Stroke prevention programs are typical secondary stroke prevention programs that enroll persons who have already suffered a stroke. If persons at high-risk for a first time stroke could be identified, they can be enrolled in stroke prevention programs immediately instead of waiting until after the stroke occurs. However, current stroke risk calculators are developed from non-VA populations and are typically implemented on individuals with full data, not on entire populations where data may be incomplete or missing.
To determine whether the Framingham stroke calculator, the most commonly used tool for predicting stroke risk, can identify persons who will develop a stroke in the following year using VA administrative data.
Our study design was a retrospective cohort of Veterans hospitalized with a first-time stroke compared to a random sample receiving care from these facilities. The stroke cohort consisted of all 380 Veterans who were hospitalized for a first-time ischemic stroke in 2008 at 5 VA Medical Centers in the Southwestern United States. The comparison group consisted of 12,390 Veterans, selected at random, receiving any care at these Medical Centers in 2008. For all Veterans, we obtained VA administrative data available as of 2007 to implement several variations of the Framingham stroke risk calculator. We calculated c statistics in order to compare the predictive validity of this tool in this setting compared to the originally published studies on the Framingham cohort.
Persons with a 2008 stroke had a higher risk profile than the comparison group on several components of the general cardiovascular Framingham calculator: older age, more likely to be male sex, higher systolic blood pressure, higher total cholesterol, and more likely to have diabetes (all p<0.05), resulting in a greater overall Framingham score (16.4 versus 14.4, p<0.0001) among persons with stroke. The c statistic for the Framingham cardiovascular calculator was 0.60 for identifying persons who would develop a stroke in 2008. However, there were 27% of cases and 20% of controls who did not have any blood pressure or cholesterol levels in 2007. After deleting persons with missing clinical data, the c statistic improved to 0.67 but was still lower than the c statistics of 0.77 to 0.84 reported in the original Framingham stroke studies.
Identifying persons at high-risk for a first-time stroke represents a change in perspective from a reactive VA healthcare system to a proactive one. This study represents a first-step toward developing an accurate stroke prediction tool using administrative databases on a VA population.
- Keyhani S, Cheng E, Ofner S, Williams L, Bravata D. The underuse of carotid interventions in veterans with symptomatic carotid stenosis. The American journal of managed care. 2014 Jul 1; 20(7):e250-6.
- Lich KH, Tian Y, Beadles CA, Williams LS, Bravata DM, Cheng EM, Bosworth HB, Homer JB, Matchar DB. Strategic planning to reduce the burden of stroke among veterans: using simulation modeling to inform decision making. Stroke. 2014 Jul 1; 45(7):2078-84.
- Ekundayo OJ, Vassar SD, Williams LS, Bravata DM, Cheng EM. Using administrative databases to calculate Framingham scores within a large health care organization. Stroke. 2011 Jul 1; 42(7):1982-7.
- Cheng E, Vasser S, Ekundayo J, Williams LS, Bravata DM. Can Framingham Calculators Identify Persons at High-Risk for Stroke in a Large Healthcare Organizations? [Abstract]. Stroke; A Journal of Cerebral Circulation. 2010 Feb 22; 41:e365.
Treatment - Observational, Prevention
Cardiovasc’r disease, Screening, Stroke
Prevention, Patient Education