Personalization based on grouping strategies for short-term cardiovascular event risk assessment

Abstract

Cardiovascular diseases are the main cause of death in all Europe, representing 47% of themortality rate. This number could decrease, if each patient would undergo the most adequate treatment. For this to happen, it is important to determine the patient's risk of having a cardiovascular event. This is known as risk assessment, and can be done by tools called risk scores. However, there are several risk scores with similar classification ability, which makes difficult the process of choosing the most adequate risk score for each situation. One possible solution would be a combination of the risk scores. This could improve the overall classification, while eliminating the need to choose from among them. This paper uses this concept to develop a combination of the risk scores using a personalization based on groups. New patients are be assigned to the most similar group and consequently to a risk score, improving the overall classification. This eliminates the need to choose a standard, and improves the overall performance. Using a dataset of patients made available by the Santa Cruz Hospital it was possible to maintain the highest sensitivity among the risk scores used (GRACE, TIMI, PURSUIT), while improving specificity by 13%.