Estimating ASCVD risk in patients with diabetes could guide intensity of risk factor treatment, but existing risk calculators (2013 ACC/AHA ASCVD pooled cohort equations [PCE] and RECODe risk equations [RECODe]) have not been evaluated in a contemporary American population with diabetes. Using validated phenotyping algorithms, we evaluated prediction models for the development of a composite “hard” ASCVD outcome (non-fatal myocardial infarction, non-fatal stroke, cardiovascular death) over 5 years in 184,823 diabetic U.S. Veterans without baseline ASCVD receiving primary care in the VA 2002-2007 (2.4% women, 19.5% black, mean age 61 years, 9851 ASCVD events). We compared discrimination using the C-statistic and calibration using the Greenwood-Nam-D’Agostino test (GND) of the PCE variables (sex, age, race, cholesterol, blood pressure [BP], smoking status, BP medications, statin treatment, and aspirin), RECODe, and a series of models using PCE variables with VA sample-specific coefficients. Models with PCE variables and VA-specific coefficients had better discrimination (C=0.642) than the PCE (C=0.582) or RECODe (C=0.633). Successive addition of hemoglobin A1c, estimated glomerular filtration rate, diabetes medications, and statin therapy improved the model (full model, C=0.669), with only minimal enhancement by semisupervised variable selection using LASSO (C=0.672). The C-statistic of the full model was similar in those <60 or ≥60 years old, but predicted risk was higher for black than for white patients, and for men than for women. The full model also had improved calibration (GND p=0.82) compared to the PCE and RECODe (p<0.0001 for both; higher p indicates better calibration).
Conclusion: Health system-specific tailoring of prediction equations provides improved discrimination and calibration compared to existing models. This risk predictor could be embedded in the electronic health record and used to guide intensification of treatment aimed to improve ASCVD outcomes.
S. Raghavan: None. Y. Ho: None. D. Posner: None. J.L. Vassy: None. D.R. Gagnon: None. L.S. Phillips: Advisory Panel; Self; Janssen Pharmaceuticals, Inc. Research Support; Self; AbbVie Inc., GlaxoSmithKline plc., Kowa Pharmaceutical Europe Co. Ltd., Novartis Pharmaceuticals Corporation, Novo Nordisk Inc., Pfizer Inc. Stock/Shareholder; Self; Diasyst Inc. Other Relationship; Self; Diasyst Inc., Janssen Pharmaceuticals, Inc. P.W. Wilson: None.
U.S. Department of Veterans Affairs; American Heart Association