PREVENT Risk Score vs the Pooled Cohort Equations in MESA

Abstract

Background

In 2023, the American Heart Association developed the PREVENT (Predicting Risk of CVD Events) equations to estimate risk of atherosclerotic cardiovascular disease (ASCVD) and heart failure (HF).

Objectives

Assess the comparative performance of PREVENT-ASCVD vs current guideline-recommended Pooled Cohort Equations (PCE). Evaluate the performance of the PREVENT-HF risk algorithm.

Methods

In 6,098 individuals from the MESA (Multi-Ethnic Study of Atherosclerosis) cohort, we calculated baseline PCE, and PREVENT predicted 10-year ASCVD event percentages, observed event percentages at 10 years, discordance between observed and expected percentages, discrimination using Harrell’s C index, and calibration using mean absolute error.

Results

Observed ASCVD event rate (6.0%) was closer to the predicted PREVENT event rate (5.7%) than the PCE (10.8%). PREVENT was more accurate in women than men (3.3% vs −11.6% discordance between observed and PREVENT predicted ASCVD), nonsmokers compared to smokers (2.4% vs −37.0% discordance), chronic kidney disease stages 3/4 (discordance 3.2%), and those with high social deprivation scores (discordance −5.0%). Forty-two percent of this cohort would be re-classified to a lower ASCVD risk category using the PREVENT equation vs the PCE. PREVENT-HF overestimates HF events by 2.1%, a relative risk overestimation of 62.6%.

Conclusions

PREVENT-ASCVD equations demonstrated a more accurate ASCVD risk-prediction stratification than the PCE. PREVENT performs best in women, nonsmokers, those with a greater degree of renal dysfunction, social deprivation, and Black individuals. PREVENT-HF overestimates risk of incident HF in a Multi-Ethnic Study of Atherosclerosis.

Central Illustration

For decades, atherosclerotic cardiovascular disease (ASCVD) has remained a leading cause of morbidity and mortality worldwide.¹ Risk calculators are essential to guide primary prevention management strategies. In 2013, the American College of Cardiology (ACC) and American Heart Association (AHA) developed the Pooled Cohort Equations (PCE) to estimate 10-year risk of development of ASCVD, which has been the guideline-recommended strategy for risk stratification clinically.² Despite its clinical utility, there have been several drawbacks identified; in particular, it has been shown to have poor predictive value in modern, diverse populations and includes race as a risk factor, which has been shown to be noncontributory to the risk-prediction models in some cases.^3,4

In an effort to improve and optimize risk prediction, in 2023, the AHA developed the PREVENT equations (AHA Predicting Risk of CVD Events) to estimate risk of cardiovascular disease (CVD).⁵ The PREVENT risk calculator gives separate 10- and 30-year risk estimates of ASCVD, heart failure (HF) events, and total CVD, defined as a composite of ASCVD and HF. In contrast to the PCE, PREVENT provides risk estimates for a broader age range (30-79 years of age vs 40-79) and is race-free in its prediction. In addition, the PREVENT calculator incorporates kidney function, hemoglobin A1c, and social determinants of health in its algorithm, when available.

Within the cohort used to create the PREVENT calculator, there were a low number of Hispanic and Asian individuals (6.0% and 2.6% female participants, and 5.3% and 2.5% male participants, respectively), which has been identified as a limitation in the risk calculator validation and an area of further research.⁶ Furthermore, comparative analyses between the new proposed risk-prediction model and the current guideline-recommended PCE in modern diverse cohorts is lacking, particularly in validating HF outcomes. As such, we aimed to assess the PREVENT risk equations performance against the PCE using a diverse, modern cohort with special attention to subgroup performance.

