) and the Midwest region had the lowest overall c-index (0.637). The Continental, Southeast, and North Atlantic had c-indices in-between. Discrimination was the highest in the Pacific and the lowest in the Midwest across all eGFR groupings.
Discussion
In this large national cohort of US Veterans, we found that the PREVENT equations offer useful risk prediction in patients with less advanced kidney disease. The base PREVENT equations had c-indices between 0.633 and 0.673 and across all three equations, discrimination decreased with lower eGFR, dropping to nondiscriminatory range in those with GFR 15–29 (CKD G4). PREVENT-ASCVD had superior discrimination when compared with PCE, particularly in those with less advanced kidney disease. In those with UACR measurements, UACR add-on equations had superior discrimination compared with their base counterparts in the overall population and specifically in those with UACR >30 mg/g.
Cardiovascular risk prediction in patients with CKD has been an area of interest for decades, as existing risk indices were not designed with the incorporation of kidney function parameters or with populations with a representative sample of patients with CKD patients. In fact, existing risk indices applied to CKD populations consistently showed poorer performance than when compared with a non-CKD population.4,10,20 The CKM scientific statement put forth by the AHA called for risk algorithms that reflect the multiple cardiovascular outcomes that patients with CKM experience to better target therapies in the subpopulations that would benefit most.21 It has been long recognized that both eGFR and albuminuria (assessed as UACR) are independent predictors of CVD; however, up until now, there were no readily available equations that had a priori included CKD variables in the original models. The PREVENT equations sought to include kidney and metabolic-specific variables that are known risk enhancers to improve risk prediction and predict outcomes beyond ASCVD.
Our findings suggest that the PREVENT equations have the greatest clinical utility among individuals with less advanced kidney disease (CKD G1–G3), where discrimination and calibration are relatively more robust. Although c-indices in our analyses did not exceed 0.7, this likely reflects the high-risk nature of our veteran population with older age and greater comorbidity burden compared with the general population in which these models were developed. This was particularly noticeable in the lower eGFR groupings. However, in these groups, there is higher baseline risk because of greater comorbidity. Thus, discrimination becomes more difficult with greater baseline risk within a subgroup. This also explains why the discrimination was enhanced in the overall population as opposed to the GFR subgroups as the ability to distinguish those with and without events was greater when there was a wider distribution of risk. This was also seen in a recent analysis of cardiovascular risk estimation using ceramide biomarkers in the ESKD population. In this secondary analysis of the A Study to Evaluate the Use of Rosuvastatin in Subjects on Regular Hemodialysis: An Assessment of Survival and Cardiovascular Events and Die Deutsche Diabetes Dialyze Studie trials, peak c-index when an established risk score+biomarker score was used was 0.62 (0.60–0.64). Either of the scores by themselves resulted in c-index <0.6 universally.22
These findings underscore the importance of lower eGFR and greater UACR alone in the indication of heightened CVD risk. Once the eGFR is <60 ml/min per 1.73 m2 and especially <45 ml/min per 1.73 m2, or UACR is >30 mg/g, there is sufficient elevation in CVD risk to intensify risk-reduction measures. Risk prediction, then, is best reserved for those with more preserved kidney function or those in early kidney disease who may be reluctant to add on to medication burden. To illustrate how these equations may be applied in practice, consider a common clinical scenario: a 45-year-old woman with a SBP of 140 mm Hg, an eGFR of 40 ml/min per 1.73 m2, and a UACR of 80 mg/g. Using the PREVENT equations, her estimated 10-year cardiovascular risk is approximately 11%, with a 30-year risk of 35%, correlating to 11 and 35 events per 100 similar individuals, respectively. These estimates support the intensification of therapy, including tight BP control with renin-angiotensin aldosterone system blockade (oral medications indicated in those with 10-year PREVENT score >7.5%23), initiation of statin therapy (moderate-intensity statin for those with PREVENT score between 5% and 10% and high-intensity statin in those with a PREVENT score >10%24), and consideration of an sodium-glucose cotransporter 2 inhibitor. Such a patient may be largely asymptomatic and could benefit from additional objective data if there is a question as to why they would need intensified therapy. In addition, presenting both short- and long-term risk estimates may enhance patient understanding of disease trajectory and reinforce the importance of early and sustained risk factor modification.
