Phenotypes of Epilepsy Etiology and Drug Resistance (PEER)

Principal Investigator: KENNEDY, EAMONN

Proposal Number: EP220089

Award Number: HT9425-23-1-0221

Period of Performance: 7/1/2023 - 6/30/2026

PUBLIC ABSTRACT

Post-traumatic epilepsy (PTE) refers to epilepsy that emerges and persists for more than 7 days after a traumatic brain injury (TBI). The likelihood of PTE depends on the severity of the brain injury, and there is normally a significant duration of time, often years, between a brain injury and the onset of recurring seizures. The delay between injury and epilepsy may offer a clinical window of opportunity since it represents a period of time prior to the emergence of epilepsy where we can affect change for those at higher risk for epilepsy following TBI. For example, if we compiled the right information together in the right ways, like the severity of brain injury, relevant medical history, and medications, it may be possible to infer who will develop epilepsy after brain injury. Very high accuracy is not required for this to be useful. For example, if we estimated the top one thousand Service Members we think are at risk of epilepsy after TBI, and only one hundred of them go on to develop epilepsy, then we will still have given some advanced warning to a hundred people that they are going to have a serious life-changing disease in the future. This means they can be allocated resources, targeted for interventions or trials before epilepsy emerges, and they can begin to prepare for different eventualities.

However, a simple model that looks at the medical history of Veterans and tries to guess who will be diagnosed with epilepsy in the future may not work for a few reasons. First, reporting rates of head injury are very low, and injury information from before military service or elsewhere may not be available. We propose to solve this problem directly by using the responses of thousands of Veteran volunteers with epilepsy and TBI who were asked about the dates and details of their epilepsy and history of lifetime head trauma in a study conducted by project co-investigator Dr. Mary Jo Pugh. We have agreements to reuse this data, which also documents measures of quality of life, medication use, combat exposure, and many other factors. Combined with decade-long medical histories, we can use this data offered by Veterans to greatly improve our model timing and event date accuracies.

Aside from event timing, there is another challenge. There is no guarantee that prior medical conditions and demographics contain enough information to make good future predictions. The strongest warning signs that come before an epilepsy diagnosis are often subjective, like unexplained fatigue, absent mindedness, spacing out, or changes in behavior. To capture these valuable pre-diagnostic warning signs, we propose to examine “multimodal” data. We will capture this information by proxy by analyzing big data repositories that detail how people use health services over time, dates, and dosages of medications that are administered, what health problems emerge, and what services are used and how frequently. A core feature of this study is that we believe the order of all these things combined might add a lot of predictive power. Whether a medication is prescribed in the presence or absence, or before or after certain medical diagnosis is informative.

Little work has explored whether specific health comorbidities interact with injury characteristics to increase risk for PTE, and military health questions of interest to the Department of Defense remain unanswered. We propose to enact the solutions to these hurdles we have described and specifically focus on medication differences and outcomes in our first aim. In our second aim, we will develop risk scores for PTE and drug-resistant epilepsy (DRE) following TBI. We want to try and exploit the delayed window of time between TBI and epilepsy diagnosis for detection and to buy time to prepare. The right preceding information could provide a summary measure of epilepsy risk following TBI, which could offer new opportunities for intervention and health care planning. The causal nature of epilepsy acquisition after TBI requires specific methods like network models that can map the directed links between TBI, epilepsy, comorbidities, treatments, and key like seizure freedom, and quality of life. Aim 2 will focus on ways to fine tune these data and models to maximize predictive performance.

After developing these risk scores, we will test how well they perform in new datasets in our third aim. We will trial the risk scores in two ongoing studies that are currently screening for epilepsy. If we can predict epilepsy before it happens in ongoing studies, that is strong evidence that the risk scores are working well and could even be rolled out to the whole population. It is not easy to guess how well these risk scores will perform up front, but there are encouraging correlates, and our group has the clinical and technical expertise to achieve success. Our proposal offers new perspectives on how to enhance treatment and innovations in data collection and analysis. We will use multimodal data to build and deploy new tools that seek to flag the early warnings for epilepsy risk with prospective validation building on our extensive clinical and technical expertise.

