computer methods and programs in biomedicine 101 ( 2 0 1 1 ) 126–134 journal homepage: www.intl.elsevierhealth.com/journals/cmpb Analysis of multiple linear regression algorithms used for respiratory mechanics monitoring during artificial ventilation Adam G. Polak ∗ Chair of Electronic and Photonic Metrology, Wroclaw University of Technology, ul. B. Prusa 53/55, 50-317 Wroclaw, Poland article info Article history: Received 10 December 2009 Received in revised form 28 July 2010 Accepted 3 August 2010 Keywords: Artificial ventilation Respiratory mechanics monitoring Multiple linear regression Forward-inverse modelling Accuracy analysis abstract Many patients undergo long-term artificial ventilation and their respiratory system mechan- ics should be monitored to detect changes in the patient’s state and to optimize ventilator settings. In this work the most popular algorithms for tracking variations of respiratory resis- tance (R rs ) and elastance (E rs ) over a ventilatory cycle were analysed in terms of systematic and random errors. Additionally, a new approach was proposed and compared to the previ- ous ones. It takes into account an exact description of flow integration by volume-dependent lung compliance. The results of analyses showed advantages of this new approach and enabled to form several suggestions. Algorithms including R rs and E rs dependencies on airflow and lung volume can be effectively applied only at low levels of noise present in measurement data, otherwise the use of the simplest model with constant parameters is preferable. Additionally, one should avoid including the resistance dependence on airflow alone, since this considerably destroys the retrieved trace of R rs . Finally, the estimated cyclic trajectories of R rs and E rs are more sensitive to noise present in pressure than in the flow signal, and the elastance traces are estimated more accurately than the resistance ones. © 2010 Elsevier Ireland Ltd. All rights reserved. 1. Introduction Although artificial ventilation is generally a rescue therapy, there are many patients who require long-term ventilatory support. Knowledge of their respiratory system mechanics is of great importance for two main purposes: to detect changes in the patient’s state, so pharmacological therapy can be administered, and to adjust the ventilator settings in order to minimize the mechanical stress and to prevent ventilator induced lung injury. The main approach to the assessment of respira- tory mechanics consists in matching the equation of a single-compartment first-order model characterized by total ∗ Tel.: +48 71 320 65 81; fax: +48 71 321 42 77. E-mail address: Adam.Polak@pwr.wroc.pl. respiratory resistance and elastance (or its inverse – compli- ance) to the measurements of pressure, flow, and volume, using multiple linear regression (MLR). The model assumes constant parameters over the data (e.g. [1–5]), resistance dependence on flow or volume [1,2,4–6], volume-dependent elastance [2,4,5,7,8] or all the mentioned effects together [5]. Usually data from the last ventilatory cycle are used by batch algorithms (working off-line with a set of past measurements) to estimate latest values of respiratory resistance and elas- tance. The success of this approach in routine clinical practice is due to its simplicity, immediate physiological interpretation, and sensitivity to changes in lung mechanics. Although the dependencies of pulmonary parameters on flow and volume are generally nonlinear, the numerical algorithms assume lin- 0169-2607/$ – see front matter © 2010 Elsevier Ireland Ltd. All rights reserved. doi:10.1016/j.cmpb.2010.08.001