Comparison of Short‐ and Long‐Term Cardiovascular Disease Risk Assessment Tools in US Young Adults

Abstract

Background

In 2023, the American Heart Association published the Predicting Risk of Cardiovascular Disease Events (PREVENT) equations for estimating atherosclerotic cardiovascular disease (ASCVD) risk in adults aged 30 to 79 years. We compared PREVENT's performance with existing US guideline recommended models—Pooled Cohort Equations for 10‐year ASCVD risk and FHS (Framingham Heart Study) equations for 30‐year ASCVD risk—among young adults.

Methods

We analyzed adults aged 20 to 39 years without baseline ASCVD from 2 sources: (1) pooled data from 2 large epidemiologic cohorts (CARDIA [Coronary Artery Risk Development in Young Adults] and FHS, n=7763), and (2) electronic health records from Kaiser Permanente Southern California (n=266 378). Incident ASCVD events were defined as nonfatal myocardial infarction, coronary heart disease death, and fatal/nonfatal stroke at 10 and 30 years.

Results

PREVENT improved 10‐year risk discrimination over Pooled Cohort Equations in both the epidemiologic cohorts (∆Harrell's C, 0.057 [95% CI, 0.013–0.101]) and Kaiser Permanente Southern California (∆Harrell's C=0.041 [95% CI, 0.034–0.049]). The Pooled Cohort Equations overestimated 10‐year risk (mean calibration 3.26 in the epidemiologic cohorts; 1.73 in Kaiser Permanente Southern California), whereas PREVENT was well calibrated in the epidemiologic cohorts (1.10 [95% CI, 0.83–1.73) but underestimated risk in KPSC (0.91 [95% CI, 0.86–0.96]), particularly among Black individuals (0.54 [95% CI, 0.48–0.61]). For 30‐year risk, PREVENT and FHS had similar discrimination, but PREVENT underestimated 30‐year risk (mean calibration 0.63) whereas FHS had good calibration (mean calibration 0.90 to 0.99]).

Conclusions

PREVENT may be a better tool for short‐term ASCVD risk assessment in young adults than the PCEs, whereas the FHS equations may be better for long‐term risk assessment than PREVENT in this age group.

Nonstandard Abbreviations and Acronyms

CARDIA

Coronary Artery Risk Development in Young Adults

FHS

Framingham Heart Study

PCEs

Pooled Cohort Equations

PREVENT

Predicting Risk of Cardiovascular Disease Events

Clinical Perspective.

What Is New?

• In this analysis of >270 000 US young adults aged 20 to 39 years from large diverse populations, the newly developed Predicting Risk of Cardiovascular Disease Events (PREVENT) equations demonstrated improved discrimination and calibration for predicting 10‐year atherosclerotic cardiovascular disease risk compared with the Pooled Cohort Equations.

• For 30‐year risk prediction, although PREVENT showed similar discrimination to the FHS (Framingham Heart Study) equations, PREVENT underestimated 30‐year risk whereas the FHS equations had good calibration.

What Are the Clinical Implications?

• These findings suggest that PREVENT may be a better tool for short‐term atherosclerotic cardiovascular disease risk assessment in young adults than the PCEs whereas the FHS equations may be better for long‐term risk assessment than PREVENT in this age group.

The overall rates of atherosclerotic cardiovascular disease (ASCVD) have declined over the past 2 decades in the United States; however, this favorable trend has not extended to young adults aged 20 to 39 years. ¹ , ² , ³ , ⁴ , ⁵ Early identification of young adults at high‐risk of developing ASCVD is of critical importance to prevent premature morbidity and mortality, reduce health care costs, and ultimately contribute to a healthier and more productive population. ⁶ , ⁷ ASCVD risk assessment is at the center of decision‐making to guide ASCVD primary prevention efforts and is recommended by multisociety guidelines. ⁸ , ⁹ Both the 2018 American Heart Association/American College of Cardiology cholesterol guideline and the 2019 cardiovascular disease primary prevention guideline recommend using the Pooled Cohort Equations (PCEs) to estimate 10‐year ASCVD risk. ⁸ , ⁹ , ¹⁰ However, the PCEs may not be directly applicable to young adults because they were derived in and recommended for adults 40 to 79 years of age. ¹⁰ The recently developed American Heart Association's Predicting Risk of Cardiovascular Disease Events (PREVENT) equations expanded the age range to include US adults aged 30 to 79 years. ¹¹ However, it remains unclear how the PREVENT equations compare to the PCEs in predicting 10‐year ASCVD risk in young adults.