Methods

Study population

MESA (Multi-Ethnic Study of Atherosclerosis) is a community-based prospective cohort study originally designed to study the natural history of subclinical cardiovascular disease. Its methods have been previously described.⁷ In short, MESA enrolled 6,814 men and women ages 45 to 84 years who were free of clinical coronary heart disease at baseline enrollment from 2000 to 2002. Individuals were enrolled at 6 academic U.S. centers in New York, Baltimore, Chicago, Los Angeles, Minnesota, and Winston Salem. All participants provided informed consent, and institutional review board approval was achieved at each participating site. Participants with missing data or no follow-up after baseline were excluded from the analysis (N = 691).

Risk factors

The risk factors included in the PCE equations are age, sex, race (White, Black, or other), smoking status, systolic blood pressure, hypertension treatment status, diabetes status, and total and high-density lipoprotein cholesterol levels (Supplemental Table 1). The PREVENT equations include traditional risk factors as included in the PCE in addition to kidney function (estimated glomerular filtration rate) and the option to tailor the model with urine albumin-to-creatinine ratio or hemoglobin A1c when indicated and available. In addition, the PREVENT calculator has the ability to include social determinants of health via the social deprivation index (SDI) when available.⁸ The SDI is a composite deprivation score of 7 demographic variables collected in the American Community Survey including poverty level, less than high school education, single-parent households, living in rented homes, overcrowding in homes, families without a car, and unemployment rates within a U.S. Postal Service Zip Code area.⁸ The higher the SDI score, the greater the level of social deprivation. We included these expanded risk factors, when available, in calculation of the PREVENT risk score. Furthermore, the PREVENT equations offer the option to calculate risk for participants with missing values for the optional variables. We used this in cases when we did not have urine albumin to creatinine, A1c, or SDI. Of note, hemoglobin A1c was collected at exam 2 (approximately 17-20 months later) for MESA participants; however, it is treated as if collected at baseline for the purpose of this analysis.

Risk score calculators and outcomes

The risk algorithms used in this analysis include the PCE,² the full PREVENT-ASCVD equation, and the full PREVENT-HF equation.⁶ ASCVD events in the PREVENT-ASCVD equation were defined as coronary artery disease (which includes myocardial infarction and fatal coronary disease) and stroke, which mirrors the PCE outcomes. HF outcomes were defined as incident HF and did not specify reduced vs preserved ejection fraction. Events within MESA were collected by periodic follow-up via scheduled follow-up exams and phone calls. Classification of events was based on medical records, autopsy reports, death certificates, or interviews with participants' physicians or family/friends and adjudicated by the Morbidity and Mortality Committee. We assessed the predictive performance in the overall population of MESA and within the following subgroups: sex, race/ethnic groups (White, Black, Hispanic, Chinese), smoking status, hypertension history, kidney function per the Kidney Disease, Improving Global Outcomes (KDIGO) chronic kidney disease (CKD) stages, and SDI.

Statistical analysis

Descriptive statistics collected at baseline enrollment are reported as median with interquartile ranges for continuous variables and number with percentages for categorical variables. To compare the number of observed events with the number of expected events as predicted by PCE and PREVENT, we followed previous studies and scaled the risk scores based on the length of follow-up (for those with <10 years of follow-up) using an exponential survival function. Discordance between predicted and observed event percentages was calculated by comparing the absolute difference between event percentages as a percentage of the observed event rate [(predicted % − observed %)/observed %) × 100]. To assess the discrimination ability of PCE and PREVENT equations, we reported Concordance statistics (C-statistics). To evaluate the accuracy of predictions, we stratified the PCE and PREVENT estimated risk scores into 10 groups (deciles) and calculated the cumulative risk of ASCVD and HF over 10 years using the Kaplan-Meier method for each group. We then plotted and compared the average PCE and PREVENT risk predictions with the Kaplan-Meier estimated cumulative risk for each group. In addition, smooth calibration plots were presented by using flexible adaptive hazard regression approach to relate PCE- and PREVENT-estimated ASCVD and HF probability at 10 years to observed event percentages at 10 years. Mean absolute errors (MAEs) were also reported.

