Behavioral and autonomic dynamics during propofol-induced unconsciousness

Abstract

During general anesthesia, both behavioral and autonomic changes are caused by the administration of anesthetics such as propofol. Propofol produces unconsciousness by creating highly structured oscillations in brain circuits. The anesthetic also has autonomic effects due to its actions as a vasodilator and myocardial depressant. Understanding how autonomic dynamics change in relation to propofol-induced unconsciousness is an important scientific and clinical question since anesthesiologists often infer changes in level of unconsciousness from changes in autonomic dynamics.

Therefore, we present a dataset combining physiology-based indices for heart rate variability and electrodermal activity with a robust statistical tool to compare behavioral and multimodal autonomic changes before, during, and after propofol-induced unconsciousness. This dataset was recorded from nine healthy volunteers during computer-controlled administration of propofol. 

Background

During general anesthesia, both behavioral and autonomic changes are caused by the administration of anesthetics. Therefore, it is reasonable to hypothesize that the brain circuits affecting behavior and autonomic activity are invoked in parallel, causing observable effects in parallel. However, the neural circuitry governing autonomic activity and behavioral changes, while complex and interconnected, is separate [1-2]. Understanding this relationship between autonomic dynamics and behavior before, during, and after anesthesia has real implications for clinical practice.

For example, during surgery, consciousness is most often directly assessed through behavioral markers. Autonomic markers present a valuable additional source of information, but they cannot be assumed to be direct equivalents to behavioral markers. Once the patient is administered a muscle relaxant, anesthesiologists who are not using the EEG or EEG-based indices infer changes in level of unconsciousness by tracking changes in autonomic dynamics, namely heart rate and blood pressure. This is provided that there are no overt changes in respiratory or cardiovascular state due to surgical or non-anesthetic pharmacological causes such as vasopressor or vasodilators. Therefore, a systematic exploration of the relationship between autonomic dynamics and consciousness in anesthesia is a highly relevant scientific and clinical question.

Propofol, one of the most common anesthetics, affects the brain to produce unconsciousness by acting primarily on GABAergic circuits to broadly inhibit neuronal firing in the cortex, thalamus, and brainstem [1-2]. In addition to the behavioral changes associated with unconsciousness, propofol has a variety of autonomic effects. It is also a vasodilator, which leads to a decrease in blood pressure. Due to the baroreflex, while there may be a transient increase in heart rate, over time, propofol is a myocardial depressant, and thus, decreases heart rate [1-2]. Finally, experiments have shown that propofol and many other anesthetics reduce baseline filling levels of sweat glands and increase thresholds for spontaneous and evoked sweating activity [3]. Advances in non-invasive sensing modalities and statistical signal processing methods have created the opportunity for detailed study of the interplay between behavioral and autonomic dynamics during propofol-induced unconsciousness.

In this dataset, we characterize autonomic state using a multimodal model that considers both heart rate variability (HRV) and electrodermal activity (EDA) to capture sympathetic and parasympathetic activity and their slow and fast dynamics. Our work has been documented in [4-6].

Methods

Data was collected from nine healthy volunteers during a study of propofol-induced unconsciousness, collected under protocol approved by the Massachusetts General Hospital (MGH) Human Research Committee [7]. All subjects provided written informed consent. For all subjects, approximately 3 hours of data were recorded while using target-controlled infusion protocol. Further details of the data collection process are described in [7].

The infusion rate was increased and then decreased in a total of ten stages of roughly equal lengths (approximately 15 minutes each) to achieve target propofol concentrations of: 0 mg/kg/hr, 1, 2, 3, 4, 5, 3.75, 2.5, 1.25, 0. The two stages of 0 mg/kg/hr are hereby referred to as baselines before and after anesthesia respectively. Continuous electrocardiogram (ECG) and EDA were collected. LOC and ROC times were annotated based on lack of response to a button-pressing task. There were other interventions included in the study for patient safety, such as administering phenylephrine (a vasopressor). All data were analyzed using Matlab R2020b.

Multimodal autonomic indices were computed for heart rate variability (HRV) and electrodermal activity (EDA) using the point process methods developed and described in [6, 8-13].

Data Description

The Data folder contains the following files:

• EDA_temp_amp_#.csv: Matrix of 5 EDA covariates for each subject. In order, EDA tonic, mean pulse rate, standard deviation of pulse rate, mean pulse amplitude, standard deviation of pulse amplitude.
• LOC_ROC.csv: Times of loss of consciousness and recovery of consciousness in seconds for each subject as a table. LOC is the first column, ROC the second.
• S#_eda_tonic.csv: Tonic component of EDA
• S#_events: Timestamps of the beginning of the stages of the propofol experiment. There are 6 increasing stages and 4 decreasing stages.
• S#_LF.csv and S#_HF.csv: LF (low frequency) and HF (high frequency) heart rate variability. LF is thought to relate to both sympathetic and parasympathetic activity and HF to parasympathetic alone.
• S#_LFnu.csv and S#_HFnu.csv: Normalized forms of LF and HF respectively.
• S#_LOC.csv and S#_ROC.csv: Times of loss and recovery of consciousness respectively.
• S#_muRR.csv and S#_sigmaRR.csv: Mean and standard deviation of the RR interval respectively.
• S#_muHR.csv and S#_sigmaHR.csv: Mean and standard deviation of the heart rate respectively.
• S#_pow_tot.csv: Total power for HRV.
• S#_ratio.csv: LF/HF for HRV.
• S#_muPR.csv and S#_sigmaPR.csv: Mean and standard deviation of the EDA pulse rate respectively.
• S#_mu_amp.csv and S#_sigma_amp.csv: Mean and standard deviation of the EDA pulse amplitude respectively.
• S#_t_EDA_tonic: Timestamps to accompany tonic EDA.
• S#_t_EDA: Timestamps to accompany muPR, sigmaPR, mu_amp, and sigma_amp.
• S#_t_HRV: Timestamps to accompany muRR, sigmaRR, muHR, sigmaHR, pow_tot, ratio, LF, HF, LFnu, and HFnu