Furthermore, because ASCVD is a cumulative, lifelong process, young adults with adverse risk factors may have low short‐term but high long‐term ASCVD risks. ¹² , ¹³ , ¹⁴ To address this issue, the 2018 American Heart Association/American College of Cardiology cholesterol guideline recommends considering the estimation of 30‐year or lifetime ASCVD risk for young adults. ¹⁵ The FHS (Framingham Heart Study) 30‐year ASCVD risk prediction equations were derived from a single cohort of White individuals aged 20 to 59 years, which may limit its applicability to other populations and has been identified as a limitation in the 2018 cholesterol guideline. ⁸ , ¹² The PREVENT equations also included models for estimating 30‐year ASCVD risk. ¹¹ However, considering that the PREVENT equations were derived based on data with a mean follow‐up of only 4.8 years, their prediction performance on long‐term ASCVD risk compared with the FHS equations remains unknown, particularly among racial and ethnic minority groups.

This study sought to compare the performance of the PCEs and the PREVENT equations in predicting 10‐year ASCVD risk in young adults, as well as to compare the performance of the FHS equations and the PREVENT equations in predicting 30‐year ASCVD risk. We analyzed data from 2 prospective cohort studies as well as from Kaiser Permanente Southern California (KPSC), a large US integrated health care system.

METHODS

Data Availability Statement

Because of the sensitive nature of the data collected for this study, access to the KPSC data set is available to qualified researchers trained in human subject confidentiality protocols. Requests may be submitted via the “Contact Us” form (https://www.kpscalresearch.org/aboutus/contact‐us/). Requests to access data from the CARDIA (Coronary Artery Risk Development in Young Adults) study and FHS from qualified researchers can be sent to the coordinating center of each cohort (CARDIA: coc@uab.edu; FHS: fhs@bu.edu).

Study Design and Cohorts

The present analysis was based on 2 complementary data sources: (1) pooled data from 2 large prospective cohort studies: CARDIA ¹⁶ and FHS (including data from the Offspring, Third Generation, Omni 1, and Omni 2 cohorts), ¹⁷ , ¹⁸ , ¹⁹ and (2) electronic health records from KPSC, a large US integrated health care system serving a diverse population. KPSC's comprehensive electronic health records integrate care provided to their members, including medical appointments, procedures, pharmacy, and laboratory services. Details of the design of each study are reported in Data S1. All study protocols were approved by the institutional review boards at participating institutions. All participants in the epidemiologic cohorts provided written informed consent, and informed consent for the KPSC population was waived given the retrospective nature of the study.

The current analysis included young adults aged 20 to 39 years and without a history of ASCVD at baseline (Figures S1 and S2). In the epidemiologic cohorts, the baseline visit was defined as the first visit that participants met this inclusion criteria. In the KPSC, the baseline visit was defined as the first day between January 1, 2008, and December 31, 2009, on which individuals met the inclusion criteria and had at least 6 months of continuous eligibility as a KPSC member. For both the epidemiologic cohorts and KPSC, we excluded individuals who had any missing values in risk predictions used in the 10‐year or 30‐year ASCVD risk prediction equations, or with missing follow‐up for ASCVD events.

Clinical Data Collection

Demographic characteristics, lipids, blood pressure (BP), and other ASCVD risk factors were assessed following standardized protocols in each study. Estimated glomerular filtration rate was calculated using the Chronic Kidney Disease Epidemiology Collaboration 2021 creatinine equation. ²⁰ Consistent with the development of the PREVENT equations, individuals with extreme values for systolic BP (SBP <90 or >200 mm Hg), total cholesterol (<130 or >320 mg/dL), high‐density lipoprotein cholesterol (HDL‐C <20 or >100 mg/dL), estimated glomerular filtration rate (<15 or >140 mL/min per 1.73 m²), or body mass index (BMI <18.5 or ≥40.0 kg/m²) were excluded from the analysis. ¹¹

ASCVD Events

The primary outcome of interest was incident ASCVD defined as nonfatal myocardial infarction (MI), fatal coronary heart disease (CHD), or fatal or nonfatal stroke, consistent with the definition used in the original derivation of the PCEs, PREVENT, and FHS equations. In the epidemiologic cohorts, events were ascertained and adjudicated within each study following a specific protocol. The diagnosis of an MI required at least 2 of the following: symptoms indicative of myocardial ischemia, ECG, or other imaging abnormalities consistent with MI, and a rising or falling pattern of cardiac biomarkers over at least 6 hours with a peak above the upper limit of normal. The diagnosis of stroke required a persistent central neurologic deficit lasting >24 hours or brain imaging consistent with acute stroke.