To examine how the PCE and PREVENT equations classify ASCVD risk, we presented risk reclassification tables (stratified by ASCVD status) between PCE and PREVENT using current clinical cutoffs of <5%, 5% to 7.5%, 7.5% to 20%, and >20% risk. We used a risk of >7.5% as the cutoff for statin eligibility based on current U.S. guidelines. Finally, we evaluated these risk scores stratified by sex, race, smoking status, history of hypertension, KDIGO CKD stages, and SDI.

Results

Study population

There were 6,814 individuals enrolled in MESA at baseline; 691 of which were excluded due to missing or extreme values that were outside the range of the risk calculators, and 20 were excluded for missing event outcome data or time, leaving 6,098 participants included in this analysis. Baseline characteristics can be found in Table 1. Women were 52% of our study population, and the average age at enrollment was 61 years. Racial/ethnic distribution of our population includes 39% White, 27% Black, 22% Hispanic, and 12% Chinese participants. Average SDI was 7.5 across the entire cohort.

Table 1.

Baseline Characteristics for Included Participants Within the Multi-Ethnic Study of Atherosclerosis (MESA)

	Total (N = 6,098)	White (n = 2,351)	Chinese (n = 730)	Black (n = 1,660)	Hispanic (n = 1,357)
Age (y)	61 (9.6)	62 (9.6)	62 (9.7)	61 (9.5)	60 (9.6)
Female	3,187 (52%)	1,208 (51%)	368 (50%)	918 (55%)	696 (51%)
Diabetes	1,594 (26%)	394 (17%)	214 (29%)	537 (32%)	449 (33%)
Family history of heart attack	2,459 (40%)	1,146 (49%)	132 (18%)	663 (40%)	518 (38%)
Lipid-lowering medications	982 (16%)	432 (18%)	99 (14%)	275 (17%)	176 (13%)
Hypertension medications	2,223 (37%)	756 (32%)	202 (28%)	824 (50%)	441 (33%)
Current smoking	795 (13%)	272 (12%)	44 (6%)	298 (18%)	181 (13%)
Systolic blood pressure (mm Hg)	126.4 (20.2)	123.6 (19.1)	124.3 (19.9)	131.4 (20.6)	126.1 (20.5)
Total cholesterol (mg/dL)	195.4 (32.3)	196.7 (31.9)	193.9 (30.8)	191.5 (32.1)	198.5 (33.6)
HDL cholesterol (mg/dL)	50.6 (13.8)	51.8 (14.7)	49.3 (12)	52 (13.9)	47.3 (12.3)
SDI score	7.5 (2.6)	6.6 (2.5)	6.5 (2.8)	8.5 (2.2)	8.4 (2.2)
eGFR (mL/min/1.73 m2)	78.4 (15.7)	74.5 (13.8)	80.3 (15.1)	81.0 (17.4)	80.9 (15.7)
Urine albumin to creatinine ratio (mg/g)	24.8 (150.1)	12.6 (50.3)	31.7 (174.3)	26.8 (130.7)	39.9 (243.5)
Hemoglobin a1c	5.7 (1.0)	5.4 (0.6)	5.8 (0.9)	5.9 (1.1)	5.9 (1.2)

Categorical variables listed as n (%). Continuous variables listed as mean (SD).

eGFR = estimated glomerular filtration rate; HDL = high-density lipoprotein; SDI = social deprivation index.

Overall performance of the PREVENT-ASCVD equations vs PCE

Figure 1 displays the predicted and observed events for the PREVENT equations as compared to the guideline-directed PCE. The PCE estimated an event rate of 10.8%, compared to 5.7% by the PREVENT equation, whereas we observed a 6.0% event rate in this population. There was an absolute risk difference of 4.8% with the PCE prediction vs −0.3% difference in PREVENT (Table 2), which portended an 79.8% relative risk overestimation in PCE vs a 5.5% relative underestimation by PREVENT. Discrimination was the same between both risk calculators, c-statistic 0.73 PCE and PREVENT. Calibration was improved in PREVENT as demonstrated by an MAE 0.001 compared to 0.058 with PCE. Within PCE, there is a greater risk of overestimation at higher risk deciles, whereas PREVENT performs more consistently across the risk spectrum (Figure 2).

