Long-term epilepsy outcome dynamics revealed by natural language processing of clinic notes.

OBJECTIVE: Electronic medical records allow for retrospective clinical research with large patient cohorts. However, epilepsy outcomes are often contained in free text notes that are difficult to mine. We recently developed and validated novel natural language processing (NLP) algorithms to automatically extract key epilepsy outcome measures from clinic notes. In this study, we assessed the feasibility of extracting these measures to study the natural history of epilepsy at our center. METHODS: We applied our previously validated NLP algorithms to extract seizure freedom, seizure frequency, and date of most recent seizure from outpatient visits at our epilepsy center from 2010 to 2022. We examined the dynamics of seizure outcomes over time using Markov model-based probability and Kaplan-Meier analyses. RESULTS: Performance of our algorithms on classifying seizure freedom was comparable to that of human reviewers (algorithm F1 = .88 vs. human annotator κ = .86). We extracted seizure outcome data from 55 630 clinic notes from 9510 unique patients written by 53 unique authors. Of these, 30% were classified as seizure-free since the last visit, 48% of non-seizure-free visits contained a quantifiable seizure frequency, and 47% of all visits contained the date of most recent seizure occurrence. Among patients with at least five visits, the probabilities of seizure freedom at the next visit ranged from 12% to 80% in patients having seizures or seizure-free at the prior three visits, respectively. Only 25% of patients who were seizure-free for 6 months remained seizure-free after 10 years. SIGNIFICANCE: Our findings demonstrate that epilepsy outcome measures can be extracted accurately from unstructured clinical note text using NLP. At our tertiary center, the disease course often followed a remitting and relapsing pattern. This method represents a powerful new tool for clinical research with many potential uses and extensions to other clinical questions.

Characterizing the treatment gap in the United States among adult patients with a new diagnosis of epilepsy.

OBJECTIVE: Epilepsy is largely a treatable condition with antiseizure medication (ASM). Recent national administrative claims data suggest one third of newly diagnosed adult epilepsy patients remain untreated 3 years after diagnosis. We aimed to quantify and characterize this treatment gap within a large US academic health system leveraging the electronic health record for enriched clinical detail. METHODS: This retrospective cohort study evaluated the proportion of adult patients in the health system from 2012 to 2020 who remained untreated 3 years after initial epilepsy diagnosis. To identify incident epilepsy, we applied validated administrative health data criteria of two encounters for epilepsy/seizures and/or convulsions, and we required no ASM prescription preceding the first encounter. Engagement with the health system at least 2 years before and at least 3 years after diagnosis was required. Among subjects who met administrative data diagnosis criteria, we manually reviewed medical records for a subset of 240 subjects to verify epilepsy diagnosis, confirm treatment status, and elucidate reason for nontreatment. These results were applied to estimate the proportion of the full cohort with untreated epilepsy. RESULTS: Of 831 patients who were automatically classified as having incident epilepsy by inclusion criteria, 80 (10%) remained untreated 3 years after incident epilepsy diagnosis. Manual chart review of incident epilepsy classification revealed only 33% (78/240) had true incident epilepsy. We found untreated patients were more frequently misclassified (p < .001). Using corrected counts, we extrapolated to the full cohort (831) and estimated <1%-3% had true untreated epilepsy. SIGNIFICANCE: We found a substantially lower proportion of patients with newly diagnosed epilepsy remained untreated compared to previous estimates from administrative data analysis. Manual chart review revealed patients were frequently misclassified as having incident epilepsy, particularly patients who were not treated with an ASM. Administrative data analyses utilizing only diagnosis codes may misclassify patients as having incident epilepsy.

Predictors of Audio-Only Versus Video Telehealth Visits During the COVID-19 Pandemic.