Conversely, our findings also highlight that in individuals with more advanced CKD (particularly G4), the PREVENT equations provide limited incremental value beyond established risk enhancers such as eGFR and UACR. In these patients, lower kidney function and elevated albuminuria are themselves strong indicators of high cardiovascular risk and should prompt treatment intensification irrespective of calculated risk. KDIGO CKD and KDIGO-American Diabetes Association guidelines both emphasize aggressive BP, diabetes, and cholesterol risk factor modification based on a diagnosis of CKD alone.25,26
Albuminuria has been reported by many studies as a strong predictor of CVD,27 independent of eGFR across baseline kidney function strata (Supplemental Figure 3). As such, one of the novel features of the PREVENT equations was the opportunity to add albuminuria, if available, to enhance risk prediction. The current analysis demonstrates that the UACR performance was best when UACR was available. The addition of UACR value to PREVENT UACR add-on equations showed superior discrimination than PREVENT base equations in those with eGFR >30 and in those with UACR values ≥30 mg/g. These findings strengthen the point that routine screening with collection of UACR in the primary care setting might play an important role in both early diagnosis of CKM and improved risk prediction of the PREVENT equations.
The PREVENT equations were also novel in the addition of heart failure as a primary outcome. Contrary to ASCVD, there are few validated heart failure risk prediction models. The PCE to prevent heart failure while widely used has only moderate discrimination and poor calibration in a CKD population.28 In our analysis, PREVENT-heart failure had the strongest discrimination and calibration compared with other base equations and PCE, but similarly declined significantly with advanced kidney disease. These outcomes highlight the value of the PREVENT equations in their ability to predict heart failure outcomes as a subset of CVD in the growing CKM population. Furthermore, the better calibration of this equation suggests that kidney-specific variables are important markers in the prediction of heart failure events. Future studies with heart failure models are needed to determine data cutoffs for therapeutic interventions.
This study had several strengths, including a large, national, multiethnic patient population. There was available data on kidney function in a majority of the patient sample. This study also has several important limitations. The analysis was conducted in a predominantly male, older age US veteran population, which may limit generalizability to broader, more diverse populations including women, younger individuals, and non-US populations. Overestimation of events may have occurred by utilization outpatient diagnostic codes rather than hospitalization records. Alternative statistics such as model-based c-index, to account for case-mix diversity, was not possible in this cohort. It is important to note that 95% CIs from standards statistics only include uncertainty because of sampling variation. Measurement errors and systematic errors in event coding are not included in the error estimates. The effect of mental health disorders in the VA population such as post-traumatic-stress-disorder and major depressive disorder was not accounted for in this study.
The PREVENT equations build upon previously available risk tools to better identify high-risk patients with CKM risk factors. The ability to predict multiple outcomes including total CVD, ASCVD, and heart failure caters to the present and ever-changing landscape of CVD. These equations may perform best in those with less advanced stages of CKD and CKM, presenting an opportunity to apply disease-preventing and disease-modifying therapeutics in a timely manner. Ultimately, eGFR <60 ml/min per 1.73 m2 and UACR >30 mg/g remain sufficient predictors of CVD underscoring the importance of having both monitored as routine risk prediction markers in the general population.
Supplementary Material
Disclosures
Disclosure forms, as provided by each author, are available with the online version of the article at https://links.lww.com/CJN/C835.
Author Contributions
Data curation: Roy O. Mathew.
Formal analysis: Roy O. Mathew.
Methodology: Roy O. Mathew.
Software: Roy O. Mathew.
Supervision: Roy O. Mathew.
Validation: Roy O. Mathew, Nikitha Murthy.
Writing – original draft: Roy O. Mathew, Nikitha Murthy, Alyssa Sanchez.
Writing – review & editing: Sadiya S. Khan, Roy O. Mathew, Nikitha Murthy, Janani Rangaswami, Alyssa Sanchez.
Funding
None.
Data Availability Statements
Original data cannot be shared. Explanation Why Data Cannot Be Shared: Confidential data belonging to the VA health care database.
Supplemental Material
This article contains supplemental material online, published as provided by the authors, at https://links.lww.com/CJN/C836.
Supplemental Table 1. Delta C-index performance by CKD stage (CI) for PREVENT base equations compared with PREVENT UACR add-on equation and PREVENT-ASCVD compared to PCE.
Supplemental Table 2. C-indices for PREVENT base, UACR add-on equations, and PCE stratified by eGFR grouping among those with UACR values.
Supplemental Table 3. C-indices for PREVENT base, UACR add-on equations, and PCE stratified by those with UACR <30 mg/g and UACR ≥30 mg/g, among those with UACR values.
Supplemental Table 4. Regional analysis of c-indices for PREVENT-CVD base equation across.
Supplemental Figure 1. CKD-specific cumulative index function plots for the outcomes.
