P6420Can machine learning help us improve risk stratification of diabetic patients with acute coronary syndromes? The answer will blow your mind

Background Risk stratification following an acute coronary syndrome (ACS) is of utmost importance, in order to identify patients at higher risk of subsequent cardiovascular events. Diabetic patients have a significantly worse prognosis, so new risk prediction tools are important to better identify and risk stratify high risk patients within this important ACS subpopulation. Aim The aim of this study was to identify the best predictors of a new ACS, in a single-center database of ACS, resorting to machine learning and artificial intelligence, and to compare the Global Registry of Acute Coronary Events (GRACE) risk score's relevance for risk discrimination in a general ACS population versus a subpopulation of diabetic patients. Methods In a single center, 5977 patients admitted due to ACS between 2004 and 2017 and alive at discharge were studied. In the subpopulation of diabetic patients (n=3429), each covariate present in the database was analyzed separately with a Cox proportional hazard model with three terms – subpopulation belonging indicator, covariate, interaction term. The p-value of the interaction term was used to rank variables. The more significant the interaction term, the stronger the change in relationship between patients in the subpopulation and the risk of a new ACS, compared to the one in the general population. Results During long term follow-up, 13% of patients (n=771) experienced a second event. Kaplan-Meier curve represents how ACS free-survival depends on the GRACE risk score and group of interest. In the general population and in the subpopulation of diabetic patients, the GRACE score was used to further divide patients into 3 terciles, of which only the lower and upper tercile are shown (GRACE ≤113 and GRACE >144, respectively). The solid lines represent Kaplan-Meier curves for diabetic patients, and the dotted lines in the general population. Pink or grey colour of the curves represent the stratification level of the covariate. Conclusions In our model, the GRACE risk score was found to be a better discriminator of risk of futher ACS in diabetic patients than in the general ACS population. Strikingly, a higher GRACE score predicts a lower rate of readmission, probably because many patients will die in the index hospitalization or out of hospital. This finding reinforces the usefulness of the GRACE score in high risk patients and may improve risk stratisfication in diabetic post-ACS patients, making sure that they are closely followed and submitted to optimal risk factor management, in order to improve their post-ACS prognosis.