BACKGROUND: Most health insurance organizations reimbursed both video and audio-only (i.e., phone) visits during the COVID-19 pandemic, but may discontinue phone visit coverage after the pandemic. The impact of discontinuing phone visit coverage on various patient subgroups is uncertain. OBJECTIVE: Identify patient subgroups that are more probable to access telehealth through phone versus video. DESIGN: Retrospective cohort. PATIENTS: All patients at a U.S. academic medical center who had an outpatient visit that was eligible for telehealth from April through June 2020. MAIN MEASURES: The marginal and cumulative effect of patient demographic, socioeconomic, and geographic characteristics on the probability of using video versus phone visits. KEY RESULTS: A total of 104,204 patients had at least one telehealth visit and 45.4% received care through phone visits only. Patient characteristics associated with lower probability of using video visits included age (average marginal effect [AME] -6.9% for every 10 years of age increase, 95%CI -7.8, -4.5), African-American (AME -10.2%, 95%CI -11.4, -7.6), need an interpreter (AME -19.3%, 95%CI -21.8, -14.4), Medicaid as primary insurance (AME -12.1%, 95%CI -13.7, -9.0), and live in a zip code with low broadband access (AME -7.2%, 95%CI -8.1, -4.8). Most patients had more than one factor which further reduced their probability of using video visits. CONCLUSIONS: Patients who are older, are African-American, require an interpreter, use Medicaid, and live in areas with low broadband access are less likely to use video visits as compared to phone. Post-pandemic policies that eliminate insurance coverage for phone visits may decrease telehealth access for patients who have one or more of these characteristics.

Polypharmacy composition and patient- and provider-related variation in patients with epilepsy.

OBJECTIVE: To describe polypharmacy composition, and the degree to which patients versus providers contribute to variation in medication fills, in people with epilepsy. METHODS: We performed a retrospective study of Medicare beneficiaries with epilepsy (antiseizure medication plus diagnostic codes) in 2014 (N = 78,048). We described total number of medications and prescribers, and specific medications. Multilevel models evaluated the percentage of variation in two outcomes (1. number of medications per patient-provider dyad, and 2. whether a medication was filled within thirty days of a visit) due to patient-to-patient differences versus provider-to-provider differences. RESULTS: Patients filled a median of 12 (interquartile range [IQR] 8-17) medications, from median of 5 (IQR 3-7) prescribers. Twenty-two percent filled an opioid, and 61% filled at least three central nervous system medications. Levetiracetam was the most common medication (40%), followed by hydrocodone/acetaminophen (27%). The strongest predictor of medications per patient was Charlson comorbidity index (7.5 [95% confidence interval (CI) 7.2-7.8] additional medications for index 8+ versus 0). Provider-to-provider variation explained 36% of variation in number of medications per patient, whereas patient-to-patient variation explained only 2% of variation. Provider-to-provider variation explained 57% of variation in whether a patient filled a medication within 30 days of a visit, whereas patient-to-patient variation explained only 30% of variation. CONCLUSION: Patients with epilepsy fill a large number of medications from a large number of providers, including high-risk medications. Variation in medication fills was substantially more related to provider-to-provider rather than patient-to-patient variation. The better understanding of drivers of high-prescribing practices may reduce avoidable medication-related harms.

Inpatient long-term video-electroencephalographic monitoring event capture audiovisual diagnostic quality.

OBJECTIVES: Long-term video-electroencephalographic monitoring (LTVEM) represents the gold-standard method to evaluate whether events represent electrographic seizures, but limited work has evaluated the quality of inpatient event capture. We evaluated the frequency of audiovisual factors impairing the ideal electroclinical correlation of seizure-like episodes during LTVEM. METHODS: We retrospectively reviewed consecutive inpatient LTVEM studies (11/2019-12/2019) from three academic epilepsy centers. We evaluated all pushbutton events for audiovisual characteristics such as whether the event was narrated, whether the patient was blocked on camera, and what diagnostic challenges impaired the electroencephalographer's ability to understand either the reason the event button was pushed or clinical semiology ("electroclinical correlation"). We determined the percent of events and studies with each outcome. RESULTS: There were 154 studies with 520 pushbutton events. The pushbutton was most commonly activated by patients (41%), followed by nurses (31%) or family (17%). Twenty-nine percent of events represented electrographic seizures, and 78% occurred in the Epilepsy Monitoring Unit. The reason for the push was not stated in 45% of events, and inadequate narration impaired electroclinical correlation in 19% of events. At least one relevant part of the patient's body was blocked during 12% of events, but this impaired electroclinical correlation in only 1% of events. There was at least one factor impairing electroclinical correlation in 21% of events, most commonly due to incomplete narration (N = 99), lights off (N = 15), or blankets covering the patient (N = 15). At least one factor impaired electroclinical correlation for any event in 36% of studies. CONCLUSION: Audiovisual factors impairing the electroencephalographer's ability to render an electroclinical correlation were common, particularly related to inadequate narration from bedside observers to explain the reason for pushing the button or event semiology. Future efforts to develop targeted countermeasures should address narration challenges and improve inpatient seizure monitoring quality metrics.

