Kidney Disease Blunts Accuracy of Heart Risk Scores in Older Women

Commonly used cardiovascular risk calculators perform poorly in postmenopausal women, and they perform worst of all in those living with chronic kidney disease, according to an observational cohort study published in the Journal of General Internal Medicine. Drawing on data from more than 21,000 women enrolled in the Women's Health Initiative Biomarker Studies, researchers found that five widely used prediction models—the Framingham risk score, the Adult Treatment Panel (ATP) III equations, the ACC/AHA Pooled Cohort Equations (PCE), QRISK3, and the newer PREVENT model—showed only modest ability to distinguish women who later experienced cardiovascular events from those who did not, and most models systematically underestimated risk in women with reduced kidney function. The analysis included 21,548 postmenopausal women aged 50 to 79 who were free of cardiovascular disease at baseline.

Within this cohort, 1,095 women had chronic kidney disease, defined in the study as an estimated glomerular filtration rate below 60 mL/min per 1.73 m². The researchers evaluated how well each model discriminated between women who did and did not go on to have a cardiovascular event, using the C-statistic, and how well predicted probabilities matched observed event rates at clinically relevant decision-making thresholds. Because each model was designed to predict a different set of outcomes, the number of events counted varied by definition.

In women without kidney disease, events ranged from 846 (4.1 percent) under the ATP III definition of fatal and nonfatal coronary heart disease to 1,435 (7.0 percent) under the QRISK3 definition, which included coronary heart disease, stroke, or transient ischemic attack. Among women with chronic kidney disease, events ranged from 93 (8.5 percent) for ATP III to 145 (13.2 percent) for QRISK3. The Framingham model counted myocardial infarction, angina, or coronary heart disease death, while the PCE and PREVENT models counted coronary heart disease or ischemic stroke.

Discrimination was unimpressive across the board. In the group with chronic kidney disease, C-statistics ranged from 0.57 for the Framingham model to 0.67 for PREVENT. In women without kidney disease, values ranged from 0.61 for Framingham to 0.71 for both QRISK3 and PREVENT.

The figures observed here leave substantial room for error in individual risk classification. Calibration—the agreement between predicted and observed risk—was generally poor in both groups. Most models underpredicted cardiovascular risk in women with chronic kidney disease, meaning the calculators suggested lower risk than the women actually experienced.

Notably, the PREVENT model, the most recent of the five, underpredicted risk for women both with and without kidney disease, suggesting that even the newest equations may not fully capture risk in this population of older women. If risk is underpredicted in women with chronic kidney disease, risk stratification across this high-risk group becomes less reliable, and predicted probabilities may not reflect the true burden of cardiovascular disease these women experience. Chronic kidney disease is a well-established risk factor for cardiovascular events, yet the performance of cardiovascular prediction models in people with reduced kidney function, and in women with the condition in particular, has been understudied.

This secondary analysis of a prospective cohort is among the efforts to fill that gap, comparing five models head-to-head within the same population rather than relying on studies conducted in different cohorts with different populations. The study has limitations inherent to its design. As a secondary analysis of the Women's Health Initiative Biomarker Studies, it reflects a cohort of postmenopausal women aged 50 to 79, and results may not generalize to men, younger women, or other populations.

The event counts also varied by outcome definition, complicating direct comparison between models. Even so, the central message is sobering: none of the five models achieved strong discrimination in women with chronic kidney disease, and calibration failures were the norm rather than the exception. The authors conclude that common cardiovascular risk prediction models have poor discrimination and calibration in women with chronic kidney disease, and that the PREVENT model generally underpredicted risk regardless of kidney status.

For researchers, the results underscore how existing equations perform in this high-risk group and where their predictions fall short.

Subject of Research: Performance of cardiovascular risk prediction models in postmenopausal women with and without chronic kidney disease

Article Title: Cardiovascular Risk Prediction in Postmenopausal Women With and Without Chronic Kidney Disease: An Observational Cohort Study

Article References: Gallo, R. J., Qin, F., Hedlin, H., Stefanick, M. L., Chertow, G. M., Manson, J. E., Allison, M., Franceschini, N., Shadyab, A. H., Saquib, N., Rhee, J. J., Bress, A. P., & Maron, D. J. (2026). Cardiovascular Risk Prediction in Postmenopausal Women With and Without Chronic Kidney Disease: An Observational Cohort Study. Journal of General Internal Medicine. https://doi.org/10.1007/s11606-026-10946-x

Image Credits: AI Generated

DOI: 10.1007/s11606-026-10946-x

Keywords: cardiovascular risk prediction, chronic kidney disease, postmenopausal women, Women's Health Initiative, Pooled Cohort Equations, PREVENT model, QRISK3, Framingham risk score, calibration, C-statistic, observational cohort, Journal of General Internal Medicine