Devin Ma

Epilepsy care is often guided by incomplete snapshots: patient-reported seizures, intermittent drug levels and medication histories that may not reflect current exposure. Between these snapshots, clinicians have little objective information about how the brain is responding to antiseizure medication. In this study, we tested whether intracranial EEG synchrony, a measure of co-ordinated neural activity across intracranial electrodes, tracks antiseizure medication load in patients with drug-resistant epilepsy. We retrospectively studied 80 consecutive patients with drug-resistant epilepsy who underwent presurgical intracranial EEG monitoring at the Hospital of the University of Pennsylvania. Antiseizure medication load was estimated continuously from medication administration records using a validated pharmacokinetic model. Synchrony was computed from intracranial EEG using the Kuramoto order parameter. We first compared synchrony within the same patient during high and low medication exposure. We then tested the continuous association between synchrony and medication load across epilepsy monitoring unit admission, adjusting for time since admission, seizure timing, wakefulness, time of day and interictal spike rate. Analyses accounted for repeated measurements within patients and tested robustness to temporal structure in the recordings. We also assessed whether synchrony dynamics were preserved when fewer electrodes were sampled. Forty-five patients had sufficient high- and low-exposure data after peri-ictal exclusion. Synchrony was higher during low medication exposure in 32 of 45 patients, corresponding to a median 7.4% increase and a median paired difference of 0.0082 on the 0–1 synchrony scale (sign test P = 0.007; Wilcoxon P < 0.001). In the continuous analysis of 67 patients and 23 402 repeated 10-min observations, higher synchrony was associated with lower medication load (β = −0.153, 95% confidence interval −0.293 to −0.012, P = 0.033); interictal spike rate was not independently associated with medication load. The inverse association remained evident in circular-shift, temporally binned and likelihood-based sensitivity analyses. Relative synchrony dynamics were preserved under electrode subsampling, with a median correlation of 0.85 with full-array synchrony at 40% sampling. These findings support synchrony as a candidate continuous physiological readout of antiseizure medication load in patients with drug-resistant epilepsy. By linking medication exposure to a measurable brain-network state, synchrony may help move epilepsy monitoring beyond seizure counting alone and towards continuous assessment of how the brain is responding to treatment.

Background and ObjectivesInpatient video-EEG epilepsy monitoring aims to provide a safe environment for antiseizure medication (ASM) tapering to precipitate seizures, which is often required for di...

With new, minimally invasive continuous EEG and wearable monitoring systems for epilepsy nearing regulatory approval, there is a rush to determine how to make these devices most useful to patients. One application is to track biomarkers of antiseizure medication (ASM) levels to help manage seizure risk and side effects. In this paper, we validate one such biomarker, phase synchronization, during medication taper and presurgical evaluation in a cohort of 80 consecutive patients recorded with intracranial EEG (iEEG) at the University of Pennsylvania. While previous investigators have demonstrated that synchrony is negatively correlated with ASM load at the resolution of 1 day, we hypothesize that synchrony continuously tracks and can predict ASM load on a clinically useful time scale. We test this hypothesis using a pharmacokinetic model generating continuous ASM load values previously published by our group and correlate it with continuous synchrony values derived from the iEEG. We use a linear mixed effect model to examine the relationship between ASM load, synchrony, vigilance, and other time dependent factors. We use dominance analysis to rank the relative importance of predictors based on their influence on ASM load. We find that synchrony not only can predict ASM load but is also the most significant predictor. Our study highlights the potential of synchrony as a biomarker for ASM load, and its utility in an ambulatory implantable device that can alert patients to conditions lowering their medications (e.g., adherence, drug interactions, generic change, pregnancy, etc.) or potential medication toxicity. We propose that measures like synchrony, packaged in an appropriate interface, could bring substantial value to ambulatory epilepsy management devices.

Objective. New implantable and wearable devices hold great promise to help patients manage their seizure disorders. One proposed application is measuring the rate of interictal epileptiform discharges as a biomarker of medication levels and seizure risk. This study aims to determine whether interictal epileptiform spike rates (spikes) are independently associated with anti-seizure medication (ASM) levels and evaluate whether spike rates are a reliable biomarker for ASM levels. Approach. We conducted a retrospective analysis of 69 patients with drug resistant epilepsy undergoing intracranial EEG monitoring during ASM taper in the epilepsy monitoring unit. An automated spike detection algorithm, validated seizure annotations, and a model of ASM load were used to assess the relationship between spike rates, ASM level, state of consciousness, and seizure timing using linear mixed effects models. Event related analysis and seasonal autoregressive models were applied to evaluate both short-term and longer-term effects of ASMs on spike rates, respectively. Main Results. Spike rates were found to increase following seizures, during sleep, and in patients experiencing early seizures, with no significant association between ASM load and spike rates after controlling for these confounders. Initial models without controls showed a positive association between ASM load and spike rates. However, this relationship disappeared when the effects of sleep and seizures were accounted for. Medication-specific analysis revealed that only levetiracetam showed a significant impact on spike rates during taper. Significance. This study demonstrates that interictal spike rates are more strongly influenced by seizure activity and sleep than by ASM levels, suggesting limited utility of spike rates as a biomarker for ASM load. However, spikes may still serve as important markers for seizure control and disease severity.

