ARX modeling of drug effects on brain signals during general anesthesia

The effect of drugs' interaction on the brain signal Bispectral Index (BIS) of the EEG, is of great importance for an anesthesia control drug infusion system. In this study, the objective was to investigate if an autoregressive with exogenous inputs model (ARX) could be a suitable approach to predicting BIS according to the anesthetic drugs concentrations. Data were collected in 45 neurosurgeries with total intravenous anesthesia every 5s. A stochastic ARX model was fitted to the data of each patient. The models structure that performed better as predictor used a 30s lag for BIS, 1min lag for propofol and 2min lag for remifentanil. The models had a good performance with statistical zero errors (P < 0.05) in 31 patients. The average of absolute errors was 8.2 ± 2.5, showing that the model captures the brain signal trend. This model proved to be effective in modeling and one step prediction of the BIS signal capturing unique characteristics. The results show that the previous brain response trend has influence on the present value, in addition the drugs concentrations from the previous 2min still have influence. This is an important conclusion for the development of drug infusion controller algorithms.