In KPSC, we used previously validated algorithms with high positive predictive values (>90%) to identify MI and stroke. ²¹ , ²² , ²³ , ²⁴ Specifically, MI was identified by hospital discharge diagnoses with International Classification of Diseases, Ninth Revision (ICD‐9) codes 410.x0, 410.x1 and Tenth Revision (ICD‐10) codes I21.x, ²¹ and stroke by hospital discharge diagnoses with ICD‐9 codes 430, 431, 432.x, 433.x1, 434.x1, 436.xx and ICD‐10 codes I60.x, I61.x, I62.x, I63.x, G46.3, G46.4. ²² , ²³ , ²⁴ Deaths from CHD and stroke were identified using ICD‐10 codes (I20–I25, I60–I69) from KPSC sources and national death index files.

Eligible individuals were followed from study entry (defined as the first eligible visit in the epidemiologic cohorts and in KPSC) for an incident ASCVD event. Follow‐up was censored at 10 years when assessing 10‐year risk prediction and censored at 30 years when assessing 30‐year risk prediction. Given that individuals in KPSC had a maximum of 15‐years of follow‐up, KPSC data were used only for evaluating the 10‐year risk. Follow‐up time was also censored at death, end of study, loss to follow‐up, or disenrollment (for KPSC members), whichever occurred first.

Additionally, in sensitivity analysis, we examined an alternative definition of ASCVD as nonfatal MI, fatal CHD, or fatal and nonfatal ischemic stroke, rather than total stroke as used in the main analysis.

Short‐Term and Long‐Term ASCVD Risk Prediction

To estimate short‐term ASCVD risk, we used both the PCEs and the PREVENT equations to predict 10‐year ASCVD risk. ¹⁰ , ¹¹ The PCEs are race‐ and sex‐specific equations incorporating age, smoking, SBP, total cholesterol, HDL‐C, antihypertensive medication use, and diabetes. ¹⁰ The PREVENT 10‐year ASCVD equations are sex specific and incorporate age, smoking, SBP, total cholesterol, HDL‐C, antihypertensive medication use, statin use, diabetes, and estimated glomerular filtration rate. ¹¹

To estimate long‐term ASCVD risk, we used the FHS equations and the PREVENT equations to predict 30‐year ASCVD risk. ¹¹ , ¹² There are 2 versions of the FHS equations for 30‐year ASCVD risk: one equation included age, smoking, SBP, antihypertensive medication use, diabetes, total cholesterol and HDL‐C (hereinafter referred to as “FHS with lipid”); and the other equation included age, smoking, SBP, antihypertensive medication use, diabetes, and BMI instead of cholesterol levels (hereinafter referred to as “FHS with BMI”). ¹² The PREVENT 30‐year ASCVD equations are sex specific and included age, smoking, SBP, total cholesterol, HDL‐C, antihypertensive medication use, diabetes, and estimated glomerular filtration rate. ¹¹

Because the PCEs are applicable only to individuals aged 40 to 75 years, we replaced age to 40 years for all participants when calculating 10‐year ASCVD risk using the PCEs. ²⁵ Similarly, because the PREVENT equations are applicable only to individuals aged 30 to 79 years, we replaced age to 30 years when age was <30 years when calculating 10‐year and 30‐year risks using the PREVENT equations. Given that the FHS equations were developed in individuals aged 20 to 59 years, the actual age was used when calculating 30‐year risks using the FHS equations.

Statistical Analysis

All analyses were performed in the epidemiologic cohorts and in KPSC separately. Individual characteristics at baseline were described for the epidemiologic cohorts and for KPSC. We assessed the distributions of predicted 10‐year and 30‐year ASCVD risks by each risk assessment tool for the overall population and by age (20–29 years, 30–39 years), sex (female, male), and racial and ethnic subgroups (non‐Hispanic White, non‐Hispanic Black, Hispanic, and Asian/Pacific Islander).

To compare the performance of the 10‐year and 30‐year risk equations in young adults, discrimination was assessed by estimating Harrell's C‐statistics, which quantifies the concordance between predicted risks and observed survival times. ²⁶ Calibration was assessed by mean calibration (also known as calibration‐in‐the‐large), which measures systematic over‐ or underprediction by comparing the mean predicted 10‐year or 30‐year ASCVD risk to the observed cumulative incidence. Observed ASCVD cumulative incidence was estimated using the nonparametric cumulative incidence function via the “cuminc” function in the “cmprsk” R package, which accounts for the competing risk of non‐CVD death. Calibration was also assessed by plotting observed versus predicted ASCVD risk across deciles of predicted risk. We used nonparametric bootstrapping to estimate the 95% CIs for Harrell's C, the difference in Harrell's C, and mean calibration. A 2‐sided P<0.05 determined statistical significance. Analyses were performed using R version 4.0.2 (Vienna, Austria) and SAS version 9.4 (SAS Institute Inc, Cary, NC).

