Adapted Framingham Models Accurately Reproduce Cardiovascular Risk Using Self-Reported Data

Adapted cardiovascular risk assessment models that substitute self-reported medical diagnoses for laboratory and clinical measurements can reliably reproduce the Framingham Risk Score across large populations, according to a peer-reviewed study analyzing data from the Brazilian National Health Survey. Researchers found that models relying on self-reported hypertension, dyslipidemia, and diabetes mellitus demonstrated high diagnostic agreement and strong discrimination when compared directly against the standard clinical Framingham formulation, offering a viable alternative for cardiovascular surveillance when biochemical testing and physical evaluations are unavailable.

The investigation focused on evaluating whether self-reported chronic condition diagnoses could serve as reliable proxies in the Framingham Risk Score, examining both the general population and a specific subgroup of adults with higher education. The Framingham Risk Score is commonly utilized in clinical settings to estimate individual cardiovascular risk profiles, but calculating the score in its standard form requires direct clinical examinations and laboratory biochemical measurements that are frequently unavailable in large epidemiological health surveys. To overcome this limitation in public health research, the study team assessed whether adapted self-reported models could closely reproduce the risk scores generated by the original clinical reference standard.

To test the viability of these adaptations, the investigators analyzed cross-sectional health records from 6,555 participants aged between 30 and 74 years who took part in the 2013 Brazilian National Health Survey, known as the Pesquisa Nacional de Saúde. Three distinct adapted risk models were developed and compared against the original Framingham Risk Score as the benchmark reference: the FRS-SAH model, which replaces clinical blood pressure measures with self-reported systemic arterial hypertension; the FRS-DLP-DM model, which incorporates self-reported dyslipidemia and diabetes mellitus; and the FRS-DLP-SAH-DM model, which substitutes self-reported data across all three conditions simultaneously.

The discriminatory performance and statistical calibration of the three adapted formulations were evaluated across three standard cardiovascular risk cutoff thresholds: risk probabilities of at least 5 percent, at least 10 percent, and at least 20 percent. The researchers conducted rigorous statistical evaluations utilizing receiver operating characteristic analysis to generate area under the curve metrics, Brier scores to measure probability calibration error, intraclass correlation coefficients to quantify continuous agreement, Bland–Altman plots to evaluate systematic and proportional bias, and bootstrap resampling analysis to establish internal model stability.

Within the total study population, the FRS-SAH adaptation demonstrated the highest capacity to reproduce the original Framingham score across all evaluated cardiovascular risk thresholds. This model yielded an area under the curve of 0.963 at the 5 percent risk threshold, 0.954 at the 10 percent threshold, and 0.917 at the 20 percent threshold. In addition to high discrimination, the FRS-SAH adaptation achieved the lowest Brier score values across every cutoff—recording 0.4110 at the 5 percent cutoff, 0.2443 at the 10 percent cutoff, and 0.1375 at the 20 percent cutoff—alongside the highest continuous agreement reflected by the intraclass correlation coefficient.

The FRS-DLP-DM model also exhibited strong discriminatory capability across the full sample, recording area under the curve values of 0.967 at the 5 percent threshold, 0.952 at the 10 percent threshold, and 0.914 at the 20 percent threshold. Conversely, the FRS-DLP-SAH-DM model, which replaced all three clinical and laboratory measures with self-reported diagnoses, showed slightly lower discrimination, particularly at the highest cardiovascular risk tier. That model produced area under the curve values of 0.928 at the 5 percent cutoff, 0.902 at the 10 percent cutoff, and 0.834 at the 20 percent cutoff.

When examining continuous score agreement against the original Framingham metric, all three adapted models showed high correlation, though differences in mean bias were documented across the specifications. The FRS-SAH model showed the smallest mean bias at +0.42 percent, whereas FRS-DLP-DM produced a mean bias of +2.25 percent and FRS-DLP-SAH-DM recorded a mean bias of +2.65 percent. Bland–Altman agreement plots demonstrated that proportional bias emerged primarily at elevated cardiovascular risk levels above 15 to 20 percent. Nonetheless, bootstrap analysis confirmed that all three models possessed high internal stability and maintained solid discriminative performance.

The researchers observed high agreement between the adapted scores and the original Framingham model in both the general population and among individuals with higher educational attainment. The authors concluded that simplified versions of the Framingham Risk Score based on self-reported health data perform with high fidelity relative to the clinical standard. These results support the practical implementation of self-reported risk models in large-scale epidemiological surveys and public health surveillance programs where laboratory testing, biochemical analyses, and direct physical measurements cannot be readily obtained.

Subject of Research: Validation of adapted Framingham Risk Score models utilizing self-reported health survey data

Article Title: Adaptation of the Framingham score with self-reported data on hypertension, dyslipidemia, and diabetes mellitus (Brazilian National Health Survey, 2013): a possibility for use in large epidemiological surveys

Article References: Rocha, M. L. D. D., Paixão, L. T., Menezes-Júnior, L., Meireles, A. L., & Carraro, J. C. C. (2026). Adaptation of the Framingham score with self-reported data on hypertension, dyslipidemia, and diabetes mellitus (Brazilian National Health Survey, 2013): a possibility for use in large epidemiological surveys. BMC Cardiovascular Disorders. https://doi.org/10.1186/s12872-026-06887-2

Image Credits: AI Generated

DOI: 10.1186/s12872-026-06887-2

Keywords: Framingham Risk Score, cardiovascular risk, self-reported data, hypertension, dyslipidemia, diabetes mellitus, Brazilian National Health Survey, epidemiological surveillance, ROC analysis, area under the curve, Brier score, intraclass correlation coefficient, Bland-Altman analysis, bootstrap validation, population health, cardiovascular disease screening, model adaptation