Figure 1.

PREVENT Risk Score Performance vs Pooled Cohort Equation

Ten-year ASCVD event percentages as predicted by the PCE and PREVENT-ASCVD equations and observed events in the Multi-Ethnic Study of Atherosclerosis (MESA). ASCVD = atherosclerotic cardiovascular disease; PCE = Pooled Cohort Equations.

Table 2.

PREVENT Risk Score vs PCE in Predicting 10-Year ASCVD Events

	Number	Predicted Events	Observed Events	Absolute Difference	Discordance (%)a	C-Statistic	Mean Absolute Error (MAE)
Total
PCE	6,098	658 (10.8%)	366 (6%)	4.8	79.8	0.73	0.058
PREVENT		346 (5.7%)	366 (6%)	−0.3	−5.5	0.73	0.001
White
PCE	2,351	237 (10.1%)	138 (5.9%)	4.2	71.7	0.71	0.052
PREVENT		121 (5.1%)	138 (5.9%)	−0.8	−12.3	0.71	0.005
Black
PCE	1,660	201 (12.1%)	111 (6.7%)	5.4	81.1	0.74	0.064
PREVENT		102 (6.1%)	111 (6.7%)	−0.6	−8.1	0.73	0.004
Hispanic
PCE	1,357	142 (10.5%)	92 (6.8%)	3.7	54.3	0.76	0.049
PREVENT		80 (5.9%)	92 (6.8%)	−0.9	−13.0	0.74	0.006
Chinese
PCE	730	78 (10.7%)	25 (3.4%)	7.3	212.0	0.74	0.088
PREVENT		43 (5.9%)	25 (3.4%)	2.5	72.0	0.74	0.029

The table shows 10-year ASCVD event percentages as predicted by the PCE and PREVENT equations and observed events in MESA, stratified by race/ethnicity.

ASCVD = atherosclerotic cardiovascular disease; PCE = pooled cohort equations; PREVENT = Predicting Risk of Cardiovascular Disease Events.

^aDiscordance is defined as [(predicted % − observed %)/observed %) × 100].

Figure 2.

Calibration of PREVENT-ASCVD and PCE

Calibration of 10-year predicted ASCVD risk with the PCE (A) and the PREVENT-ASCVD (B) equation compared to observed event rate displayed as deciles of predicted risk. The dotted line represents a perfectly calibrated model, and the solid line represents the calibration of our models within MESA.

Performance of the PREVENT-ASCVD equations in subgroups

PCE and PREVENT performances were assessed in the following subgroups: sex, smoking status, history of hypertension, KDIGO CKD stage, SDI, and race/ethnicity groups (Figure 3). Overall, the PREVENT-predicted event percentages are more similar to observed event percentages in all specified subgroups, with the exception of individuals who smoke. The PREVENT-ASCVD equation’s predicted event rate was most similar to the observed event rate in women (0.2% absolute risk difference and relative 3.3% risk overestimation), nonsmokers (0.2% absolute risk difference and 2.4% relative risk overestimation), individuals with CKD stages 3 and 4 (absolute risk difference 0.3%, corresponding to 3.2% relative risk overestimation), and those who are most disadvantaged with SDIs of 7 to 10 (absolute risk difference −0.3%, with −5.0% absolute risk underestimation). In all these subgroups, the PCE overestimated relative risk by 80.1% to 93.5%. Notably PREVENT underestimated risk in men (−0.9% absolute risk difference, −11.6% relative risk underestimation), current smokers (−3.4% absolute risk difference, −37.0% relative risk underestimation), and those with a history of hypertension (−1.2% absolute risk difference, −13.8% relative risk underestimation). Discrimination was similar amongst all subgroups between PCE and PREVENT (Supplemental Table 2). Calibration for all subgroups is found in Supplemental Figures 1 to 6. The PREVENT-ASCVD prediction models had improved calibration as compared to the PCE prediction models in all subgroups analyzed.

