Abstract P071: Creating Claims-based Risk Scores for Incident Coronary Heart Disease
Background: A claims-based coronary heart disease (CHD) risk score may be useful to control for confounding and identify high risk groups in research using administrative data.
Objective: To derive and validate a claims-based risk score for incident CHD in 2004-2011 Medicare 5% national samples and validate in the REasons for Geographic and Racial Differences in Stroke (REGARDS)-Medicare link population.
Methods: We included Medicare beneficiaries with no evidence of CHD who had fee-for-service health insurance. When examining prescriptions, Medicare part D coverage was required. CHD was defined as the first inpatient claim for acute myocardial infarction (AMI). Three groups of predictors were included in the risk models (group 1: traditional CHD risk factors, group 2: other diagnoses and demographic characteristics, group 3: medications). Using Cox proportional hazards models, c-statistics and 95% CI were calculated for model discrimination and CHD rates were calculated across deciles of predicted risk testing calibration.
Results: Of 990,836 individuals in the derivation cohort, 28,086 (2.8%) developed AMI over 5 years. The average age was 67.5 years, 57.6% (570,722/990836) were female, and 83.6% (828,338/990,836) were white, 10.7% (106,020/990,836) black, and 5.7% (56,477/990,836) other ethnicities. Models with group 1 variables had a moderate discrimination (Table). Adding group 2 variables improved model discrimination. Comparing the Framingham CHD risk score to our model with group 1 + 2 variables yielded similar discrimination. Group 3 variables (prescriptions) improved discrimination to 0.727 (0.714 - 0.741). Prediction models overestimated risk in the top decile in the Medicare 5% sample and the REGARDS-Medicare link population.
Conclusion: In patients with no history of CHD, models including demographics and diagnoses from claims data exhibited discrimination and calibration similar to the Framingham Risk Score. The inclusion of prescription information further improved discrimination.
Author Disclosures: M. Mefford: None. L. Huang: B. Research Grant; Significant; Amgen. M. Safford: B. Research Grant; Significant; Amgen. P. Muntner: B. Research Grant; Significant; Amgen. E. Levitan: B. Research Grant; Significant; Amgen.
- © 2016 by American Heart Association, Inc.