Model-based administration of inhalation anaesthesia. 3. Validating the system model

Background. We quantified the predictive performance of our computer model of the administration of inhalation anaesthesia from a Datex-Ohmeda Modulus CD circle-absorber system. Methods. In 50 patients, desflurane anaesthesia was maintained with a fresh-gas flow (FGF) of 0.5 litres min(-1) of both nitrous oxide and oxygen, preceded by fast (n=14) or slow (n=36) induction: FGF greater than total ventilation, Group F; FGF equal to 1.0 litres min(-1), Group S. The two versions of the model studied differed in the size of their inter-tissue diffusion, as 0 (version 1) and 3% (version 2) of the cardiac output was shifted from the viscera to adipose tissue. Model performance was judged by comparing measured and predicted gas concentrations in terms of three variables for each gas concentration in each patient: root mean squared error (rmse=total error), bias (mean predicted - measured) (systematic error), and scatter (error around bias). These variables were then averaged over all patients. These measures were calculated overall, and separately for each group and each stage (1 =induction or 2=maintenance). Results. Model predictions were in reasonable to very good agreement with clinically obtained data. Version 2 performed better than version 1. Differences between groups were not demonstrated. The model performed better for stage 2, but only for desflurane. In group S, results (mean (SD); as percentages of the measured values for nitrous oxide, oxygen and desflurane) in the order rmse, bias, and scatter were for end-tidal concentrations of nitrous oxide; 8 (4), 8 (5), 2 (1)%; oxygen: 11 (4), -10 (6), 2 (1. 1)%; nitrogen: 0.9 (0.6), -0.8 (0.6), 0.2 (0. 1) vol%; carbon dioxide: 1.8 (0.6), 1.8 (0.6), 0.2 (0. 1) vol%; desflurane, stage 2: 8 (4), 4 (7), 4 (2)%, vs 15 (6), - 10 (8), 9 (4)% for stage 1. Conclusion. Administration of inhalation anaesthesia can be based on version 2 of this model, but must be guided by active monitoring.