Abstract 3178: Improvement in the Prognostic Ability of a Validated Post-MI Risk Model with the Addition of Health Status Measures
The GRACE mortality prediction model used at the time of discharge after hospitalization for a myocardial infarction (MI) is a validated means of stratifying patients’ risk. We examined whether the addition of other variables known to be associated with mortality, diabetes and health status, could further improve the stratification of patients’ risk. We studied 2,481 MI patients who survived hospitalization in the prospective, 19-center PREMIER study. A risk prediction model for 3-year mortality was constructed by combining patient health status scores and diabetes with the GRACE mortality risk score. The net improvement in the predictive value of the model was quantified using the integrated discrimination improvement (IDI) statistic, which measures the average increase in sensitivity of the model offset by any average decrease in specificity. The Short Form-12 (SF-12) mental component score (MCS) and physical component score (PCS) at baseline were used to measure health status. The GRACE model c-statistic for discriminating 3-year mortality in the PREMIER cohort was 0.759 (versus 0.75 in published validation). With the combination of the SF-12 score with the GRACE score, the c-statistic improved to 0.79, increasing IDI by 3%. IDI increased another 1.2% with the addition of diabetes, for a total improvement in IDI of 4.2% and a c-statistic of 0.798. Addition of SF-12 scores increased predicted risks of death by 2.4% (absolutely, on average) among patients who died and decreased them by 0.6% among survivors. Adding diabetes further changed predicted risks of death by +1.0% and −0.2%, respectively. Addition of health status scores and diabetes to the GRACE mortality risk model significantly improves its ability to predict 3-year mortality after MI. Routine use of these models may improve medical decision making and patient-physician communication.