The incremental value of angiographic features for predicting recurrent cardiovascular events: Insights from the Duke Databank for Cardiovascular Disease.

BACKGROUND AND AIMS: Identifying patient subgroups with cardiovascular disease (CVD) at highest risk for recurrent events remains challenging. Angiographic features may provide incremental value in risk prediction beyond clinical characteristics.

METHODS

We included all cardiac catheterization patients from the Duke Databank for Cardiovascular Disease with significant coronary artery disease (CAD; 07/01/2007-12/31/2012) and an outpatient follow-up visit with a primary care physician or cardiologist in the same health system within 3 months post-catheterization. Follow-up occurred for 3 years for the primary major adverse cardiovascular event endpoint (time to all-cause death, myocardial infarction [MI], or stroke). A multivariable model to predict recurrent events was developed based on clinical variables only, then adding angiographic variables from the catheterization. Next, we compared discrimination of clinical vs. clinical plus angiographic risk prediction models.

RESULTS

Among 3366 patients with angiographically-defined CAD, 633 (19.2%) experienced cardiovascular events (death, MI, or stroke) within 3 years. A multivariable model including 18 baseline clinical factors and initial revascularization had modest ability to predict future atherosclerotic cardiovascular disease events (c-statistic = 0.716). Among angiographic predictors, number of diseased vessels, left main stenosis, left anterior descending stenosis, and the Duke CAD Index had the highest value for secondary risk prediction; however, the clinical plus angiographic model only slightly improved discrimination (c-statistic = 0.724; delta 0.008). The net benefit for angiographic features was also small, with a relative integrated discrimination improvement of 0.05 (95% confidence interval: 0.03-0.08).

CONCLUSIONS

The inclusion of coronary angiographic features added little incremental value in secondary risk prediction beyond clinical characteristics.