RESULTS

We included 7736 young adults from the epidemiologic cohorts (4537 from CARDIA and 3199 from FHS) and 266 378 young adults from KPSC (Table 1 and Table S1). In the epidemiologic cohorts, mean±SD age at baseline was 29.2 (5.4) years, 53.1% were women, and 29.8% self‐identified as Black. In KPSC, mean±SD age at baseline was 31.6 (5.4) years, 60.1% were women, 8.0% were non‐Hispanic Black, and 46.7% were Hispanic. Compared with individuals included in the final KPSC sample, those excluded due to incomplete data were more likely to be younger, male, never smokers, without diabetes, have lower total cholesterol, and not use lipid‐lowering or antihypertensive medications (Table S2).

Table 1.

Baseline Characteristics of Study Population

Characteristics	Epidemiologic cohorts (N=7736)	KPSC (N=266 378)
Age, y, mean±SD	29.2±5.4	31.6±5.4
Sex, no. (%)
Female	4104 (53.1)	160 075 (60.1)
Male	3632 (46.9)	106 303 (39.9)
Race and ethnicity, no. (%)
Asian/Pacific Islander	41 (0.5)	28 433 (10.7)
Hispanic	83 (1.1)	124 431 (46.7)
Non‐Hispanic Black	2307 (29.8)	21 296 (8.0)
Non‐Hispanic White	5297 (68.5)	68 020 (25.5)
Other*	8 (0.1)	24 198 (9.1)
Smoking status, no. (%)
Never	4226 (54.6)	222 241 (83.4)
Former	1367 (17.7)	22 725 (8.5)
Current	2143 (27.7)	21 412 (8.0)
Body mass index, kg/m2, mean±SD	25.1±4.3	28.0±5.1
Diabetes, no. (%)	63 (0.8)	9205 (3.5)
Cholesterol, mg/dL, mean±SD
Total cholesterol	182.4±31.2	188.9±33.6
High‐density lipoprotein cholesterol	52.81±3.5	50.81±3.0
Systolic blood pressure, mm Hg, mean±SD	112±11	119±14
Antihypertensive treatment, no. (%)	137 (1.8)	13 622 (5.1)
Lipid‐lowering treatment, no. (%)	38 (0.5)	9067 (3.4)
Estimated glomerular filtration rate, mL/min per 1.73 m2, mean±SD	107.7±19.1	108.1±16.0

KPSC indicates Kaiser Permanente Southern California.

^*The “Other” group include Native American/Alaska Native, multiple or other races and ethnicities, as well as those unreported.

The mean predicted 10‐year ASCVD risk was higher when estimated by the PCEs compared with PREVENT (1.40% versus 0.47% in the epidemiologic cohorts; 1.18% versus 0.62% in KPSC) (Figure 1, Tables S3 and S4). Mean predicted 30‐year risk was 4.38% by the FHS with lipid equation, 4.80% by the FHS with BMI equation, and 3.05% by PREVENT (Figure 1, Table S5). Distributions of the predicted 10‐year and 30‐year risks by age, sex, and racial or ethnic subgroups are shown in Tables S3 through S5.

Figure 1. Distribution of 10‐year and 30‐year ASCVD risks estimated by the PCEs, PREVENT, and the FHS equations.

A, Epidemiologic cohorts; (B), KPSC. ASCVD indicates atherosclerotic cardiovascular disease; FHS, Framingham Heart Study; KPSC, Kaiser Permanente Southern California; PCEs, Pooled Cohort Equations; and PREVENT, Predicting Risk of Cardiovascular Disease Events.

In analyses assessing 10‐year ASCVD risk prediction, a total of 33 incident ASCVD events occurred in the epidemiologic cohorts and 1037 events occurred in KPSC, during a median follow‐up of 10 years. Compared with the PCEs, PREVENT improved discrimination in the overall population in both the epidemiologic cohorts (Harrell's C was 0.804 for PREVENT versus 0.748 for PCEs; ∆Harrell's C, 0.057 [95% CI, 0.013–0.101]) and KPSC (Harrell's C was 0.718 for PREVENT versus 0.676 for PCEs; ∆Harrell's C, 0.041 [95% CI, 0.034–0.049]) (Table 2). The improvement in discrimination was observed across all racial and ethnic subgroups. The PCEs overestimated 10‐year risk (mean calibration in the epidemiologic cohorts, 3.26 [95% CI, 2.46–5.13]; in KPSC, 1.73 [95% CI, 1.64–1.84]). The mean calibration of PREVENT was 1.10 (95% CI, 0.83–1.73) in the epidemiologic cohorts and 0.91 (95% CI, 0.86–0.96]) in KPSC, with the lowest mean calibration in non‐Hispanic Black individuals in KPSC (mean calibration, 0.54 [95% CI, 0.48–0.61]) (Table 2, Figure 2, Figures S3 and S4).

