GRACE PLUS: A data fusion-based approach to improve GRACE score in the risk assessment of Acute Coronary Syndrome.

Cardiovascular Diseases (CVDs) are the world's leading cause of morbidity and mortality, being responsible for almost 17 million deaths each year. In Europe, let alone it is estimated that 20% of all citizens suffer from one form of CVD, namely cerebrovascular disease or heart failure and Coronary Artery Disease (CAD). Among the latter, Acute Coronary Syndrome (ACS) is of particular importance since it is deadly and, hence, requires a prompt diagnosis and immediate medical attention. Aiming to deal with prognostication and promote consistency in managing patients with ACS, the Global Registry of Acute Coronary Events (GRACE) risk score has been proposed. This tool is based on eight independent risk factors roughly accounting for 89.9% of prognostic information. Nevertheless, as some other risk factors, not included in GRACE, are also known to be important vectors in the stratification of patients, namely haemoglobin at admission, it is expected that by embedding additional risk factors information into GRACE it will lead to a better characterisation of a patient's risk. Making use of data-fusion techniques, the present work proposes a generalisable framework to improve the classification performance of GRACE in predicting the risk of death in the course of six months after an ACS event, while preserving its interpretability and applicability. Considering haemoglobin concentration at admission, as an additional risk factor, it is shown that the discrimination performance of new GRACE Plus score outperformed that of GRACE in a database of cohorts comprising 1506 patients admitted with ACS, showing a F-1 score of 0.6033 for GRACE Plus against 0.5828 for GRACE, which is corroborated by one-tailed t-test in terms of correct stratification of death and survival endpoints, namely, t = − 9. 1876 and p < 0. 001. • GRACE neglects the information associated with other relevant risk factors. • Haemoglobin at admission has the potential to increase GRACE predictive performance. • Data fusion techniques allow combining complementary information into a single model. • Machine learning can be used to correct GRACE, while preserving its interpretability. • GRACE++ promotes parsimonious resource management and patient' tailored treatments. [ABSTRACT FROM AUTHOR]