PREDICTING SIX-YEAR HEART FAILURE RISK IN DIABETIC PATIENTS

Purpose: The purpose of this project was to create a web-based tool for predicting the risk of congestive heart failure (CHF) in patients with type 2 diabetes mellitus.

Method: This study was conducted on a retrospective cohort of 28,862 patients with type 2 diabetes identified in the Electronic Health Record at the Cleveland Clinic. All of the patients were being treated with a single oral anti-diabetic medication (biguanide, meglitinide, sulfonylurea, or thiazolidinedione) at baseline and were followed for the development of CHF for up to 6 years. Patients with a history of CHF or baseline left ventricular ejection fraction <40% were excluded. We considered 17 predictor variables based on their theoretical association with the development of CHF in patients with diabetes. Missing values were imputed without regard to the outcome using multiple imputations with chained equations. A competing risk regression model was used to predict the time to development of CHF. The “full” model was reduced using the “stepdown” method suggested by Harrell. In this method, the full model is fit in order to obtain the linear predictor, which is then used as the outcome for an ordinary least squares (OLS) regression using the same variables. The variables are ranked according to their importance by determining their impact on the correlation coefficient in the OLS model. The variable ranking is then used to find the survival model that has the best accuracy in terms of the bootstrap corrected concordance index. The most accurate reduced model contained the following 14 variables: estimated glomerular filtration rate, history of coronary artery disease, hemoglobin A1C, body mass index, systolic blood pressure, smoking status, insulin, aspirin, household income, age, and oral anti-diabetic medication class. The model coefficients were used to create a web based version of this calculator that is available at: http://simpal.com/RCC/links/CHF_DM2.html.

Result: There were 1214 patients who developed evidence of new CHF. Following 10-fold cross validation, the model showed good discrimination (c index =0.747) and calibration (assessed graphically). 

Conclusion: CHF is a common complication in patients with type 2 diabetes and this is the first prediction tool to address this issue. Models such as this will be increasingly used in the future to guide anti-diabetic therapy and will hopefully lead to improved clinical outcomes.