Implementation of a Bayesian based advisory tool for target-controlled infusion of propofol using qCON as control variable

This single blinded randomized controlled trial aims to assess whether the application of a Bayesian-adjusted CePROP (effect-site of propofol) advisory tool leads towards a more stringent control of the cerebral drug effect during anaesthesia, using qCON as control variable. 100 patients scheduled for elective surgery were included and randomized into a control or intervention group (1:1 ratio). In the intervention group the advisory screen was made available to the clinician, whereas it was blinded in the control group. The settings of the target-controlled infusion pumps could be adjusted at any time by the clinician. Cerebral drug effect was quantified using processed EEG (CONOX monitor, Fresenius Kabi, Bad Homburg, Germany). The time of qCON between the desired range (35-55) during anaesthesia maintenance was defined as our primary end point. Induction parameters and recovery times were considered secondary end points and coefficient of variance of qCON and CePROP was calculated in order to survey the extent of control towards the mean of the population. The desired range of qCON between 35 and 55 was maintained in 84% vs. 90% (p = 0.15) of the case time in the control versus intervention group, respectively. Secondary endpoints showed similar results in both groups. The coefficient of variation for CePROP was higher in the intervention group. The application of the Bayesian-based CePROP advisory system in this trial did not result in a different time of qCON between 35 and 55 (84 [21] vs. 90 [18] percent of the case time). Significant differences between groups were hard to establish, most likely due to a very high performance level in the control group. More extensive control efforts were found in the intervention group. We believe that this advisory tool could be a useful educational tool for novices to titrate propofol effect-site concentrations.

Machine learning in anesthesiology: Detecting adverse events in clinical practice

The credibility of threshold-based alarms in anesthesia monitors is low and most of the warnings they produce are not informative. This study aims to show that Machine Learning techniques have a potential to generate meaningful alarms during general anesthesia without putting constraints on the type of procedure. Two distinct approaches were tested - Complication Detection and Anomaly Detection. The former is a generic supervised learning problem and for this a simple feed-forward Neural Network performed best. For the latter, we used an Encoder-Decoder Long Short-Term Memory architecture that does not require a large manually-labeled dataset. We show this approach to be more flexible and in the spirit of Explainable Artificial Intelligence, offering greater potential for future improvement.

Moderate-to-Deep Sedation Using Target-Controlled Infusions of Propofol and Remifentanil: Adverse Events and Risk Factors: A Retrospective Cohort Study of 2937 Procedures

BACKGROUND

In the University Medical Center Groningen in Groningen, the Netherlands, moderate-to-deep sedation is provided by nursing staff trained and supervised by the anesthesia department using protocol-based target-controlled infusions (TCIs) of propofol and remifentanil. The aim of this retrospective cohort study was to investigate the incidence of events with potential adverse health consequences within this service model and the risk factors for the occurrence of these events.

METHODS

We retrospectively interrogated a database containing the computerized anesthetic records of 2937 procedures where moderate-to-deep sedation was provided using TCI administration of propofol and remifentanil between May 2014 and October 2017. The primary outcome measures were the incidence of sedation-related events with potential adverse health consequences and risk factors for the occurrence of such events. The events under investigation were unplanned intensive care unit (ICU) admission, need for cardiopulmonary resuscitation (CPR), death on the day of the procedure due to sedation-related events, cardiovascular events (arrhythmias, hypertension, and hypotension), pulmonary events (aspiration, desaturation, unplanned tracheal intubation), anaphylactic or allergic reactions, and the termination of the procedure due to sedation-related events. Cardiovascular and pulmonary events were classified as severe, significant, or moderate. Events were identified by using computer algorithms to search the computerized records from all included procedures.

RESULTS

Data from 2937 procedures were analyzed. No patients suffered catastrophic events (death, need for CPR, or unplanned ICU admission). Thirty-two severe sedation-related events occurred in 32 procedures. Severe desaturation (0.6%; 95% confidence interval [CI], 0.4-0.9) and severe hypertension (0.2%; 95% CI, 0.04-0.37) were the most common severe events. Significant hypotension (8.8%; 95% CI, 7.73-9.77) and significant desaturation (1.6%; 95% CI, 1.12-2.02) were found to be the most common events with potential adverse health consequences. No patient suffered lasting health consequences. Average mean and maximum targeted effect-site concentrations (Cet) for propofol were 2.6 ± 0.83 and 3.3 ± 1.09 µg·mL, respectively, and for remifentanil 0.84 ± 0.18 and 0.99 ± 0.22 ng·mL, respectively. Maximum Cets of propofol were lower among patients with higher body mass index (BMI) and were higher among patients of younger age. Higher BMI was a risk factor for desaturation. Increased age and lower BMI were risk factors for hypotension. Longer procedure time was a risk factor for both desaturation and hypotension.

CONCLUSIONS

Moderate-to-deep sedation by propofol and remifentanil TCI has a low incidence of catastrophic and severe events.