Code

Code has been provided to reproduce the analysis in our associated study [6]. To run a regression with the project data:

1. Open LogReg_propofol_regression_MAIN.m
2. Change the parameters in the first section to reflect the conditions you want to use (history or not, mean or median within each window, which question - see below, which covariates - multimodal or unimodal, window length in seconds, amount of history in seconds)
3. Run the script. The script is designed to input a 1x1 Matlab struct for each subject, with fields eda_tonic, events, HF, LF, HFnu (normalized HF), LFnu (normalized LF), LOC, ROC, mu_amp, sigma_amp, muHR, sigmaHR, muPR, sigmaPR, muRR, sigmaRR, pow_tot, ratio, t_HRV, t_EDA, t_EDA_tonic taken from the respective csv files.

To recreate the EDA specific figures for each subject in Subramanian et al. [6]:

1. Open EDA_deepdive_viz.m
2. Run the script. The script is designed to input a Matlab struct that is 1x9, each with field X that is from EDA_temp_amp_#.csv. It will generate figures for all 9 subjects in order from 1-9 showing the different EDA point process indices, temporal and amplitude related.

Usage Notes

These data and code have been used to explore relationships between behavioral and autonomic dynamics before during and after changes in consciousness in [4-6]. Those interested in experimenting with machine learning models using these indices or in exploring the dynamics of these indices with changing depth of sedation can do so. One limitation of these data is that they only contain HRV and EDA related information and do not contain EEG or blood pressure, for example. Those further interested in the point process indices themselves can check out [12-13] for the HRV point process model and [8-11] for development of the EDA point process model. Related datasets on PhysioNet [14-15] also have EDA data that can be used to experiment with the point process model properties.

Release Notes

Version 1.0

Conflicts of Interest

A utility patent application titled “Tracking Nociception under Anesthesia using a Multimodal Metric” was filed with S.S., R.B., and E.N.B. on July 15, 2020 (Application No. PCT/US2020/042031). Some of the methods used to create this database were included in the patent application.

References

1. Brown EN, Lydic R, Schiff ND. General anesthesia, sleep, and coma. N Engl J Med. 2010;363(27):2638-50.
2. Brown EN, Purdon PL, Van Dort CJ. General anesthesia and altered states of arousal: a systems neuroscience analysis. Annu Rev Neurosci. 2011;34: 601-628.
3. Kunimoto M et al. Neuroeffector characteristics of sweat glands in the human hand activated by regular neural stimuli. Journal of Physiology. 1991 Oct;442(1):391-411.
4. Subramanian S, Barbieri R, Purdon PL, Brown EN. Detecting loss and regain of consciousness during propofol anesthesia using multimodal indices of autonomic state. Proc. 42nd IEEE International Conf on Eng in Biol and Med (EMBC). 2020 Jul.
5. Subramanian S, Barbieri R, Purdon PL, Brown EN (2020). Analyzing transitions in anesthesia by multimodal characterization of autonomic state. Proc. 11th Conf. of the European Study Group on Cardiovascular Oscillations (ESGCO). 2020 Apr.
6. Subramanian S, Purdon PL, Barbieri R, Brown EN (2020). Quantitative Assessment of the Relationship between Behavioral and Autonomic Dynamics during Propofol-Induced Unconsciousness. bioRxiv doi: 10.1101/2020.11.03.367607
7. Purdon PL et al. Electroencephalogram Signatures of Loss and Recovery of Consciousness from Propofol. PNAS. 2013;110: E1142-1151.
8. Subramanian S, Barbieri R, Brown EN. A systematic method for preprocessing and analyzing electrodermal activity. Proc. 41st IEEE International Conf on Eng in Biol and Med (EMBC). 2019 Jul.
9. Subramanian S, Barbieri R, Brown EN. A point process characterization of electrodermal activity. Proc. 40th IEEE International Conf on Eng in Biol and Med (EMBC). 2018 Jul.
10. Subramanian S, Barbieri R, Brown EN. Point process temporal structure characterizes electrodermal activity. PNAS. 2020 Oct;117(42):26422-26428. DOI: 10.1073/pnas.2004403117.
11. Subramanian S, Purdon PL, Barbieri R, Brown EN. A model-based approach for pulse selection from electrodermal activity. IEEE Trans Biomed Eng. 2021 Apr 6. doi: 10.1109/TBME.2021.3071366.
12. Point Process Models of Human Heart Beat Interval Dynamics [Internet]. Available from: http://users.neurostat.mit.edu/barbieri/pphrv
13. Barbieri R, Matten EC, Alabi AA, Brown EN. A point-process model of human heartbeat intervals: new definitions of heart rate and heart rate variability. Am. J. Physiol. Heart Circ. Physiol. 2005 Jan;288(1):H424–435.
14. Subramanian, S., Barbieri, R., & Brown, E. (2020). Electrodermal Activity of Healthy Volunteers while Awake and at Rest (version 1.0). PhysioNet. https://doi.org/10.13026/arty-2540.
15. Subramanian, S., Purdon, P., Barbieri, R., & Brown, E. (2021). Pulse Amplitudes from electrodermal activity collected from healthy volunteer subjects at rest and under controlled sedation (version 1.0). PhysioNet. https://doi.org/10.13026/r9p1-bk90.