Cardiovascular disease risk, awareness, and treatment in people with epilepsy.

OBJECTIVE: To evaluate whether cardiovascular risk, risk awareness, and guideline concordant treatment differ in individuals with versus without epilepsy. METHODS: This was a retrospective cross-sectional study using the National Health and Nutrition Examination Survey. We included participants ≥18 years for 2013-2018. We classified participants as having epilepsy if reporting ≥1 medication treating seizures. We calculated 10-year atherosclerotic cardiovascular disease (ASCVD) risk using the revised pooled cohort equation. We compared unadjusted and adjusted risk for participants with versus without epilepsy. We then assessed hypertension and diabetes disease awareness and control, plus statin guideline-concordance. We assessed mediators for both ASCVD risk and cardiovascular disease awareness. RESULTS: Of 17,961 participants, 154 (0.9%) had epilepsy. Participants with epilepsy reported poorer diet (p = 0.03), fewer minutes of moderate-vigorous activity per day (p < 0.01), and increased frequency of cardiovascular conditions (e.g. coronary heart disease, myocardial infarction, stroke). There was no difference in control of individual examination and laboratory risk factors between groups (A1c, systolic blood pressure, diastolic blood pressure, high-density lipoprotein, low-density lipoprotein, total cholesterol). However, epilepsy was associated with 52% (95% confidence interval [CI]: 0-130%) increase in ASCVD risk, which became nonsignificant after adjusting for health behaviors. No single studied variable (income, Patient Health Questionnaire-9 (PHQ-9), diet, smoking) had a significant indirect effect. Participants with epilepsy reported increased hypertension awareness which was trivially but significantly mediated by having a routine place of healthcare (indirect effect: 1% absolute increase (95% CI: 0-1%), and they reported increased rates of hypertension treatment and guideline-concordant statin therapy. Participants with versus without epilepsy reported similar rates of blood pressure control and diabetes awareness, treatment, and control. CONCLUSIONS: Participants with epilepsy had increased ASCVD risk, despite similar or better awareness, treatment, and control of individual risk factors such as diabetes and hypertension. Our results suggest that epilepsy is associated with numerous health behaviors leading to cardiovascular disease, though the causal pathway is complex as these variables (income, depression, diet, exercise, smoking) generally served as confounders rather than mediators.

Disability in people with epilepsy: A nationally representative cross-sectional study.

OBJECTIVE: The objective of this study was to explore the prevalence and predictors of limitations causing disability in patients treated for seizures or epilepsy compared with patients without epilepsy. METHODS: This was a retrospective cross-sectional study using the National Health and Nutrition Examination Survey (NHANES). We included all participants ≥20 years old for 2013-2018. We classified patients as having epilepsy if they reported taking at least one prescription medication to treat seizures or epilepsy. Physical, mental, and social limitations were determined from interview questions. We report the prevalence of any limitation and total number of limitations for participants without vs. with epilepsy using serial negative binomial regressions and severity of individual limitations according to epilepsy status. RESULTS: We included 17,057 participants, of whom 148 (0.8%) had epilepsy. Overall, 80% (95% confidence interval [CI]: 73%-86%) with epilepsy vs. 38% (95% CI: 36%-39%) without epilepsy reported at least 1 limitation (p < 0.01). The mean number of limitations was 7.5 (95% CI: 6.2-8.8) for those with epilepsy vs. 2.4 (95% CI: 2.3-2.6) for those without epilepsy (p < 0.01). Epilepsy was associated with an incidence rate ratio (IRR) of 3.1 (95% CI: 2.6-3.7) in an unadjusted negative binomial regression. After adjusting for demographics and comorbidities, this association was no longer significant (IRR: 1.2, 95% CI: 0.9-1.7). Limitations cited by 40-50% of participants with epilepsy included stooping/kneeling/crouching, standing for long periods of time, and pushing/pulling objects. Limitation severity was consistently higher in patients with epilepsy. CONCLUSIONS: Patients with epilepsy had 3.1 times as many physical, mental, or social limitations compared with those without epilepsy, and disability severity was consistently higher. This effect was attenuated after considering baseline variables such as smoking and depression severity. Our work implies the importance of structured mental health screening and self-management programs targeting mood, weight, and lifestyle as potential leverage points towards alleviating epilepsy-related disability.