Table 2.

Harrell's C‐Statistics (95% CI) and Mean Calibration (95% CI) of the PCEs and the PREVENT Equations in Predicting 10‐Year ASCVD Risk

	Overall	Age		Sex		Race or ethnicity
		20–29 y	30–39 y	Female	Male	Non‐Hispanic White	Non‐Hispanic Black	Hispanic	Asian/Pacific Islander
Epidemiologic cohorts
No.	7736	4468	3268	4104	3632	5297	2307	NA
No. of events	33	10	23	12	21	24	8
Person‐years	76 578	44 327	32 251	40 769	35 809	52 499	22 811
PCEs
Harrell's C	0.748 (0.649 to 0.844)	0.659 (0.475 to 0.841)	0.786 (0.670 to 0.889)	0.716 (0.531 to 0.870)	0.747 (0.613 to 0.862)	0.777 (0.647 to 0.884)	0.699 (0.493 to 0.881)	NA
Mean calibration	3.26 (2.46 to 5.13)	6.10 (3.70 to 15.24)	2.03 (1.42 to 3.39)	2.31 (1.45 to 4.61)	3.81 (2.72 to 6.21)	2.81 (1.99 to 4.56)	4.88 (2.77 to 13.10)
PREVENT
Harrell's C	0.804 (0.725 to 0.884)	0.718 (0.519 to 0.868)	0.822 (0.728 to 0.908)	0.795 (0.683 to 0.896)	0.810 (0.710 to 0.893)	0.821 (0.714 to 0.905)	0.773 (0.590 to 0.921)	NA
∆Harrell's C=PREVENT—PCEs	0.057 (0.013 to 0.101)	0.058 (−0.067 to 0.109)	0.036 (−0.001 to 0.076)	0.079 (0.009 to 0.166)	0.063 (−0.012 to 0.143)	0.044 (0.002 to 0.088)	0.074 (0.014 to 0.149)
Mean calibration	1.10 (0.83 to 1.73)	1.60 (0.98 to 3.99)	0.89 (0.62 to 1.48)	1.07 (0.67 to 2.14)	1.12 (0.81 to 1.83)	1.13 (0.80 to 1.83)	1.09 (0.62 to 2.93)
KPSC
No.	266 378	91 743	174 635	160 075	106 303	68 020	21 296	124 431	28 433
No. of events	1037	143	894	479	558	277	156	465	121
Person‐years	1 547 317	441 185	1 106 132	937 187	610 130	412 396	132 258	751 774	190 382
PCEs
Harrell's C	0.676 (0.661 to 0.696)	0.636 (0.593 to 0.685)	0.669 (0.651 to 0.690)	0.648 (0.623 to 0.680)	0.671 (0.651 to 0.694)	0.665 (0.632 to 0.696)	0.688 (0.647 to 0.730)	0.669 (0.643 to 0.698)	0.736 (0.686 to 0.780)
Mean calibration	1.73 (1.64 to 1.84)	3.09 (2.61 to 3.66)	1.57 (1.48 to 1.66)	1.26 (1.17 to 1.38)	2.11 (1.91 to 2.30)	1.79 (1.61 to 2.07)	1.50 (1.32 to 1.69)	1.74 (1.59 to 1.88)	1.59 (1.30 to 1.96)
PREVENT
Harrell's C	0.718 (0.700 to 0.734)	0.650 (0.606 to 0.699)	0.697 (0.682 to 0.715)	0.703 (0.678 to 0.729)	0.719 (0.694 to 0.740)	0.705 (0.669 to 0.739)	0.754 (0.711 to 0.789)	0.704 (0.675 to 0.731)	0.769 (0.725 to 0.806)
∆Harrell's C=PREVENT—PCEs	0.041 (0.034 to 0.049)	0.014 (−0.001 to 0.033)	0.028 (0.019 to 0.038)	0.055 (0.041 to 0.068)	0.048 (0.035 to 0.063)	0.040 (0.025 to 0.054)	0.066 (0.047 to 0.086)	0.035 (0.026 to 0.045)	0.033 (0.008 to 0.050)
Mean calibration	0.91 (0.86 to 0.96)	1.26 (1.06 to 1.49)	0.90 (0.85 to 0.95)	0.77 (0.71 to 0.84)	1.02 (0.92 to 1.11)	0.95 (0.85 to 1.09)	0.54 (0.48 to 0.61)	0.98 (0.90 to 1.05)	0.91 (0.74 to 1.13)

ASCVD indicates atherosclerotic cardiovascular disease; KPSC, Kaiser Permanente Southern California; PCEs, Pooled Cohort Equations; and PREVENT, Predicting Risk of Cardiovascular Disease Events.

