Abstract 11323: Improving Risk Assessment With Generalized Cardiovascular Disease Risk Score

Introduction: Clinical practice usually divides cardiovascular (CV) disease prevention patients into distinct primary (1 ^o ) and secondary (2 ^o ) prevention groups.

Hypothesis: That a generalized CVD risk prediction model will find overlap in risk between primary and (2 ^o ) prevention patients and variation within them, allowing better targeting of interventions for CV risk.

Methods: We followed all patients aged 40-75 who visited a Veterans Affairs outpatient clinic during the year 2010. The primary outcome was having a new ASCVD event in 5 years of follow-up. We used one risk score based on existing risk scores (traditional model) and developed a second using elastic net logistic regularization that with over 100 variables from the electronic health record (VARS+). All models were assessed for predictive accuracy. Models were developed on an 80% sample of the dataset and tested on the remaining 20%.

Results: The sample had 742,787 patients, over 20,000 of whom were women. 7.7% had a history of heart attack, 5.2% stroke, and 27% had diabetes. 30% of male participants and 23% of women an event during follow-up. The c-statistic is 0.75 for the traditional model and 0.77 for the VARS+ model, with good calibration in both. There was substantial overlap between 1 ^o and 2 ^o prevention (Figure). There was a wide range of risk, such that secondary prevention patients in the 25 ^th and 75 ^th percentiles for overall risk had 5-year event rates of 14.1% and 32.1%, respectively.

Discussion: CV disease prevention could be better targeted in some cases by reducing the distinction between 1 ^o and 2 ^o prevention in favor of a focus on generalized risk.