TECHNICAL ABSTRACT

Background: Recent research suggests post-traumatic epilepsy (PTE) is associated with the failure of antiseizure medications (ASMs) to provide seizure freedom, but the underlying mechanisms, cofactors, and the broader extent of treatment-PTE relationships remain unclear. Whether specific health comorbidities interact with traumatic brain injury (TBI) to increase risk for PTE has not been extensively evaluated, and military health questions of interest to the Department of Defense (DOD) remain unanswered. This proposal addresses these research gaps by developing risk scores for PTE and drug-resistant epilepsy (DRE) following TBI. The right preceding information could provide a summary measure of epilepsy risk following TBI, which could offer new opportunities for intervention and health care planning. The causal nature of epilepsy acquisition after TBI requires specific methods that can map the directed links between TBI, epilepsy, comorbidities, treatments, and key outcomes including suicide ideation/attempt, mortality, seizure freedom, and quality of life (QOL).

Hypothesis: Our overarching hypothesis is that the magnitude and mechanisms of PTE, PTE comorbidities, and treatment act in concert within a dynamic network where timing and directionality are critical. This proposal aims to use both existing and ongoing Epilepsy Research Program (ERP) study data on self-reported lifetime history of TBI to enhance TBI timing/severity estimates in directed multimodal networks. Our analysis of this ERP study data will use new software tools developed by Drs. Kennedy and Vega Yon to complete our specific aims.

Specific Aims: Aim 1 considers an epilepsy cohort and focuses on treatment, comorbidity, and outcomes. Aim 1 will use existing survey/interview data and merged heath records to develop directed health networks of epilepsy etiology, comorbidity, drug administration, and QOL. Aim 2 considers a TBI cohort and focuses on risk factors for epilepsy including comorbidities and drug administration, which may correlate with subjective risk factors. Aim 2 will develop risk scores for PTE and DRE following TBI by identifying what multimodal signals discriminate Veterans who do/do not acquire an epilepsy after TBI. Aim 3 is a detection/risk-scoring objective that will incorporate two ongoing studies (one large, one rich) chosen for prospective validation of risk scores for epilepsy and treatment failure. Aim 3 will validate risk score performance in two ongoing studies by measuring how well risk scores detect future epilepsy diagnoses and better/worse outcomes including seizure freedom.

Aim 1: Develop models of comorbidity and medication use in a VWE (Veterans with epilepsy) cohort with PTE/DRE and determine the drivers of medication success/failure and QOL in PTE and NTE (non-traumatic epilepsy) cohorts.

Aim 2: Develop risk scores in a TBI+ cohort that differentiate Veterans who do/do not acquire epilepsy following TBI using multimodal data documenting injury characteristics, health conditions, and treatment after TBI.

Aim 3: will prospectively validate PTE and DRE risk scores in two ongoing studies that screen for epilepsy.

Study Design: This combined retrospective and prospective study design will use multimodal data to establish directed network models and evidence-based validated risk scores for PTE and DRE. We will develop networks combining the medication treatment and comorbid health conditions for three cohorts: (1) a cohort that collected rich QOL, combat, and medication use measures, and merged pharmaceutical and health records (n=2,603); (2) a large existing Veteran cohort (VWE n=7,323) documenting longitudinal medication and health conditions and outcomes; and (3) the LIMBIC (Long-Term Impact of Military-Relevant Brain Injury Consortium) Prospective Longitudinal Study (PLS) study cohort (N>2200, growing), which is screening for TBI, epilepsy, and relevant outcomes. We will implement a TBI event timing model based on self-reported lifetime history of TBI to address uncertainties in TBI timing and improve our temporal models. We will infer probabilities of comorbid health conditions before and after TBI and epilepsy. We will relate these temporal exposure histories to outcomes including QOL and seizure freedom. We will implement a network analysis to relate multimorbidity and medication use and create dynamic and appropriate visualizations and media. We will identify model parameters that are most strongly associated with VWE with PTE vs. NTEs. We will deploy pretrained models of score risks for PTE and DRE in the prospective cohorts to measure and validate epilepsy/DRE risk score performance.

Innovation: We propose to develop novel epilepsy diagnosis and treatment risk scores that will enhance military health and promote new research paradigms. Our methods offer new perspectives on how to enhance treatment and outcomes for the VWE community combining innovations in data collection and analysis. Using multimodal data, this research project will pioneer etiologically relevant tools that seek to flag the early warnings for epilepsy risk with prospective validation leveraging our detection/clinical expertise and novel solutions to improve military health.