Supplemental Figure 2. Observed versus predicted events for PREVENT equations and PCE.
Supplemental Figure 3. Heat map of AHA PREVENT-CVD risk stratified by KDIGO eGFR stage and UACR category.
References
- 1.Ahmad FB, Anderson RN. The leading causes of death in the US for 2020. JAMA. 2021;325(18):1829–1830. doi: 10.1001/jama.2021.5469 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 2.Minhas AMK Mathew RO Sperling LS, et al. Prevalence of the cardiovascular-kidney-metabolic syndrome in the United States. J Am Coll Cardiol. 2024;83(18):1824–1826. doi: 10.1016/j.jacc.2024.03.368 [DOI] [PubMed] [Google Scholar]
- 3.Grundy SM Cleeman JI Daniels SR, et al. Diagnosis and management of the metabolic syndrome. Circulation. 2005;112(17):2735–2752. doi: 10.1161/CIRCULATIONAHA.105.169404 [DOI] [PubMed] [Google Scholar]
- 4.Stevens PE Ahmed SB Carrero JJ, et al. KDIGO 2024 clinical practice guideline for the evaluation and management of chronic kidney disease. Kidney Int. 2024;105(4):S117–S314. doi: 10.1016/j.kint.2023.10.018 [DOI] [PubMed] [Google Scholar]
- 5.Ndumele CE Neeland IJ Tuttle KR, et al. A synopsis of the evidence for the science and clinical management of cardiovascular-kidney-metabolic (CKM) syndrome: a scientific statement from the American Heart Association. Circulation. 2023;148(20):1636–1664. doi: 10.1161/CIR.0000000000001186 [DOI] [PubMed] [Google Scholar]
- 6.Khan SS Matsushita K Sang Y, et al. Development and validation of the American Heart Association’s PREVENT equations. Circulation. 2024;149(6):430–449. doi: 10.1161/CIRCULATIONAHA.123.067626 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 7.Matsushita K Kaptoge S Hageman SHJ, et al. Including measures of chronic kidney disease to improve cardiovascular risk prediction by SCORE2 and SCORE2-OP. Eur J Prev Cardiol. 2023;30(1):8–16. doi: 10.1093/EURJPC/ZWAC176 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 8.Matsushita K Jassal SK Sang Y, et al. Incorporating kidney disease measures into cardiovascular risk prediction: development and validation in 9 million adults from 72 datasets. EClinicalMedicine. 2020;27:100552. doi: 10.1016/j.eclinm.2020.100552 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 9.Bundy JD Rahman M Matsushita K, et al. Risk prediction models for atherosclerotic cardiovascular disease in patients with chronic kidney disease: the CRIC study. J Am Soc Nephrol. 2022;33(3):601–611. doi: 10.1681/ASN.2021060747 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 10.Major RW Grant R Shepherd D, et al. Cardiovascular risk models in chronic kidney disease: a systematic review and external validation of models. medRxiv. 2022. doi: 10.1101/2022.01.22.22268958 [DOI] [Google Scholar]
- 11.Chen SC Su HM Tsai YC, et al. Framingham risk score with cardiovascular events in chronic kidney disease. PLoS One. 2013;8(3):e60008. doi: 10.1371/JOURNAL.PONE.0060008 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 12.Khan SS Coresh J Pencina MJ, et al. Novel prediction equations for absolute risk assessment of total cardiovascular disease incorporating cardiovascular-kidney-metabolic health: a scientific statement from the American Heart Association. Circulation. 2023;148(24):1982–2004. doi: 10.1161/CIR.0000000000001191 [DOI] [PubMed] [Google Scholar]
- 13.Bevan GH, Nasir K, Rajagopalan S, Al-Kindi S. Socioeconomic deprivation and premature cardiovascular mortality in the United States. Mayo Clin Proc. 2022;97(6):1108–1113. doi: 10.1016/j.mayocp.2022.01.018 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 14.Mathew RO Khan SS Tuttle KR, et al. Performance of the American Heart Association’s PREVENT risk score for cardiovascular risk prediction in a multiethnic population. Nat Med. 2025;31(8):2655–2662. doi: 10.1038/S41591-025-03789-2 [DOI] [PubMed] [Google Scholar]
