Heart Failure (HF) is highly prevalent in diabetes patients and is associated with severe morbidity and mortality. Early identification and prediction of onset HF in an asymptomatic patient with stable diabetes have been elusive. Therefore, this study aimed to predict the first occurrence of HF among diabetes patients.

A time-varying Cox model was derived from the Action to Control Cardiovascular Risk in Diabetes (ACCORD) clinical trial to predict the risk of incident CHF, defined by CHF hospitalization. Candidate variables were chosen based on literature reviews and experts consultation, including demographic characteristics, physical exam, laboratory, medical history, drug and healthcare utilization. Predictors were selected from candidate variables using a step-wise algorithm. Model performance was evaluated by Brier score and C-index. The model was transformed into an integer-based scoring algorithm (0~100) for 5-year risk of incident HF.

RESULTS

A total of 9,649 diabetes patients without CHF history were used for model development, with median follow-up of 4 years and 299 CHF events. The model identified several predictors: college education, age at T2DM diagnosed, HbA1c, SBP, BMI, HDL, urine albumin-to-creatinine ratio, hypertension duration, myocardial infarction history, albuminuria history, revascularization history, neuropathy history, end-stage renal disease, cardiovascular medication, any hospitalization and ER visit. The model demonstrated good discrimination (C-index 0.838 [95% CI 0.821-0.855]) and calibration (Brier Score 0.0064 [95% CI 0.006-0.007]) in ACCORD data. The 5-year incident HF increased in a graded fashion with 5-quintile: ≤1% risk in quintile 1 (risk score ≤ 28), 1%-5% risk in quintile 2 (risk score 29-41), 5%-10% risk in quintile 3 (risk score 42-47), 10%-20% risk in quintile 4 (risk score 48-53), and ≥20% in quintile 5 (risk score ≥54).