Prediction of stroke in a population with high prevalence of diabetes: The strong heart study

Abstract

Background- Stroke prediction models are valuable to physicians in evaluating the risk of their patients so that preventive interventions can be promoted. The Framingham Risk Profile is a widely used stroke prediction equation. However, the contributions of some common risk factors for stroke vary across populations and some risk factors are specific to certain populations. For example, albuminuria is an important risk factor in American Indians (AIs), which is not included in the Framingham equation. The objective of the current study is to develop stroke prediction equations using routinely collected variables in AIs, a population with high rates of diabetes and stroke. Methods- The data used in the analysis are from 4507 stroke free participants at enrollment in the Strong Heart Study (SHS), the largest population-based longitudinal study of cardiovascular disease (CVD) and its risk factors in AIs in Arizona, Oklahoma, and South/North Dakota. As of December 2008, 379/4507 (8.4%) participants suffered a first stroke during an average follow-up of 17 years. Baseline potential risk factors were included in the Cox proportional-hazard models to develop gender-specific prediction equations. Backward selection was used to choose the predictors. Model performance was assessed using Harrell's C statistics based on bootstrapping methods. Results- Baseline age, untreated systolic blood pressure, treated diastolic blood pressure, HDL-C, current smoking, diabetes, macro-albuminuria, and history of CVD are significant predictors for incident stroke among women. Most of these predictors except HDL-C were also in the prediction equation for men. The equations provided good discrimination ability, as indicated by a C statistic of 0.72 for men and 0.73 for women. Conclusions- Predicted risk of stroke in 10 years can be provided for physicians and their patients. Then appropriate intervention can be implemented. The stroke prediction equations from SHS can be applied to other AIs as well as other ethnic groups with high rates of diabetes and albuminuria.