Polypharmacy in patients with epilepsy: A nationally representative cross-sectional study.

OBJECTIVE: The objective of the study was to characterize the prevalence of polypharmacy and central nervous system (CNS)-acting medications in patients with epilepsy, and particular types of medications. METHODS: This was a retrospective cross-sectional study using data from the nationally representative National Health and Nutrition Examination Survey (NHANES). We included patients who reported taking at least one prescription medication in order to treat seizures or epilepsy during NHANES survey years 2013-2016. We assessed the number and types of drugs and predictors of total number of medications using a negative binomial regression. We then assessed prevalence of polypharmacy (≥5 medications), CNS polypharmacy (≥3 CNS-acting medications) and additional CNS-acting medications, and drugs that lower the seizure threshold (i.e., bupropion and tramadol), and extrapolated prevalence to estimated affected US population. RESULTS: The NHANES contained 20,146 participants, of whom 135 reported taking ≥1 antiseizure medication (ASM) for seizures or epilepsy representing 2,399,520 US citizens using NHANES's sampling frame. Patients reported taking a mean 5.3 (95% confidence interval (CI): 4.3-6.3) prescription medications. Adjusting for race, sex, and uninsurance, both age and number of chronic conditions predicted increased number of medications (incident rate ratio (IRR) per decade: 1.16, 95% CI: 1.04-1.28; IRR per chronic condition: 1.19, 95% CI: 1.11-1.27). Polypharmacy was reported by 47% (95% CI: 38%-57%) of patients, CNS polypharmacy by 34% (23%-47%), benzodiazepine use by 21% (14%-30%), opioid use by 16% (11%-24%), benzodiazepine plus opioid use by 6% (3%-14%), and 6% (2%-15%) reported a drug that lowers the seizure threshold. Twelve percent (7%-20%) took an opioid with either a benzodiazepine or gabapentinoid. CONCLUSIONS: Polypharmacy is common in patients with epilepsy. Patients taking ASMs frequently reported also taking other CNS-acting medications (i.e., opioids, benzodiazepines, seizure threshold-lowering medications), and medication combinations with black box warnings. Central nervous system polypharmacy poses health risks. Future research is needed to explore drivers of polypharmacy and strategies to help mitigate potentially harmful prescription use in this high-risk population.

Big data in status epilepticus.

Status epilepticus care and treatment are already being touched by the revolution in data science. New approaches designed to leverage the tremendous potential of "big data" in the clinical sphere are enabling researchers and clinicians to extract information from sources such as administrative claims data, the electronic medical health record, and continuous physiologic monitoring data streams. Algorithmic methods of data extraction also offer potential to fuse multimodal data (including text-based documentation, imaging data, and time-series data) to improve patient assessment and stratification beyond the manual capabilities of individual physicians. Still, the potential of data science to impact the diagnosis, treatment, and minute-to-minute care of patients with status epilepticus is only starting to be appreciated. In this brief review, we discuss how data science is impacting the field and draw examples from the following three main areas: (1) analysis of insurance claims from large administrative datasets to evaluate the impact of continuous electroencephalogram (EEG) monitoring on clinical outcomes; (2) natural language processing of the electronic health record to find, classify, and stratify patients for prognostication and treatment; and (3) real-time systems for data analysis, data reduction, and multimodal data fusion to guide therapy in real time. While early, it is our hope that these examples will stimulate investigators to leverage data science, computer science, and engineering methods to improve the care and outcome of patients with status epilepticus and other neurological disorders. This article is part of the Special Issue "Proceedings of the 7th London-Innsbruck Colloquium on Status Epilepticus and Acute Seizures".

Timing is everything: Where status epilepticus treatment fails.