Figure 3.

Subgroup Performance of PREVENT Risk Score vs PCE

Ten-year ASCVD event rates as predicted by the PCE and PREVENT-ASCVD equations and observed events in the Multi-Ethnic Study of Atherosclerosis (MESA) stratified by sex (A), current smoking status (B), history of hypertension (C), race/ethnicity (D), chronic kidney disease (CKD) stages (E), and social deprivation index (SDI) (F).

ASCVD risk reclassification

Risk classification using current U.S. guideline cutoffs for PCE and PREVENT prediction is shown in Table 3, stratified by those who did not and did go on to have an ASCVD event within 10 years. Overall, 2,561 (42.0%) participants were reclassified from a higher-risk category to lower risk, and 348 (5.7%) were reclassified from lower to higher risk category using the PREVENT equation as compared to the PCE equation. Of the 2,561 participants reclassified from a higher to a lower risk category, 96.8% did not have an ASCVD event at 10 years. In contrast, of the 348 (5.7%) participants who were reclassified from a lower to higher risk category, 102 (29.3%) had an ASCVD.

Table 3.

ASCVD Risk Reclassification From PCE to PREVENT Risk Score

	Predicted Risk in Individuals Who Did Not Have an ASCVD Event Within 10 y
	PREVENT Predicted Risk
	0%-5%	5%-7.5%	7.5%-20%	>20%	Reclassified to Higher Risk
PCE predicted risk
0%-5%	1,972	93	67	4	246 (4.3%)
Quality of evidence and recommendations
5%-7.5%	550	134	63	3
7.5%-20%	435	664	800	16
>20%	2	34	793	102
Reclassified to lower risk	2,478 (43.2%)

	Predicted Risk in Individuals Who Had an ASCVD Event
	PREVENT Predicted Risk
	0%-5%	5%-7.5%	7.5%-20%	>20%	Reclassified to Higher Risk
PCE predicted risk
0%-5%	48	22	42	4	102 (27.9%)
5%-7.5%	10	18	22	1
7.5%-20%	7	20	103	11
>20%	1	2	43	12
Reclassified to lower risk	83 (22.7%)

Number of individuals in each risk category defined as low risk (<5% risk of ASCVD event in 10 years), borderline risk (5%-7.5%), intermediate risk (7.5%-<20%), and high risk (>20%) according to the PCE and re-classified by PREVENT equations. Numbers highlighted in blue represent the individuals whose risk was reclassified from a higher risk category with PCE prediction to lower risk category with PREVENT. Numbers highlighted in italic represent individuals whose risk was reclassified from lower risk to higher risk category when reclassified by PREVENT. The boxes colored bold represent concordant risk classifications between PCE and PREVENT predictions.

Among all subgroups assessed, there was a general reclassification to lower risk from higher risk when using the PREVENT equation as compared to the PCE equation. Of the 1,561 men reclassified from higher to lower risk, 96.4% of them did not have an ASCVD event at 10 years, similar to 97.3% of women. Among smokers, 94.1% of current smokers and 97.3% of nonsmokers reclassified to lower risk did not have an ASCVD event at 10 years. Among those with hypertension, 95.8% with hypertension and 98.1% of those without hypertension that were reclassified to a lower risk did not have an ASCVD event at 10 years of follow-up. Across the KGIDGO CKD stages, 95.6% to 97.0% of participants that were reclassified to a lower risk with PREVENT-ASCVD were event-free. Across SDI ranges, 96.2% to 97.7% of individuals did not go on to have an event despite being reclassified to lower risk with PREVENT-ASCVD. Across the various racial/ethnic groups, between 1% and 4% of those reclassified to a lower risk category went on to have an ASCVD event.

