Abstract P079: Development of a Risk Prediction Model for Incident Peripheral Artery Disease Hospitalization. The ARIC Study
Introduction: Improving clinicians’ capacity to use readily available information to predict risk for hospitalized peripheral artery disease (PAD) in the general population could lead to more successful PAD prevention efforts and a reduction in PAD-related hospitalizations. However, previously published PAD risk algorithms have been criticized, thus warranting development of a more parsimonious and easily applied model.
Aim: Identify the sociodemographic and potentially modifiable risk factors that predict incident hospitalized PAD. We restricted attention to risk factors for which information is readily available to primary care health practitioners.
Methods: After excluding those with baseline PAD (n=715), missing covariates of interest (n=1230), and race/center strata with insufficient numbers (n=103), we analyzed data from 13,744 participants of the community-based ARIC cohort. Hospitalized PAD incidence was defined as the first occurrence of a primary or secondary listing of a PAD-related ICD-9 code between 1987 and 2008. ICD codes used were those indicated to define PAD in the Mayo Clinic Algorithm for Identifying PAD Patients from EMR. The relationship between incident hospitalized PAD and covariates of interest was modeled by independently adding each covariate to the base model with age, race, and gender. Model performance was assessed using area under the receiver-operating curve (AUC) for 10-year risk and likelihood ratio (LR) testing for goodness of fit. Variables that contributed > 0.03 to incremental AUC values (IAUC) or were significant in LR testing (p< 0.10) were entered into a final model.
Results: There were 318 incident hospitalized PAD events during a median of 18.1 years of follow-up (249,023 person-years). The incidence of hospitalized PAD was 12.8 per 10,000 person-years. Variables that independently contributed > 0.03 to the base model (AUC, 0.61) included diabetes (IAUC, 0.10), smoking (IAUC, 0.06), hypertension (IAUC, 0.04), and education (IAUC, 0.03). All of the following were statistically significant (p<0.0001) with LR testing: diabetes, smoking, hypertension, education, prevalent CHD, and BMI. These variables were included in the final model (AUC, 0.79).
Conclusions: We optimized a set of risk factors that predict incident hospitalized PAD in a biracial, population based cohort of middle-aged men and women. A parsimonious model restricted to age, race, gender, and diabetes discriminates moderately well (AUC, 0.71) and its ease of use would facilitate adoption in a primary care setting. As PAD is indicative of systemic atherosclerotic burden, aggressive prevention efforts could have benefits in delaying or preventing myocardial infarction, stroke, and other major circulatory system disorders.
- © 2013 by American Heart Association, Inc.