Identifying patients at high risk of death with novel computational biomarkers

Abstract

The accurate assessment of a patient's risk of adverse events remains a mainstay of clinical care for patients with cardiovascular disease. Commonly used risk metrics have traditionally been based on simple models that incorporate various aspects of the medical history, presenting signs and symptoms, and lab values. More sophisticated methods, such as those based on signal processing and machine learning, form an attractive platform to build improved risk metrics because they can offer deeper insights into aspects of clinical data that cannot be approached by simpler methods. In particular, generalized additive models can exhibit comparable or superior performance to conventional logistic regression models, while simultaneously providing potentially useful prognostic information on the relationship between patient outcomes and clinical variables. Moreover, artificial neural networks can provide a convenient formalism for combining rich time series information from physiological signals with more conventional patient features. In this work, models based on signal processing and machine learning techniques are developed and applied to two independent datasets consisting of post-acute coronary syndrome patients to predict cardiovascular death at various endpoints. The models are shown to successfully identify high-risk patients, thereby demonstrating the potential of these techniques to improve the management of patients with cardiovascular disease.