Development of a risk score for atrial fibrillation (Framingham Heart Study): a community-based cohort study.

BACKGROUND: Atrial fibrillation contributes to substantial increases in morbidity and mortality. We aimed to develop a risk score to predict individuals' absolute risk of developing the condition, and to provide a framework for researchers to assess new risk markers. METHODS: We assessed 4764 participants in the Framingham Heart Study from 8044 examinations (55% women, 45-95 years of age) undertaken between June, 1968, and September, 1987. Thereafter, participants were monitored for the first event of atrial fibrillation for a maximum of 10 years. Multivariable Cox regression identified clinical risk factors associated with development of atrial fibrillation in 10 years. Secondary analyses incorporated routine echocardiographic measurements (5152 participants, 7156 examinations) to reclassify the risk of atrial fibrillation and to assess whether these measurements improved risk prediction. FINDINGS: 457 (10%) of the 4764 participants developed atrial fibrillation. Age, sex, body-mass index, systolic blood pressure, treatment for hypertension, PR interval, clinically significant cardiac murmur, and heart failure were associated with atrial fibrillation and incorporated in a risk score (p<0.05, except body-mass index p=0.08), clinical model C statistic 0.78 (95% CI 0.76-0.80). Risk of atrial fibrillation in 10 years varied with age: more than 15% risk was recorded in 53 (1%) participants younger than 65 years, compared with 783 (27%) older than 65 years. Additional incorporation of echocardiographic measurements to enhance the risk prediction model only slightly improved the C statistic from 0.78 (95% CI 0.75-0.80) to 0.79 (0.77-0.82), p=0.005. Echocardiographic measurements did not improve risk reclassification (p=0.18). INTERPRETATION: From clinical factors readily accessible in primary care, our risk score could help to identify risk of atrial fibrillation for individuals in the community, assess technologies or markers for improvement of risk prediction, and target high-risk individuals for preventive measures.

Duke Scholars

Author Michael J Pencina Biostatistics & Bioinformatics, Division of Biostatistics