An introduction to predictive modelling of drug concentration in anaesthesia monitors

Influence of an "Electroencephalogram-Based" Monitor Choice on the Delay Between the Predicted Propofol Effect-Site Concentration and the Measured Drug Effect

BACKGROUND

Clinicians can optimize propofol titration by using 2 sources of pharmacodynamic (PD) information: the predicted effect-site concentration for propofol (Ceprop) and the electroencephalographically (EEG) measured drug effect. Relation between these sources should be time independent, that is, perfectly synchronized. In reality, various issues corrupt time independency, leading to asynchrony or, in other words, hysteresis. This asynchrony can lead to conflicting information, making effective drug dosing challenging. In this study, we tried to quantify and minimize the hysteresis between the Ceprop (calculated using the Schnider model for propofol) and EEG measured drug effect, using nonlinear mixed-effects modeling (NONMEM). Further, we measured the influence of EEG-based monitor choice, namely Bispectral index (BIS) versus qCON index (qCON) monitor, on propofol PD hysteresis.

METHODS

We analyzed the PD data from 165 patients undergoing propofol-remifentanil anesthesia for outpatient surgery. Drugs were administered using target-controlled infusion (TCI) pumps. Pumps were programmed with Schnider model for propofol and Minto model for remifentanil. We constructed 2 PD models (direct models) relating the Schnider Ceprop to the measured BIS and qCON monitor values. We quantified the models' misspecification due to hysteresis, on an individual level, using the root mean squared errors (RMSEs). Subsequently, we optimized the PD models' predictions by adding a lag term to both models (lag-time PD models) and quantified the optimization using the RMSE.

RESULTS

There is a counterclockwise hysteresis between Ceprop and BIS/qCON values. Not accounting for this hysteresis results in a direct PD model with an effect-site concentration which produces 50% of the maximal drug effect (Ce50) of 6.24 and 8.62 µg/mL and RMSE (median and interquartile range [IQR]) of 9.38 (7.92-11.23) and 8.41(7.04-10.2) for BIS and qCON, respectively. Adding a modeled lag factor of 49 seconds to the BIS model and 53 seconds to the qCON model improved both models' prediction, resulting in similar Ce50 (3.66 and 3.62 µg/mL for BIS and qCON) and lower RMSE (median (IQR) of 7.87 (6.49-9.90) and 6.56 (5.28-8.57) for BIS and qCON.

CONCLUSIONS

There is a significant "Ceprop versus EEG measured drug effect" hysteresis. Not accounting for it leads to conflicting PD information and false high Ce50 for propofol in both monitors. Adding a lag term improved the PD model performance, improved the "pump-monitor" synchrony, and made the estimates of Ce50 for propofol more realistic and less monitor dependent.

Investigation of data-driven optical neuromonitoring approach during general anesthesia with sevoflurane

Anesthesia monitoring currently needs a reliable method to evaluate the effects of the anesthetics on its primary target, the brain. This study focuses on investigating the clinical usability of a functional near-infrared spectroscopy (fNIRS)-derived machine learning classifier to perform automated and real-time classification of maintenance and emergence states during sevoflurane anesthesia. For 19 surgical procedures, we examine the entire continuum of the maintenance-transition-emergence phases and evaluate the predictive capability of a support vector machine (SVM) classifier during these phases. We demonstrate the robustness of the predictions made by the SVM classifier and compare its performance with that of minimum alveolar concentration (MAC) and bispectral (BIS) index-based predictions. The fNIRS-SVM investigated in this study provides evidence to the usability of the fNIRS signal for anesthesia monitoring. The method presented enables classification of the signal as maintenance or emergence automatically as well as in real-time with high accuracy, sensitivity, and specificity. The features local mean HbTotal, std [Formula: see text], local min Hb and [Formula: see text], and range Hb and [Formula: see text] were found to be robust biomarkers of this binary classification task. Furthermore, fNIRS-SVM was capable of identifying emergence before movement in a larger number of patients than BIS and MAC.