Performance of PREVENT-HF

Within MESA, 195 HF events were adjudicated over 10 years of follow-up compared with 317 events predicted with the PREVENT-HF equations, a 2.1% absolute risk difference, which is a 62.6% relative risk overestimation. Across racial/ethnic groups in MESA, the highest observed event rate (4.3%) and lowest relative discordance (48.6%) was seen in Black individuals (4.3%), and the lowest observed event rate (1.5%) and greatest relative risk discordance (345.5%) was observed in Chinese participants. With regards to discrimination, the PREVENT-HF equation had a C-statistic of 0.77 within MESA (Supplemental Table 3). Calibration organized by deciles is represented in Supplemental Figure 7, and the MAE was 0.026. HF risk was overpredicted in every decile, with greater overprediction at higher predicted event percentages.

Performance of the PREVENT-HF equations in subgroups

Supplemental Table 4 details the performance of the PREVENT-HF equations in multiple subgroups. The predicted percentages of incident HF compared to observed event percentages were most similar in men (1.6% absolute risk difference, 37.7% relative risk overestimation), smokers (0.8% absolute risk difference, 17.1% relative risk overestimation), those with hypertension (2.6% absolute risk difference, 47.3% relative risk overestimation), those in CKD stages 3 and 4 (2.6% absolute risk difference, 32.1% relative risk overestimation), and individuals in the middle range of SDI, 4 to 6 (−0.9% absolute risk difference, 16.9% relative risk underestimation). Model discrimination was best in individuals with the least amount of social disadvantage, SDI 1 to 3 (C-statistic 0.8). Calibration was best in smokers (MAE 0.019), those classified as CKD stage 1 (MAE 0.015), and those at middle levels of SDI 4 to 6 (MAE 0.016).

Discussion

New diagnostic tools should be evaluated against current practice standards before adoption into clinical practice. PREVENT performance was appropriately compared to the PCE during the initial model derivation; however, the correlation between predicted and observed event rate, particularly by various subgroups and the clinical implications of reclassification, had not been fully delineated. Within the MESA cohort, we found that the 2023 AHA PREVENT-ASCVD equations more accurately predicted observed event percentages and had similar discrimination when compared to the current guideline-recommended 2013 ACC/AHA PCE equations. This is one of the first studies to analyze the performance of the PREVENT-HF equations, and we demonstrated that the performance of PREVENT-HF risk score was fair but significantly overpredicted incident HF events (Central Illustration). Furthermore, we found that PREVENT-ASCVD more accurately predicted ASCVD events than the PREVENT-HF equation predicted incident HF events within MESA.

Central Illustration.

PREVENT Risk Scores Performance in MESA

(A) PREVENT-ASCVD risk score vs the Pooled Cohort Equation (PCE) in prediction of 10-year ASCVD events. (B) Ten-year predicted HF events per the PREVENT-HF risk score compared with 10-year observed event percentages in MESA. (C) Risk reclassification within MESA using the PREVENT-ASCVD risk score. HF = heart failure.