Status epilepticus is an emergency; however, prompt treatment of patients with status epilepticus is challenging. Clinical trials, such as the ESETT (Established Status Epilepticus Treatment Trial), compare effectiveness of antiepileptic medications, and rigorous examination of effectiveness of care delivery is similarly warranted. We reviewed the medical literature on observed deviations from guidelines, clinical significance, and initiatives to improve timely treatment. We found pervasive, substantial gaps between recommended and "real-world" practice with regard to timing, dosing, and sequence of antiepileptic therapy. Applying quality improvement methodology at the institutional level can increase adherence to guidelines and may improve patient outcomes. Ann Neurol 2017;82:155-165.

Addressing barriers to surgical evaluation for patients with epilepsy.

OBJECTIVE: Patients with poorly controlled seizures are at elevated risk of epilepsy-related morbidity and mortality. For patients with drug-resistant epilepsy that is focal at onset, epilepsy surgery is the most effective treatment available and offers a 50-80% cure rate. Yet, it is estimated that only 1% of patients with drug-resistant epilepsy undergo surgery in a timely fashion, and delays to surgery completion are considerable. The aim of this study was to increase availability and decrease delay of surgical evaluation at our epilepsy center for patients with drug-resistant epilepsy by removing process barriers. METHODS: For this quality improvement (QI) initiative, we convened a multidisciplinary team to construct a presurgical pathway process map and complete root cause analysis. This inquiry revealed that the current condition allowed patients to proceed through the pathway without centralized oversight. Therefore, we appointed an epilepsy surgery nurse manager, and under her direction, multiple additional process improvement interventions were applied. We then retrospectively compared preintervention (2014-2015) and postintervention (2016-2017) cohorts of patient undergoing the presurgical pathway. The improvement measures were patient throughput and pathway sojourn times. As a balancing measure, we considered the proportion of potentially eligible patients (epilepsy monitoring unit (EMU) admissions) who ultimately completed epilepsy surgery. RESULTS: Following our intervention, patient throughput was substantially increased for each stage of the presurgical pathway (32%-96% growth). However, patient sojourn times were not improved overall. No difference was observed in the proportion of possible candidates who ultimately completed epilepsy surgery. SIGNIFICANCE: Although process improvement expanded the number of patients who underwent epilepsy surgical evaluation, we experienced concurrent prolongation of the time from pathway initiation to completion. Ongoing improvement cycles will focus on newly identified residual sources of bottleneck and delay.

Postdiagnosis neurological care for patients with psychogenic nonepileptic spells (PNES).

OBJECTIVE: This study investigated continuity of neurological care for patients discharged from the epilepsy monitoring unit (EMU) with a diagnosis of psychogenic nonepileptic spells (PNES). Because PNES are seizure-like episodes that cannot be explained by abnormal electrical brain activity, they are challenging for patients to understand and accept. Consequently, after diagnosis, patients commonly fail to start recommended psychotherapy and instead pursue redundant medical care. As consistent relationships with healthcare providers may help, we instituted standard follow-up for patients diagnosed with PNES. METHODS: We performed a retrospective observational cohort study of adults diagnosed with PNES in our EMU. In November 2013, we began routine scheduling of postdischarge follow-up neurology appointments. We compared preintervention (November 2010-October 2013) and postintervention (November 2013-May 2016) cohorts with regard to clinic attendance, understanding the diagnosis, compliance with recommendations, and event frequency. RESULTS: We identified 55 patients in the preintervention cohort and 123 patients in the postintervention cohort. We successfully implemented the intended practice changes; more patients had follow-up scheduled by discharge (preintervention 2% vs. postintervention 36%, p<0.001), time to follow-up decreased (46days vs. 29, p=0.001), and providers more consistently queried understanding of diagnosis (38% vs. 67%, p=0.03). Explicit planning for continued care did not produce the anticipated patient-provider relationships, as follow-up in clinic was low (38% vs. 37%). For patients who attended clinic, the intervention did not improve establishment of psychiatric care, compliance with medication recommendations, understanding of diagnosis, or event frequency. The odds of reduced event frequency were nonsignificantly increased with understanding the diagnosis (OR 3.75, p=0.14). Recommending antiepileptic drug (AED) discontinuation was associated with increased odds of event freedom (OR 6.91, p<0.01). SIGNIFICANCE: Scheduling follow-up for patients diagnosed with PNES did not facilitate ongoing patient-provider relationships due to poor clinic attendance. As follow-up is unreliable, the inpatient visit is a critical window of opportunity for intervention.