Figure 2. Calibration plots for the PCEs, PREVENT, and the FHS equations in predicting 10‐year and 30‐year ASCVD risk.

A, Epidemiologic cohorts; B, KPSC. ASCVD indicates atherosclerotic cardiovascular disease; BMI, body mass index; FHS, Framingham Heart Study; KPSC, Kaiser Permanente Southern California; PCEs, Pooled Cohort Equations; and PREVENT, Predicting Risk of Cardiovascular Disease EVENTs.

In analyses assessing 30‐year ASCVD risk prediction, a total of 286 incident ASCVD events occurred in the epidemiologic cohorts during a median follow‐up of 30 years. Discrimination was similar for PREVENT (Harrell's C=0.753) and FHS with lipid equation (Harrell's C0.747 ∆Harrell's C, 0.006 [95% CI, −0.007 to 0.020]), whereas discrimination was lower for the FHS with BMI equation (Harrell's C, 0.729; ∆Harrell's C, 0.024 [95% CI, 0.007–0.043]) (Table 3). Mean calibration was 0.63 (95% CI, 0.56–0.71) for PREVENT, 0.90 (95% CI, 0.81–1.02) for the FHS with lipid equation, and 0.99 (95% CI, 0.89–1.12) for the FHS with BMI equation (Table 3, Figure 2, Figure S5).

Table 3.

Harrell's C‐Statistics (95% CI) and Mean Calibration (95% CI) of the FHS Equations and the PREVENT Equations in Predicting 30‐Year ASCVD Risk, Epidemiologic Cohorts

	Overall	Age		Sex		Race
		20–29 y	30–39 y	Female	Male	White	Black
No.	7736	4468	3268	4104	3632	5297	2307
No. of events	286	148	138	107	179	170	115
Person‐years	192 422	120 980	71 442	103 187	89 235	128 057	62 449
FHS with lipid
Harrell's C	0.747 (0.714 to 0.778)	0.715 (0.672 to 0.753)	0.775 (0.731 to 0.819)	0.717 (0.669 to 0.765)	0.739 (0.705 to 0.772)	0.792 (0.756 to 0.828)	0.715 (0.670 to 0.755)
Mean calibration	0.90 (0.81 to 1.02)	0.69 (0.59 to 0.81)	0.99 (0.84 to 1.18)	0.76 (0.64 to 0.93)	0.99 (0.87 to 1.16)	1.11 (0.95 to 1.30)	0.55 (0.47 to 0.66)
FHS with BMI
Harrell's C	0.729 (0.698 to 0.762)	0.682 (0.636 to 0.723)	0.766 (0.726 to 0.807)	0.710 (0.659 to 0.762)	0.712 (0.675 to 0.747)	0.768 (0.731 to 0.806)	0.693 (0.644 to 0.736)
Mean calibration	0.99 (0.89 to 1.12)	0.76 (0.66 to 0.89)	1.08 (0.92 to 1.29)	0.83 (0.70 to 1.01)	1.09 (0.96 to 1.26)	1.18 (1.02 to 1.39)	0.67 (0.57 to 0.79)
PREVENT
Harrell's C	0.753 (0.721 to 0.782)	0.717 (0.675 to 0.757)	0.770 (0.726 to 0.815)	0.734 (0.685 to 0.780)	0.737 (0.702 to 0.771)	0.799 (0.766 to 0.833)	0.710 (0.660 to 0.757)
∆Harrell's C=PREVENT—FHS with lipid	0.006 (−0.007 to 0.020)	0.003 (−0.020 to 0.024)	−0.005 (−0.016 to 0.006)	0.017 (−0.006 to 0.042)	−0.002 (−0.018 to 0.015)	0.007 (−0.005 to 0.022)	−0.005 (−0.031 to 0.022)
∆Harrell's C=PREVENT—FHS with BMI	0.024 (0.007 to 0.043)	0.035 (0.003 to 0.067)	0.004 (−0.019 to 0.025)	0.024 (−0.010 to 0.061)	0.025 (0.000 to 0.052)	0.031 (0.011 to 0.051)	0.017 (−0.022 to 0.053)
Mean calibration	0.63 (0.56 to 0.71)	0.60 (0.51 to 0.70)	0.60 (0.51 to 0.72)	0.59 (0.50 to 0.72)	0.65 (0.57 to 0.76)	0.74 (0.64 to 0.87)	0.44 (0.37 to 0.52)

ASCVD indicates atherosclerotic cardiovascular disease; BMI, body mass index; FHS, Framingham Heart Study; and PREVENT, Predicting Risk of Cardiovascular Disease Events.

