Abstract 18114: Differences Between Observed and Predicted Cardiovascular Event Rates Using the Framingham and REACH Equations: The Case of High-intensity Statin Users in the United Kingdom

Abstract

Background: High-intensity statin users are a select group of patients, and their observed cardiovascular (CV) event risk may not align with the predicted risk from published risk equations. Hypothesis: Published risk equations accurately characterize the observed patient risk in high intensity statin users Methods: Five cohorts were studied in the Clinical Practice Research Datalink: high-risk primary prevention (HRPP - prevalent diabetes in the absence of any other CV event), evident CV disease (ECVD - prevalent stable angina, revascularization, transient ischemic attack, carotid stenosis, abdominal aortic aneurysm, or peripheral vascular disease), incident acute coronary syndrome (ACS), incident ischemic stroke (IS), and incident heart failure (HF). Patients were followed from 1/1/2005 (HRPP, ECVD) or first qualifying event (ACS/IS/HF) until the first CV hospitalization after cohort inclusion, death, or end of observation (December 31, 2011). Predicted 10-year overall risk was based on Framingham (D’Agostino 2008) for HRPP and on REACH (Wilson, 2012) for all other cohorts. Acute (<=12 mos) and long-term rates (12+ mos) were generated for recurrent events in the ACS, IS, and HF groups (Wilson, 2012). Predicted rates were compared to the observed event rates using rate ratios, (observed rate/predicted rate). Standard errors were estimated using Poisson regression. Results: Framingham was well calibrated to the observed event rate in the HRPP cohort (RR 0.92 (95% CI 0.87 to 0.97). For all secondary prevention cohorts, REACH consistently underestimated the observed event rate, though the magnitude of the difference varied (see Table 1). Conclusions: Traditional risk scores consistently underestimated event rates in all secondary prevention populations treated with a high intensity statin. Observed events are vital for anchoring predicted rates to accurately determine country level estimates in certain patient populations.