- 15.Inker LA Eneanya ND Coresh J, et al. New creatinine- and cystatin C–based equations to estimate GFR without race. New Engl J Med. 2021;385(19):1737–1749. doi: 10.1056/NEJMoa2102953 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 16.Sumida K Nadkarni GN Grams ME, et al. Conversion of urine protein–creatinine ratio or urine dipstick protein to urine albumin–creatinine ratio for use in chronic kidney disease screening and prognosis. Ann Intern Med. 2020;173(6):426–435. doi: 10.7326/M20-0529 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 17.United States Department of Veterans Affairs . Using Data to Prevent Veteran Mortality a Guide to Accessing and Understanding Mortality Data in the VA Mortality Data Repository; 2023. Accessed June 22, 2026. https://mirecc.va.gov/MIRECC/suicideprevention/documents/VA_DoD_MDR_Flyer_022026_update.pdf [Google Scholar]
- 18.van Geloven N Giardiello D Bonneville EF, et al. Validation of prediction models in the presence of competing risks: a guide through modern methods. BMJ. 2022;377:e069249. doi: 10.1136/bmj-2021-069249 [DOI] [PubMed] [Google Scholar]
- 19.R Foundation for Statistical Computing. Comprehensive R Archive Network (CRAN). Accessed June 23, 2026. https://cran.r-project.org/ [Google Scholar]
- 20.Weiner DE Tighiouart H Elsayed EF, et al. The framingham predictive instrument in chronic kidney disease. J Am Coll Cardiol. 2007;50(3):217–224. doi: 10.1016/j.jacc.2007.03.037 [DOI] [PubMed] [Google Scholar]
- 21.Ndumele CE Rangaswami J Chow SL, et al. Cardiovascular-kidney-metabolic health: a presidential advisory from the American Heart Association. Circulation. 2023;148(20):1606–1635. doi: 10.1161/CIR.0000000000001184 [DOI] [PubMed] [Google Scholar]
- 22.Witoslawska A Drechsler C Duarte K, et al. Predictive performance of the cardiovascular event risk test 2 risk score in hemodialysis patients with ESKD. Clin J Am Soc Nephrol. 2025;20(12):1683–1695. doi: 10.2215/CJN.0000000831 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 23.Jones DW Ferdinand KC Taler SJ, et al. 2025 AHA/ACC/AANP/AAPA/ABC/ACCP/ACPM/AGS/AMA/ASPC/NMA/PCNA/SGIM guideline for the prevention, detection, evaluation and management of high blood pressure in adults: a report of the American College of Cardiology/American Heart Association Joint Committee on clinical practice guidelines. Circulation. 2025;152(11):e114–e218. doi: 10.1161/CIR.0000000000001356 [DOI] [PubMed] [Google Scholar]
- 24.Blumenthal RS Morris PB Gaudino M, et al. 2026 ACC/AHA/AACVPR/ABC/ACPM/ADA/AGS/APhA/ASPC/NLA/PCNA guideline on the management of dyslipidemia: a report of the American College of Cardiology/American Heart Association Joint Committee on clinical practice guidelines. Circulation. 2026;153(17):e1154–e1276. doi: 10.1161/CIR.0000000000001423 [DOI] [PubMed] [Google Scholar]
- 25.Kidney Disease: Improving Global Outcomes (KDIGO) CKD Work Group. KDIGO 2024 clinical practice guideline for the evaluation and management of chronic kidney disease. Kidney Int. 2024;105(4S):S117–S314. doi: 10.1016/S0085-2538(24)00110-8 [DOI] [PubMed] [Google Scholar]
- 26.De Boer IH Khunti K Sadusky T, et al. Diabetes management in chronic kidney disease: a consensus report by the American Diabetes Association (ADA) and Kidney Disease: Improving Global Outcomes (KDIGO). Kidney Int. 2022;102(5):974–989. doi: 10.1016/j.kint.2022.08.012 [DOI] [PubMed] [Google Scholar]
- 27.Matsushita K, Ballew SH, Wang AYM, Kalyesubula R, Schaeffner E, Agarwal R. Epidemiology and risk of cardiovascular disease in populations with chronic kidney disease. Nat Rev Nephrol. 2022;18(11):696–707. doi: 10.1038/s41581-022-00616-6 [DOI] [PubMed] [Google Scholar]
- 28.Mehta R Ning H Bansal N, et al. Ten-year risk-prediction equations for incident heart failure hospitalizations in chronic kidney disease: findings from the chronic renal insufficiency cohort study and the multi-ethnic study of atherosclerosis. J Card Fail. 2022;28(4):540–550. doi: 10.1016/j.cardfail.2021.10.007 [DOI] [PMC free article] [PubMed] [Google Scholar]
Associated Data
This section collects any data citations, data availability statements, or supplementary materials included in this article.
Data Availability Statement
Original data cannot be shared. Explanation Why Data Cannot Be Shared: Confidential data belonging to the VA health care database.