Objective: Mesial temporal lobe epilepsy is a common localization of drug‐resistant epilepsy in adults. Patients often undergo intracranial electroencephalographic monitoring to confirm localization and determine candidacy for focal ablation or resection. Clinicians primarily base surgical decision‐making on seizure onset patterns, with imaging abnormalities and information from interictal epileptiform discharge (spikes) used as ancillary data. How the morphology and timing of spikes within multielectrode sequences may inform surgical planning is unknown, in part due to the lack of measurement methods for large datasets. We hypothesized that patients with mesial temporal lobe epilepsy have a distinct mesial‐to‐lateral spike gradient that differentiates them from other epilepsy localizations.Methods: In a multicenter study at the University of Pennsylvania and the Medical University of South Carolina, we analyzed the timing and morphology of spikes and seizure high‐frequency energy ratio in 75 patients with drug‐resistant epilepsy. We compared these features across patients with mesial temporal lobe epilepsy, temporal neocortical epilepsy, and other localizations.Results: A logistic regression model combining all features predicted a clinical localization of mesial temporal lobe epilepsy in unseen patients with an area under the receiver operating characteristic curve of .82 (compared to an area under the receiver operating characteristic curve of .70 for seizure‐only features, DeLong test p = .08) and an average precision of .84. Spike rate was the most important feature in the combined model.Significance: These findings advance surgical planning by demonstrating that quantitative spike analysis can effectively supplement seizure data in localizing mesial temporal lobe epilepsy. This approach could reduce reliance on prolonged seizure monitoring, potentially decreasing patient risk and hospitalization costs while improving surgical targeting. Our results support incorporating automated spike analysis into standard clinical workflows for epilepsy surgery evaluation.

Mesial temporal lobe epilepsy (mTLE) is a common localization of drug-resistant epilepsy in adults. Patients often undergo intracranial EEG (iEEG) monitoring to confirm localization and determine candidacy for focal ablation or resection. Clinicians primarily base surgical decision-making on seizure onset patterns, with imaging abnormalities and information from interictal epileptiform discharge (spikes) used as ancillary data. How the morphology and timing of spikes within multi-electrode sequences may inform surgical planning is unknown, in part due to the lack of measurement methods for large datasets. We hypothesized that patients with mTLE have a distinct mesial-to-lateral spike pattern that differentiates them from other epilepsy localizations. In a multicenter study at the University of Pennsylvania and the Medical University of South Carolina, we analyzed the timing and morphology of spikes and seizure high frequency energy ratio (HFER) in 75 patients with drug-resistant epilepsy. We compared these features across patients with mTLE, temporal neocortical epilepsy, and other localizations. A logistic regression model combining all features predicted a clinical localization of mTLE in unseen patients with an AUC of 0.82 (compared to an AUC of 0.70 for seizure-only features, DeLong’s test p = 0.08). Spike rate was the most important feature in the combined model. Modeled probability of mTLE was similar between patients who had a good versus a poor 12-month outcome after undergoing a mesial temporal resection or ablation ( p = 0.34). These findings support quantitative spike analysis to supplement analysis of seizures for use in surgical planning.

While our understanding of fairness in machine learning has significantly progressed, our understanding of fairness in reinforcement learning (RL) remains nascent. Most of the attention has been on fairness in one-shot classification tasks; however, real-world, RL-enabled systems (e.g., autonomous vehicles) are much more complicated in that agents operate in dynamic environments over a long period of time. To ensure the responsible development and deployment of these systems, we must better understand fairness in RL. In this paper, we survey the literature to provide the most up-to-date snapshot of the frontiers of fairness in RL. We start by reviewing where fairness considerations can arise in RL, then discuss the various definitions of fairness in RL that have been put forth thus far. We continue to highlight the methodologies researchers used to implement fairness in single- and multi-agent RL systems and showcase the distinct application domains that fair RL has been investigated in. Finally, we critically examine gaps in the literature, such as understanding fairness in the context of RLHF, that still need to be addressed in future work to truly operationalize fair RL in real-world systems.

While our understanding of fairness in machine learning has significantly progressed, our understanding of fairness in reinforcement learning (RL) remains nascent. Most of the attention has been on fairness in one-shot classification tasks; however, real-world, RL-enabled systems (e.g., autonomous vehicles) are much more complicated in that agents operate in dynamic environments over a long period of time. To ensure the responsible development and deployment of these systems, we must better understand fairness in RL. In this paper, we survey the literature to provide the most up-to-date snapshot of the frontiers of fairness in RL. We start by reviewing where fairness considerations can arise in RL, then discuss the various definitions of fairness in RL that have been put forth thus far. We continue to highlight the methodologies researchers used to implement fairness in single- and multi-agent RL systems before showcasing the distinct application domains that fair RL has been investigated in. Finally, we critically examine gaps in the literature, such as understanding fairness in the context of RLHF, that still need to be addressed in future work to truly operationalize fair RL in real-world systems.