Use of basis functions within a non-linear autoregressive model of pulmonary mechanics

Abstract Patients suffering from acute respiratory distress syndrome (ARDS) require mechanical ventilation (MV) for breathing support. A lung model that captures patient specific behaviour can allow clinicians to optimise each patient's ventilator settings, and reduce the incidence of ventilator induced lung injury (VILI). This study develops a nonlinear autoregressive model (NARX), incorporating pressure dependent basis functions and time dependent resistance coefficients. The goal was to capture nonlinear lung mechanics, with an easily identifiable model, more accurately than the first order model (FOM). Model coefficients were identified for 27 ARDS patient data sets including nonlinear, clinically useful inspiratory pauses. The model successfully described all parts of the airway pressure curve for 25 data sets. Coefficients that captured airway resistance effects enabled end-inspiratory and expiratory relaxation to be accurately described. Basis function coefficients were also able to describe an elastance curve across different PEEP levels without refitting, providing a more useful patient-specific model. The model thus has potential to allow clinicians to predict the effects of changes in ventilator PEEP levels on airway pressure, and thus determine optimal patient specific PEEP with less need for clinical input or testing.