Comprehensive Evaluation of Atherosclerotic Cardiovascular Disease Risk Predictors Across the Pooled Cohort Equations, Predicting Risk of Cardiovascular Disease Events, and an Expanded Model.
Zhang Y, Schaefer E, et al. • Journal of the American Heart Association • 2026
Self-reported race did not improve ASCVD risk prediction, whereas small dense low-density lipoprotein cholesterol, hs-CRP, and education level added predictive value, suggesting potential utility in including additional biomarkers and social factors.
Key Findings
Results
PCE overestimated 10-year ASCVD risk while PREVENT-ASCVD underestimated risk when applied using published coefficients.
Analysis was conducted in a pooled sample of 13,108 participants across 3 prospective cohorts: ARIC, FOS, and MESA.
58.4% of participants were female, 22.6% were Black, and median age was 61 years.
873 participants (6.7%) developed ASCVD within 10 years.
Miscalibration was observed when both models were applied with their originally published coefficients rather than after refitting.
Results
Self-reported race was not selected as a predictor in the expanded ASCVD risk model across multiple variable selection methods.
Variable selection was performed using stepwise selection, elastic net, and random forest methods.
Self-reported race was not selected by any of the three variable selection approaches.
The finding suggests race does not improve ASCVD risk prediction beyond other included variables.
This contrasts with its inclusion as a race-specific coefficient in the PCE model.
Results
Small dense low-density lipoprotein cholesterol (sdLDL-C), high-sensitivity C-reactive protein (hs-CRP), and education level were selected as additional predictors beyond those in PCE and PREVENT-ASCVD.
These three variables were identified across multiple variable selection methods (stepwise selection, elastic net, and random forest).
sdLDL-C, hs-CRP, and education level represent both emerging biomarkers and a social determinant of health.
These predictors were added as candidate variables alongside standard PCE and PREVENT-ASCVD variables.
Lipoprotein(a) was also evaluated as a candidate predictor but was not consistently selected.
Results
The EXPAND model achieved consistently higher C-statistics compared with both PCE and PREVENT-ASCVD models refitted in the study sample.
EXPAND (Expanded ASCVD Non-traditional Determinants) integrated selected predictors including sdLDL-C, hs-CRP, and education.
Comparisons were made against both originally published and refitted versions of PCE and PREVENT-ASCVD.
Higher C-statistics indicate better discriminative ability of EXPAND across the full sample.
The improvement was described as consistently achieved, suggesting robustness across subgroup analyses.
Results
EXPAND showed modest calibration advantages over refitted PCE and PREVENT-ASCVD, particularly among Black men.
Calibration advantages were described as modest overall but more pronounced in the subgroup of Black men.
Both PCE and PREVENT-ASCVD were refitted in the study sample to provide a fair comparison controlling for cohort-specific differences.
Black participants comprised 22.6% of the total study population (n=13,108).
Improved calibration in Black men is notable given historical concerns about PCE performance in this group.
Results
Most predictors shared by PCE and PREVENT-ASCVD were selected across all three variable selection methods.
Candidate predictors included all variables from both PCE and PREVENT-ASCVD models.
Three selection methods were applied: stepwise selection, elastic net, and random forest.
Consistent selection of shared predictors across methods supports the validity of the core variables in existing models.
Cox proportional hazards models were used as the underlying regression framework.
Methods
The study pooled data from three prospective cohorts—ARIC, FOS, and MESA—to evaluate ASCVD risk prediction models.
Total pooled sample was 13,108 participants.
The sample was 58.4% female and 22.6% Black with a median age of 61 years.
873 participants (6.7%) developed ASCVD within 10 years, defining the outcome.
The multi-cohort design was intended to provide sufficient sample size and demographic diversity for model evaluation.
What This Means
This research suggests that current tools used to predict a person's 10-year risk of heart attack or stroke have notable limitations. The widely used Pooled Cohort Equations (PCE) tended to overestimate risk, while a newer model called PREVENT underestimated it, when tested in a pooled dataset of over 13,000 adults from three large long-term studies. The researchers then built a new exploratory model called EXPAND that incorporated both the standard predictors and additional factors including a specific type of 'bad' cholesterol particle (small dense LDL), an inflammation marker (hs-CRP), and education level as a measure of social circumstance.
One notable finding is that self-reported race—which is built into the PCE as a separate calculation for Black versus non-Black individuals—did not meaningfully improve risk prediction when other factors were accounted for. In contrast, the biological markers sdLDL-C and hs-CRP, as well as education level, consistently added predictive value across multiple statistical approaches. The EXPAND model showed better ability to distinguish who would and would not develop cardiovascular disease (higher C-statistics) and had modestly better calibration, especially among Black men, a group where existing models have historically performed poorly.
This research suggests that future cardiovascular risk prediction tools may benefit from incorporating emerging biological markers and social factors like education, rather than relying on self-reported race as a proxy. The findings also highlight that even well-established clinical tools may need recalibration when applied to new populations, and that continued refinement of risk models could help ensure more equitable and accurate predictions across diverse groups.
Zhang Y, Schaefer E, Ikezaki H, Diffenderfer M, Lloyd-Jones D, Hoogeveen R, et al.. (2026). Comprehensive Evaluation of Atherosclerotic Cardiovascular Disease Risk Predictors Across the Pooled Cohort Equations, Predicting Risk of Cardiovascular Disease Events, and an Expanded Model.. Journal of the American Heart Association. https://doi.org/10.1161/JAHA.125.049726