PREDICTING SIX-YEAR CORONARY HEART DISEASE RISK IN DIABETIC PATIENTS

Purpose

The objective of this study was to create a tool that accurately predicts the risk of Coronary Heart Disease (CHD) in Type II diabetic patients, while adjusting for oral hypoglycemic agent.

Methods

This study was based on a cohort of 23,901 type 2 diabetic patients identified in the Cleveland Clinic (CC) electronic health record (EHR). Patients prescribed an oral hypoglycemic agent between 1998 and 2006 were eligible for inclusion. The study was limited to patients who were being prescribed a single oral agent at baseline. Patients with a history of CHD or dialysis were excluded. Baseline information about past medical history, medication history, and social history were extracted from the outpatient EHR. Laboratory values were included from both inpatient and outpatient records. The most recent historical lab value was considered the baseline value. If no laboratory value was available for a specific variable, then the first value obtained up until 21 days after baseline was used. Missing values were imputed using multiple imputation with chained equations. A Cox proportional hazards regression model for time to CHD (documentation of CHD or revascularization procedure) was created using 22 predictor variables that were chosen for their clinical importance. The prediction model was validated using random subsets of the cohort that were systematically not used to create the prediction model (‘ten-fold cross validation’).

Results

The length of follow-up in the cohort ranged from 1 day to 8 years (median = 15.2 months), and 2,897 CHD events were observed. The prediction tool had a concordance index (i.e. c-statistic) of 0.724, which indicates that the tool accurately discriminated 72.4% of the discordant pairs of patients. The medication class associated with the lowest risk of CHD depends upon the age of the patient due to an interaction between age and medication in the model.  For a patient who is 50 years old, the biguanides were associated with the lowest risk of primary CHD.

Conclusions

This study successfully used a clinical EHR to create an accurate prediction of CHD risk in diabetics and is the first CHD prediction tool to incorporate medication class as a predictor. Improved CHD risk assessment in diabetics could help improve CHD prevention.