Among specified subgroups, there was minimal discordance between PREVENT-ASCVD-predicted percentages and observed event percentages in women, nonsmokers, CKD stages 2 to 4, SDI scores 4 and above, and in Black individuals. This was in contrast to the PCE, which we found to overestimate risk by 80% to 96% among these subgroups in this analysis of a sex-balanced, multiethnic population. The PREVENT equations modestly underestimate risk in men, current smokers, history of hypertension, those classified as CKD stage 1 or 2, and higher SDI scores of 7 and above by 5% to 37%. This contrasts with PCE overestimation of 34.2% to 85.0% for these subgroups of participants. Among the racial groups evaluated, the PREVENT modestly underestimated risk in White, Black, and Hispanic individuals by 8.1% to 13.0%. Among these subgroups, there are historically higher risk groups that will have risk underestimation by the PREVENT score if incorporated into clinical practice. This may lead to harmful risk underestimation and minimization of primary prevention strategies that can lead these already higher-risk groups to experience ASCVD events at higher rates. Diligent understanding of PREVENT performance on a patient-based level prior to complete incorporation will be imperative to understand limitations of this new score and limit patient harm. Furthermore, risk was more significantly overestimated in Chinese individuals in both the PCE and PREVENT equations, 212% and 72%, respectfully. Poor performance in this racial group is consistent with prior analyses of the PCE and other risk-prediction tools and a historic limitation of risk prediction in this population.^3,9 Across the overall cohort and specified subgroups analyzed, there was similar discrimination using C-statistic between PCE and PREVENT. Calibration was improved overall and across subgroups using updated PREVENT vs PCE prediction. PREVENT-HF performed best with men, smokers, those with a history of hypertension, individuals with CKD stages 3/4 and SDI 4 to 6, and White and Black individuals compared to Hispanic and Chinese participants.

The PCE has been shown to overestimate patients' risk by up to 60% to 90% when replicated in modern, diverse cohorts.^3,10,11 Our findings are similar to other cohorts who have recently evaluated comparative performance of the PREVENT equations and PCE.12, 13, 14 As such, the PREVENT equations were a timely and necessary upgrade to our available ASCVD risk-prediction tools. These equations included more modern populations, which reflected current practice patterns and provided an updated risk-prediction algorithm for clinicians to use.⁶

Studies from the National Health and Nutrition Examination Surveys (NHANES) have shown that applying the PREVENT-ASCVD equations could reduce eligibility for antihypertensives and statin medications for 15.8 million U.S. adults, which could result in 107,000 additional myocardial infarctions or stroke with current guideline-recommended cutoffs.^13,15 Similarly in our work, there was an overall trend of reclassification from a higher risk category to a lower risk category, which would affect statin eligibility in 19% of this population who were previously statin-eligible according the most recent ACC/AHA primary prevention guidelines using PCE risk stratification.¹⁶ Our analysis suggests that reclassification was generally appropriate with only 83 (3.2%) people who were reclassified from a higher risk to lower risk category went on to have an ASCVD event. Of those, only 30 individuals (81.2% of all those reclassified to lower risk) would have fallen below the 7.5% 10-year ASCVD risk cutoff proposed by ACC/AHA guidelines to consider statin therapy. Appropriate risk reclassification was also observed when stratified by various subgroups. Among the subgroups assessed, there were 1.1% to 5.9% of individuals reclassified to lower risk with the PREVENT equations who later went on to have an ASCVD event. Despite reliability reclassifying the majority of individuals to a lower risk categories, PREVENT had similar performance in predicting cardiovascular death when compared to PCE.¹⁷ Our data suggest that PREVENT reclassification will prove to be appropriate for patients in the future; however, it will be imperative to monitor primary prevention usage, event percentages, and cardiovascular mortality in coming years, particularly if the guidelines are updated to recommend risk stratification with this new risk score.

Beyond ASCVD, the PREVENT equations can also predict risk of HF events. The most recent 2022 HF guidelines recognize Stage A HF, which identified those at risk of developing HF who should be targeted for primary prevention.¹⁸ Prior prediction models to predict incident HF were developed in cohorts of individuals with coronary artery disease, who are inherently high risk for HF.¹⁹ The PREVENT-HF is among the first HF risk-prediction scores designed for use in a multi-ethnic population of men and women without known ASCVD. We found that this novel HF prediction tool significantly overestimated events in this cohort, particularly with Chinese individuals. This is one of the first validation studies including the PREVENT-HF model, and thus, further study is required in additional cohorts to elucidate its clinical accuracy. Additional modification may be required to tailor this algorithm to different patient populations.