In sensitivity analysis using the alternative ASCVD definition (ie, nonfatal MI, fatal CHD, or fatal and nonfatal ischemic stroke), model discrimination improved for all 10‐year and 30‐year risk equations compared with Harrell's C from the main analysis (Tables S6 and S7). For example, in the epidemiologic cohorts, Harrell's C for PREVENT improved from 0.804 to 0.871 when predicting 10‐year ASCVD risk and improved from 0.753 to 0.783 when predicting 30‐year risk. Mean calibration was higher when using the alternative ASCVD definition compared with the main analysis results. For example, in the epidemiologic cohorts, mean calibration for PREVENT increased from 1.10 to 1.59 when predicting 10‐year risk and increased from 0.63 to 0.72 when predicting 30‐year risk, indicating more overestimation of 10‐year risk and better prediction of 30‐year risk by PREVENT when using the alternative ASCVD definition.

DISCUSSION

In this analysis of more than 290 000 young adults from diverse populations, the PREVENT equations demonstrated improved discrimination in predicting 10‐year ASCVD risk and were overall better calibrated compared with the PCEs. Although the PREVENT equations showed similar discrimination to the FHS equations in predicting 30‐year ASCVD risk, PREVENT underestimated 30‐year risk whereas the FHS equations had good calibration. These findings suggest that PREVENT may be a better tool for short‐term ASCVD risk assessment in young adults than the PCEs and the FHS equations may be better for long‐term risk assessment than PREVENT in this age group.

Despite the overall decline in ASCVD rates in the United States over the past 2 decades, stroke and CHD rates among young adults have been increasing. ² , ⁵ This trend underscores the urgent need to enhance early prevention efforts in this demographic group, which is often overlooked in ASCVD research. ²⁷ Although multiple US guidelines put risk assessment at the center of decision‐making for ASCVD primary prevention, the best approach to assess ASCVD risk among young adults remains unclear. ⁸ , ⁹ , ¹⁰ , ²⁸ Whereas 10‐year risk assessment in young adults can help identify those with elevated short‐term risk and guide the initiation of preventive pharmacotherapy, long‐term risk assessment captures the cumulative burden of risk over the life course and can identify individuals who may have low estimated short‐term risk but high lifetime risk. The PCEs, one of the most widely used 10‐year ASCVD risk assessment tools, were developed in and recommended for adults 40 to 79 years and may not be directly applicable to young adults. ¹⁰ The newly developed PREVENT equations extended the age range to 30 to 79 years, but young adults <30 years old were still not included. ¹¹ The current study showed that the mean predicted 10‐year ASCVD risk was about 2‐fold higher when estimated by the PCEs compared with PREVENT, consistent with findings from a recent study of US adults aged 40 to 75 years. ²⁹ When predicting incident 10‐year ASCVD events, PREVENT modestly improved model discrimination compared with the PCEs. The PCEs appeared to overestimate 10‐year risk by 226% on average in the epidemiologic cohorts and by 73% in KPSC. In comparison, PREVENT was overall well calibrated in the epidemiologic cohorts but slightly underestimated risk in KPSC, particularly in non‐Hispanic Black individuals. This discrepancy in calibration may reflect differences in cohort characteristics, health care settings, or temporal trends, as the KPSC population tend to receive standardized care and are more contemporary than those from the epidemiologic cohorts. Additionally, although the mean calibration of PREVENT in the epidemiologic cohorts was close to ideal, the wide CI reflects statistical uncertainty due to the relatively small number of events. Further research is needed to validate these findings in other contemporary data sources that include diverse racial and ethnic groups.

Additionally, studies have shown that young adults with adverse risk factors may have low short‐term but high lifetime ASCVD risks, suggesting that long‐term risk assessment may be more important for guiding ASCVD prevention and intervention in this population. ¹² , ¹³ , ¹⁴ The current study showed that on average, young adults had low predicted 10‐year and 30‐year risks, with highly right‐skewed distributions. In the epidemiologic cohorts, the median 10‐year risk estimated by the PREVENT was 0.94% versus 0.34% in young adults with versus without an incident ASCVD event at 10 years; the corresponding median 30‐year risk estimated by PREVENT was 4.26% versus 2.24% in young adults with versus without an incident ASCVD event at 30 years. Further, although the current study demonstrated overall good calibration for the PREVENT equations when predicting 10‐year ASCVD risk, PREVENT appeared to underestimate 30‐year risk and had worse calibration than the FHS equations. This underestimation may be due, in part, to the PREVENT equations being derived from data with a mean follow‐up of only 4.8 years. Additionally, because PREVENT was based on more contemporary data starting from 1992, whereas the external validation data from CARDIA and FHS date back to the 1970s to 1980s, this discrepancy might also contribute to the underestimation of 30‐year risk by PREVENT.

