knowledge transfer


Decision Support for Artificial Ventilator Management via Adaptive Modelling

Submitted by D A Linkens with H F Kwok, M Mahfouf and G Mills,29.11.2001,IBA D


In hospital Intensive Care Units (ICU) the management of artificial ventilators is an important and routine requirement which absorbs much anaesthetist and nursing time. A computer-based decision-support system offers potential advantages, and particularly when the patient is being weaned from the ventilator to their own spontaneous breathing. Patient-specific models (either quantitative or qualitative) are necessary for such decision-support tools. The ventilator control variables are the inspired oxygen level, the tidal volume, respiratory rate, peak pressures and the mode of ventilation. The decisions are based on measurements of blood gases (via sampling), pH, and respiratory mechanics (e.g. lung resistance and compliance).


A detailed lung model for blood gas dynamics called SOPAVent is the basis for the decision-support advice. It has been validated against clinical data for a range of patient lung abnormalities. To make the model patient-specific requires the on-line estimation of two particular parameters, being dead-space and shunt, since they have both intra and inter patient variability. To provide on-line advice it is necessary to have a method of parameter prediction which is not dependent on blood sampling and subsequent analysis. For shunt estimation it has been shown that there is good correlation between it and a hypoxaemia index, called the respiratory index (RI) which is the ratio of the alveolar-arterial oxygen difference to the arterial oxygen tension. This can be measured on-line. For estimation of the dead-space, a similar index based on carbon dioxide measurements is being investigated.

Status and results

The shunt estimator has been validated against 9 patients in an ICU database, with 117 sets of blood gas measurement. A similar approach is being used for dead-space estimation. However, there will always be some error between the adapted SPOAVent model and the specific patient. To adjust for this discrepancy, a grey-box model is being constructed which incorporates a black-box dynamic model in parallel with the physiologically-based SOPAVent model. This will be based on population average discrepancy data, thus leaving the adaptivity in the shunt and dead-space on-line adjustment.

Adaptivity and portability

This system achieves Level 1 adaptivity in the EUNITE definition since it allows for disturbances to patient characteristics and subsequent alterations in the decision-support advice. It may become Level 2 adaptable if the SPOAVent model can be made adjustable to various abnormal lung conditions, such as pneumonia and ARDS (Accute Respiratory Distress Syndrome).

More information

Further details are given in:

H F Kwok D A Linkens, M Mahfouf and G Mills, "An adaptive approach to respiratory shunt estimation for decision-support in ICU",EUNITE 2001,Annual Symposium on "Intelligent Technologies, Hybrid Systems and their Implementation in Smart Adaptive Systems",Tenerife,13 &14 December 2002.

technical reports & case studies

executive summaries

scientific papers




search eunite with