While the authors of the PREVENT equations noted a lack of diversity in the derivation cohorts, a particular strength of our study is the racial and ethnic diversity and our demonstration of risk-prediction performance in these various groups.⁶ In our study population, we included 27% Black, 22% Hispanic, and 12% Chinese participants, which are higher proportions than those in similar studies evaluating PREVENT performance.¹² An important update of the PREVENT equations was the removal of the race component from the PCE. We found that risk prediction with the PREVENT equations was similar across racial/ethnic group studies and performed better across all racial/ethnic groups than the PCE. The exception is prediction within Chinese populations, which continues to be an area of concern for modern U.S. risk-prediction scores with both the PREVENT and PCE. MESA also collected participant-specific zip codes, which allowed incorporation of SDI in our analysis. We found that the PREVENT score performs well across the spectrum of SDI, particularly best in the higher SDI ranges; this will allow identification and incorporation of social determinants of health to augment risk prediction in vulnerable populations. Furthermore, this is one of the first validation reports of the PREVENT-HF performance comparing predicted event percentages to observed events percentages, whereas previous analyses have focused primarily on PREVENT-ASCVD performance.

Study Limitations

There are several limitations to note. The MESA cohort was used to derive the PREVENT equations, and thus, the participants contributed directly to the derivation, which may contribute to overfitting of model performance. This type of study should be replicated in an independent cohort that was not included in the PREVENT derivation to ensure external validity. In addition, MESA enrolled individuals at 6 academic centers, which may reflect a different population than those seen in rural or community practices. Furthermore, we used hemoglobin A1c from exam 2 to represent glucose control at baseline for individuals, as it was not measured at enrollment. Lastly, with regards to smoking exposure, there are differences in level of exposure by race/ethnicity, such than African Americans on average have fewer pack-years than White Americans, which may result in apparent heterogeneity.²⁰

Conclusions

In this modern, multi-ethnic cohort, the PREVENT equations portend a more accurate ASCVD risk prediction than the current guideline-recommended PCE. It performs best in women, nonsmokers, individuals with kidney disease, those with higher levels of social deprivation, and Black individuals. The PREVENT-HF equations also have fair performance in predicting HF events but inferior to the performance of PREVENT-ASCVD in predicting HF vs ASCVD events. These findings support adoption of this updated prediction tool.

Perspectives.

COMPETENCY IN MEDICAL KNOWLEDGE AND PATIENT CARE: With regards to medical knowledge, we have demonstrated the performance of the PREVENT risk-prediction model, which gives medical professionals an understanding of how these scores apply to their everyday patients. Caregivers can utilize this knowledge and apply it to patient care in everyday practice when seeing patients at risk of ASCVD and HF to guide management strategies and prevention.

TRANSITIONAL OUTLOOK: In our work, we demonstrated that the PREVENT-ASCVD risk-prediction model performs better than the current guideline-recommended standard PCE. Our study is in alignment with prior studies that demonstrate superior performance of this new score. Despite this, before implementation into guideline recommendations for prevention of ASCVD, we highlight the need for further study on underrepresented groups in medicine, as well as elucidating risk-score performance among subgroups not identified in our analysis. Furthermore, now that there will be a significant risk reclassification of individuals with regards to ASCVD risk, we will need to further investigate whether the predetermined cutoff still remains clinically relevant or warrants readjustment.

Funding support and author disclosures

This research was supported by contracts 75N92020D00001, HHSN268201500003I, N01-HC-95159, 75N92020D00005, N01-HC-95160, 75N92020D00002, N01-HC-95161, 75N92020D00003, N01-HC-95162, 75N92020D00006, N01-HC-95163, 75N92020D00004, N01-HC-95164, 75N92020D00007, N01-HC-95165, N01-HC-95166, N01-HC-95167, N01-HC-95168, and N01-HC-95169 from the National Heart, Lung, and Blood Institute, and by grants UL1-TR-000040, UL1-TR-001079, and UL1-TR-001420 from the National Center for Advancing Translational Sciences (NCATS). The authors have reported that they have no relationships relevant to the contents of this paper to disclose.