Exploring Time Series Retrieved from Cardiac Implantable Devices for Optimizing Patient Follow-Up

Abstract

Current cardiac implantable devices (IDs) are equipped with a set of sensors that can provide useful information to improve patient follow-up and prevent health deterioration in the postoperative period. In this paper, data obtained from an ID with two such sensors (a transthoracic impedance sensor and an accelerometer) are analyzed in order to evaluate their potential application for the follow-up of patients treated with a cardiac resynchronization therapy (CRT). A methodology combining spatiotemporal fuzzy coding and multiple correspondence analysis (MCA) is applied in order to: 1) reduce the dimensionality of the data and provide new synthetic indexes based on the "factorial axes" obtained from MCA; 2) interpret these factorial axes in physiological terms; and 3) analyze the evolution of the patient's status by projecting the acquired data into the plane formed by the first two factorial axes named "factorial plane." In order to classify the different evolution patterns, a new similarity measure is proposed and validated on the simulated datasets, and then, used to cluster observed data from 41 CRT patients. The obtained clusters are compared with the annotations on each patient's medical record. Two areas on the factorial plane are identified, one being correlated with a health degradation of patients and the other with a stable clinical state.