1108 — Performance of Framingham and UKPDS Risk Tools in Estimating 5-Year Cardiovascular (CV) Outcomes in Veterans with Diabetes
Wiitala WL, Sussman JB, and Hayward RA, Center for Clinical Management Research, Ann Arbor HSR&D;
Accurate prediction of patient CV risk is essential to optimal management of blood pressure and lipids, especially for those with diabetes. However, previous research suggests that prediction tools developed in one patient population often have substantially diminished accuracy when used in a dissimilar patient population. Therefore, we sought to examine the performance of two established risk models, the Framingham (1998) and the UK Prospective Diabetes Study (UKPDS; 2001) tools, in estimating fatal and non-fatal CV events in a sample of VA diabetic patients.
The sample was comprised of VA patients (N = 1,791) enrolled in the VADT clinical trial, a study of intensive glycemic control. We excluded subjects who were female (n = 52), had no follow-up time (n = 23), had a prior history of heart disease (n = 710), or were missing values for risk factors (n = 13), resulting in a final sample of N = 993. Patient enrollment occurred between 2000 and 2003. Patients’ estimated 5-year CV risk scores were calculated using Framingham and UKPDS equations. We examined calibration, discrimination, and reclassification measures to assess and compare the risk scores.
Median follow-up was 4.3 years; there were 67 CV events. We found poor performance of both tools. Although the UKPDS equation had better discrimination (c = 0.64) than the Framingham equation (c = 0.58), both were far below that reported for their internal validation. Framingham and UKPDS performed similarly by the net reclassification index (NRI); the integrated discrimination improvement (IDI) indicated a slight increase in mean sensitivity using UKPDS. Calibration of both measures was poor, with Framingham underestimating for low risk groups and overestimating for higher risk groups, while the UKPDS model overestimated events across all groups.
Although UKPDS was slightly more predictive than Framingham, neither of the risk models provided dependable estimates of risk for Veterans with diabetes.
The central goal of Veteran-centered medical care is tailoring treatment to individual Veteran risks, benefits, and preferences. This requires accurate information on patient risks; however, we show that existing tools may be ineffective in VHA. A CV risk score developed and validated on Veterans may help VHA provide more patient-centered outpatient care and reduce both under- and over-treatment.