Machine Learning Model Based on Transthoracic Bioimpedance and Heart Rate Variability for Lung Fluid Accumulation Detection: Prospective Clinical Study (Preprint)

Abstract

BACKGROUND Clinically, the most important signs and symptoms of acute decompensated heart failure (ADHF) relate to accumulation of excess body fluid, but autonomic dysregulation is another characteristic feature of ADHF physiology. Transthoracic bioimpedance (TBI) is a non-invasive, simple method for measuring fluid retention in lungs. Heart rate variability (HRV) is another widely used noninvasive tool to assess autonomic function. We hypothesize that TBI and HRV can be used for detection of fluid accumulation in ADHF participants. OBJECTIVE In this paper, we aimed to evaluate the performance of TBI and HRV parameters obtained using a fluid accumulation vest (FAV) with dry carbon black polydimethylsiloxane (CB-PDMS) electrodes in a prospective clinical study ‘System for Heart-failure Identification using an External Lung-fluid Device’ (S.H.I.E.L.D.). METHODS We computed fifteen parameters, eight calculated from the model to fit Cole-Cole plot from TBI measurements (R0, RI, R∞, R0 - R∞, FE, fc, α, and Cm), and seven based on linear (mean HR, HRVLF, HRVHF, HRVLFn, HRVHFn) and nonlinear (PDMSymp, and PDMPSymp) analysis of HRV. We compared the values of these parameters between three groups of participants: Control (non-HF hospitalized participants), Baseline (ADHF participants’ recordings taken at the time of admittance to the hospital), and Discharge (ADHF participants’ recordings acquired at the time of discharge from hospital). RESULTS Among the fifteen parameters, two TBI (R0 and R0-R∞) and three HRV (HRVHF, HRVLFn, and HRVHFn) parameters were found to be the most discriminatory between non-HF and ADHF groups. The two TBI parameters had statistically significantly lower values for ADHF participants than for non-HF participants, which is an indicator that accumulated fluids in the lungs are of higher volume for HF participants. We used several machine learning approaches to classify participants with fluid accumulation (Baseline ADHF) and without fluid accumulation (Control and ADHF participants at discharge), termed Wet vs. Dry groups, respectively. A cubic support vector machine model using TBI and HRV parameters achieved an accuracy of 92% classifying Wet and Dry groups. Looking at the parameters included in the model, the TBI parameters are related to intra and extra-cellular fluid, whereas the HRV parameters are mostly related to sympathetic activation. CONCLUSIONS This is useful, for instance, to provide in-home diagnostic wearable vest that can detect or predict fluid accumulation in HF participants. Results suggest that fluid accumulation, detection, and subsequently ADHF detection, could be performed using TBI and HRV measurements acquired with a wearable vest.