In sensitivity analyses using an alternative ASCVD definition that excluded hemorrhagic and other nonischemic stroke types, we observed modest improvements in model discrimination across all equations and subgroups. This finding is expected and likely reflects the closer pathophysiologic relationship between the risk factors included in ASCVD risk equations with ischemic stroke than with total stroke. These results suggest that restricting ASCVD outcomes to those more directly related to atherosclerosis, such as ischemic stroke, may enhance model discrimination. Mean calibration also increased across all equations and subgroups, and these changes led to improved (mean calibration closer to 1) or worsened calibration (further deviation from 1), depending on the model.

The current US guidelines classify a 10‐year ASCVD risk of ≥7.5% estimated by the PCEs as intermediate (7.5%–19.9%) or high (≥20%) risk among adults aged 40 to 75 years and use these thresholds to guide statin use. ⁸ , ⁹ However, clinically relevant 10‐year or 30‐year ASCVD risk thresholds have not been specifically developed and validated for young adults. This lack of tailored thresholds underscores a significant gap in our current risk prediction models, which were primarily designed for middle‐aged and older populations. Without these specific thresholds, young adults may be inaccurately assessed, potentially leading to undertreatment or overtreatment. Future studies are needed to establish these thresholds in young adults to enhance the precision of risk stratification and improve ASCVD outcomes in this population.

Strengths and Limitations

The main strengths of the current study include the use of large, diverse populations of young adults from both well‐characterized epidemiologic cohorts and a large integrated health care system. This allowed us to evaluate risk model performance across different settings, and the consistency of findings supports the external validity and generalizability of the results. The long‐term follow‐up data and rigorous event ascertainment also allowed us to compare the performance of the newly developed PREVENT equations with existing short‐term and long‐term risk assessment tools recommended in the 2018 American Heart Association/American College of Cardiology cholesterol guideline.

Several limitations also need to be considered. First, both CARDIA and FHS were used as part of the derivation cohorts in the development of the PCEs and PREVENT, and the FHS equations were developed solely based on the FHS Offspring cohort. As a consequence, model validation in the epidemiologic cohorts may result in more optimistic estimates of prediction performances for all 3 risk models. However, external validation of the 10‐year risk equations in KPSC found consistent results. Second, because the maximum follow‐up time in KPSC was 15 years, we were able to assess the performance of 30‐year ASCVD risk equations in the epidemiologic cohorts but not in KPSC. Although the epidemiologic cohorts provided long follow‐up, they primarily included non‐Hispanic Black and White individuals. Future research is needed to evaluate the 30‐year risk equations in other contemporary, racially and ethnically diverse data sources with long‐term follow‐up. Third, compared with young adults included in the final KPSC analytic sample, those excluded due to incomplete data were generally younger and with lower cardiovascular risk factor burden. As such, the analytic cohort may overrepresent young adults at higher baseline risk and limit the generalizability of model performance to lower‐risk populations. Fourth, the current analysis only assessed the base PREVENT model but not the optional PREVENT equations enhanced with additional predictors including urine albumin‐to‐creatinine ratio, hemoglobin A1c, and social deprivation index, as these variables were not universally available across all study samples. Lastly, given that the PCEs are applicable only to adults aged 40 and older and the PREVENT equations to adults aged 30 and older, we substituted an age of 40 years for the PCEs and an age of 30 years for PREVENT when participants were younger than these thresholds. This approach likely overestimated the predicted risk for young adults aged 20 to 39 years.

CONCLUSIONS

In this analysis of young adults from large diverse populations, the newly developed PREVENT equations demonstrated improved discrimination and calibration for predicting 10‐year ASCVD risk compared with the PCEs. For 30‐year risk prediction, although PREVENT showed similar discrimination to the FHS equations, PREVENT underestimated 30‐year risk whereas the FHS equations had good calibration. These findings suggest that PREVENT may be a better tool for short‐term ASCVD risk assessment in young adults than the PCEs, and the FHS equations may be better for long‐term risk assessment than PREVENT in this age group. Future research is needed to evaluate the 30‐year risk equations in more contemporary and racially and ethnically diverse young adult populations.

Sources of Funding

This work was supported by National Institutes of Health grants R01HL155081, R01HL168379, and R01HL158790 (Zhang, An). The Coronary Artery Risk Development in Young Adults Study is supported by contracts 75N92023D00002, 75N92023D00003, 75N92023D00004, 75N92023D00005, and 75N92023D00006 from the National Heart, Lung, and Blood Institute. The Framingham Heart Study was supported by the National Heart, Lung, and Blood Institute (HHSN268201500001I and 75N92019D00031) and the Boston University School of Medicine.

Disclosures

Kristi Reynolds received research support from Merck Sharp & Dohme LLC unrelated to this work.

Supporting information

Data S1

Tables S1–S7

Figures S1–S5