Multivariable Regression Model to Estimate Tidal Volume for Different Respiratory Patterns.

Abstract

Respiratory patterns present great variability, both in healthy subjects and in patients with different diseases and forms of nasal, oral, superficial or deep breathing. The analysis of this variability depends, among others, on the device used to record the signals that describe these patterns. In this study, we propose multivariable regression models to estimate tidal volume (V_T) considering different breathing patterns. Twenty-three healthy volunteers underwent continuous multisensor recordings considering different modes of breathing. Respiratory flow and volume signals were recorded with a pneumotachograph and thoracic and abdominal respiratory inductive plethysmographic bands. Several respiratory parameters were extracted from the volume signals, such as inspiratory and expiratory areas (Area_ins, Area_exp), maximum volume relative to the cycle start and end (VT_ins, VT_exp), inspiratory and expiratory time (T_ins, T_exp), cycle duration (T_tot), and normalized parameters of clinical interest. The parameters with the greatest individual predictive power were combined using multivariable models to estimate V_T. Their performance were quantified in terms of determination coefficient (R²), relative error (E_R) and interquartile range (IQR). Using only three parameters, the results obtained for the thoracic band (VT_exp, T_tot, Area_exp) were better than those obtained from the abdominal band (VT_exp, T_ins, Area_ins) with R² = 0.94 (IQR: 0.07); E_R = 6.99 (IQR: 6.12) vs R² = 0.91 (IQR: 0.09), E_R = 8.70 (IQR: 4.62). Overall performance increased to R² = 0.97 (IQR: 0.02) and E_R = 4.60 (IQR: 3.68) when parameters from the different bands were combined, further improving when was applied to segments with different inspiration-expiration patterns. In particular, the nose-nose E_R = 1.39 (IQR: 0.73), nose-mouth E_R = 2.11 (IQR: 1.23) and mouth-mouth E_R = 2.29 (IQR: 1.44) patterns showed the best results compared to those obtained for basal, shallow and deep breathing.Clinical relevance- Respiratory pattern variability can be described using multivariable regression model for tidal volume.