The Use of a Patient Simulator for Knowledge Acquisition from the Clinicians

The first objective of this study is to observe how the clinicians adjust ventilator settings in simulated mechanically ventilated patients. This information was then used to construct the fuzzy rule-bases of a ventilator advisory system. A simulator of patients on artificial ventilation was developed using the MATLAB/SIMULINK environment based on a grey-box mathematical model of the human respiratory system. A graphic user interface (GUI) was designed to improve its usability. Eleven simulated patients were constructed. Each simulated patient consists of 7 to 44 tuned simulated events based on the data recorded in the intensive care unit (ICU) patient data management system (PDMS). Four ICU consultants were invited to participate in the simulation study. The fuzzy input partitions for the rule-bases were derived from the probability distribution of the acceptable input values defined by the clinicians. The rules were derived after correlation analysis of the simulation results and supplemented by the review of the rule-bases by a clinical expert. The rule-bases were validated using independent retrospective clinical data. The performance measures for the FiO2, PEEP and Ventilatory rate rule-bases were good as in more than 70% of the cases, the fuzzy rulebase output matched exactly the clinician’s action recorded in the PDMS. The PINSP rule-base did not perform well. In the future, further tuning of the rule-bases should be done to improve its performance.