Computer simulation to predict patient responses to alterations in the ventilation regime.

We have developed a technique for simulating artificially ventilated patients using a computer model of the respiratory system and have investigated its ability to predict the patient's response to alterations in the ventilation regime. The majority of those studied were in a stable condition following cardiac surgery. Predictions for PaO2 showed a reasonable correlation with measured values (r = 0.94), although individual predictions could differ appreciably (SEE = +/- 17.3 mmHg). Predictions for PaCO2 correlated less well (r = 0.89, SEE = +/- 2.0 mmHg) and there was a consistent tendency to overestimate this variable. The least accurate predictions were those for PvO2 (r = 0.61, SEE = +/- 3.8 mmHg). Errors in prediction were attributable to deficiencies in the model (most importantly the three-compartment lung and the modelling of CO2 production, storage and excretion), compounded by unpredictable alterations in the patient's physiology (mainly Vo2, Vco2 and Qt). However, allowing for accepted clinical variability and routine measurement accuracy, the predictions are generally reasonable. This suggests that, with some further modifications, computer modelling of this, and other, systems may prove to be a clinically and educationally useful technique.