Determining the Optimal Calibration Point for the Clinical Application of a Simple Model of Human Gas Exchange

Jörn Kretschmer, Axel Riedlinger, and Knut Möller


Simulation and optimization, Mathematical modelling, Medical decision support, Model calibration


Mathematical models of the human physiology allow both the identification of hidden patient parameters and the prediction of patient reactions towards changes in the therapy regime. The knowledge of hidden patient parameters allows a doctor to further evaluate a patient’s stage of illness, while the prediction of patient reactions allows optimization of therapy settings to reach a desired therapy goal. Such physiological modeling might be applied to mechanically ventilated patients in order to optimize the applied ventilator settings. Mechanical ventilation is a life-saving routine therapy; however it bears the risk of further injuries if ventilator settings are not set properly. We have previously presented a simple model of human gas exchange that is able to predict the inspired oxygen fraction (FiO2) to reach a desired level of arterial oxygen partial pressure (PaO2) in a patient. The model is calibrated using one blood gas measurement, thus measurement noise may strongly influence predictive quality. We therefore applied artificial and real patient data to determine the optimal calibration point where influence of measurement noise is minimal and prediction quality is optimal. Results show, that prediction quality strongly increases when calibration is done at 60% to 70% FiO2 or 150mmHg to 200mmHg PaO2, respectively.

Important Links:

Go Back