Career Center AES Connect

Abstracts

EEG-Derived Insights into Pharmacokinetics: Mapping Neural Dynamics to Drug Levels

Abstract number : 3.421
Submission category : 3. Neurophysiology / 3G. Computational Analysis & Modeling of EEG
Year : 2023
Submission ID : 1405
Source : www.aesnet.org
Presentation date : 12/4/2023 12:00:00 AM
Published date :

Authors :
Presenting Author: David Geng, BS – University of Pittsburgh School of Medicine, Carnegie Mellon University

Shobhit Singla, MD, PhD – Dept. of Neurology – University of Pittsburgh School of Medicine; James Castellano, MD, PhD – Dept. of Neurology – University of Pittsburgh School of Medicine; Avniel Ghuman, PhD – Dept. of Neurosurgery – University of Pittsburgh School of Medicine, Carnegie Mellon University

Rationale: Our understanding of the neural network effects of anti-seizure drugs (ASDs) primarily comes from studies comparing individuals on ASDs to those not on them. Thus, little is known about the continuous relationship between neural activity and pharmacodynamics, especially at the individual patient level. Understanding the dynamics of brain activity changes with ASD blood levels offers insights into both the mechanisms of action of AEDs and the neural network effects when patients transition on and off ASDs in epilepsy monitoring units (EMUs). Here, we assessed whether pharmacodynamics can be tracked using scalp electroencephalography (EEG) and identified key EEG markers of drug blood levels.

Methods: We retrospectively analyzed seven-day EEG recordings from a single patient admitted to the EMU for suspected epilepsy but was later diagnosed as non-epileptic by the clinical team. During the stay, the patient received four doses of oral levetiracetam (Keppra) without concurrent neurotropic or psychotropic medications. EEG Feature Extraction: After preprocessing and artifact removal from the EEG data, we segmented the continuous EEG data into 5-second non-overlapping epochs. For each epoch, we extracted 21 time and frequency domain features from each of the 28 electrodes (588 features). Pharmacokinetic Modeling: We built a two-compartment pharmacokinetic model to represent Keppra blood levels over time. This model combined Keppra blood concentration at admission, the specific dosage schedule, and established pharmacokinetic parameters of oral Keppra. Machine Learning Analysis: We utilized linear discriminant analysis to reduce the high-dimensional EEG data to a discriminant dimension. This enabled us to evaluate the potential of predicting Keppra blood levels using EEG features and identify the most significant EEG markers.

Results: Our analysis revealed a clear correspondence between brain-wide EEG-derived predictions and actual Keppra blood levels. Notably, we predicted with over 95% accuracy whether the patient had high or low levels of Keppra in their blood, and the EEG signals that allowed for this classification tracked the waxing and waning of the blood levels over the entire week (See Figure). Notably, aperiodic features, previously associated with EEG background activity and the excitatory-inhibitory (E/I) balance, were the top predictors in the classifier.

Conclusions: Our preliminary findings indicate that scalp EEG signals hold promise as pharmacokinetic predictors. The importance of aperiodic features in our findings further supports the hypothesis that ASDs modulate the E/I balance and EEG background activity. We are expanding our dataset to include more patients to determine the generalizability of these findings across both non-epileptic and epileptic conditions. Our findings underscore the significant potential of using scalp EEG to monitor neural dynamics for continuous and long-term pharmacokinetic applications, suggesting a promising direction for better understanding the mechanisms of action of ASDs in large scale brain networks.

Funding:
R01MH132225

Neurophysiology