Plateau Pressure Prediction in ARDS Patients. Issue 1 (July 2017)

• Record Type: Journal Article
• Authors: Langdon, R.; Docherty, P.D.; Schranz, C.; Chase, J.G.
• Is Part Of: IFAC-PapersOnLine. Volume 50:Issue 1(2017)
• Page Start: 5480
• Page End: 5485
• Publication Date: 2017-07
• Subjects: Physiological models -- Physiology -- Biomedical control -- Parameter identification -- Parameter optimisation -- Pressure volume relationships; Automatic control -- Periodicals; 629.805
• DOI: 10.1016/j.ifacol.2017.08.1086
• Languages: English
• ISSNs: 2405-8963

Abstract

Abstract: In mechanical ventilation, the optimal level of positive end expiratory pressure (PEEP) is widely debated in the treatment of acute respiratory distress syndrome (ARDS). While PEEP is often necessary to maintain recruitment, excessively high PEEP can cause high pressure at the alveoli, indicated by the measurement of plateau pressure (PP). High PP can indicate over distension and damage to healthy alveoli, which can lead to ventilator induced lung injury (VILI). Model based methods allow the estimation of patient specific parameters that can aid clinicians in selecting optimal patient specific PEEP. A model that accurately and precisely predicts the outcome of an increase in PEEP may allow dangerously high PP to be avoided. This paper examines the PP prediction capability of two variations of a model of respiratory mechanics that uses basis functions to capture pulmonary elastance. BFM(I), uses four basis functions, and BFM(II) uses two. The models were identified on one or more adjacent PEEP steps, and extrapolated to predict PP at PEEP levels that were 2, 4, and 6 cmH2 O above the highest PEEP in the training data. A comparison using the same training data was made using the first order model of pulmonary mechanics (FOM(I)). A further method more consistent with typical use of the FOM was also tested, in which the FOM was trained on a single PEEP prior to prediction (FOM(II)). The method was conducted on 23 ARDS patients that underwent a recruitment manoeuvre. All four models yielded high specificity in all scenarios (≥ 0.90). The FOM(I), FOM(II) and BFM(I) sensitivity were similar across all prediction horizons, and generally decreased as the prediction PEEP increased. However the reduced parameterization of BFM(II) compared with BFM(I) allowed sensitivity to remain high at ≥ 0.95 at all prediction horizons. The results highlight the importance of finding a balance between model fitting and predictive ability when considering optimal model parameterization.