Use of Antiepileptic Drugs in Patients with Chronic Kidney Disease and End Stage Renal Disease.

Epilepsy is a disorder with an approximate worldwide prevalence of 1%. Due to complexities of metabolism, protein-binding, renal elimination, and other pharmacokinetic parameters, the dosing of antiepileptic drugs (AEDs) in patients with chronic kidney disease (CKD) or end stage renal disease (ESRD) deserves special attention. This is a review of the most commonly prescribed AEDs with special focus on their indication, pharmacokinetics, and unique considerations for use in patients with CKD and ESRD. A review of their renal toxicities is also included.

Burden of Neurologic Health Care and Incident Neurologic Diagnoses in the Year After COVID-19 or Influenza Hospitalization.

BACKGROUND AND OBJECTIVE: Following the outbreak of viral infections from the severe acute respiratory syndrome coronavirus 2 virus in 2019 (coronavirus disease 2019 [COVID-19]), reports emerged of long-term neurologic sequelae in survivors. To better understand the burden of neurologic health care and incident neurologic diagnoses in the year after COVID-19 vs influenza, we performed an analysis of patient-level data from a large collection of electronic health records (EMR). METHODS: We acquired deidentified data from TriNetX, a global health research network providing access to EMR data. We included individuals aged 18 years or older during index event, defined as hospital-based care for COVID-19 (from April 1, 2020, until November 15, 2021) or influenza (from 2016 to 2019). The study outcomes were subsequent health care encounters over the following year for 6 neurologic diagnoses including migraine, epilepsy, stroke, neuropathy, movement disorders, and dementia. We also created a composite of the 6 diagnoses as an incident event, which we call "incident neurologic diagnoses." We performed a 1:1 complete case nearest-neighbor propensity score match on age, sex, race/ethnicity, marital status, US census region patient residence, preindex years of available data, and Elixhauser comorbidity score. We fit time-to-event models and reported hazard ratios for COVID-19 vs influenza infection. RESULTS: After propensity score matching, we had a balanced cohort of 77,272 individuals with COVID-19 and 77,272 individuals with influenza. The mean age was 51.0 ± 19.7 years, 57.7% were female, and 41.5% were White. Compared with patients with influenza, patients with COVID-19 had a lower risk of subsequent care for migraine (HR 0.645, 95% CI 0.604-0.687), epilepsy (HR 0.783, 95% CI 0.727-0.843), neuropathies (HR 0.567, 95% CI 0.532-0.604), movement disorders (HR 0.644, 95% CI 0.598-0.693), stroke (HR 0.904, 95% CI 0.845-0.967), or dementia (HR 0.931, 95% CI 0.870-0.996). Postinfection incident neurologic diagnoses were observed in 2.79% of the COVID-19 cohort vs 4.91% of the influenza cohort (HR 0.618, 95% CI 0.582-0.657). DISCUSSION: Compared with a matched cohort of adults with a hospitalization or emergency department visit for influenza infection, those with COVID-19 had significantly fewer health care encounters for 6 major neurologic diagnoses over a year of follow-up. Furthermore, we found that COVID-19 infection was associated with a lower risk of an incident neurologic diagnosis in the year after infection.

Disparities in seizure outcomes revealed by large language models.

OBJECTIVE: Large-language models (LLMs) can potentially revolutionize health care delivery and research, but risk propagating existing biases or introducing new ones. In epilepsy, social determinants of health are associated with disparities in care access, but their impact on seizure outcomes among those with access remains unclear. Here we (1) evaluated our validated, epilepsy-specific LLM for intrinsic bias, and (2) used LLM-extracted seizure outcomes to determine if different demographic groups have different seizure outcomes. MATERIALS AND METHODS: We tested our LLM for differences and equivalences in prediction accuracy and confidence across demographic groups defined by race, ethnicity, sex, income, and health insurance, using manually annotated notes. Next, we used LLM-classified seizure freedom at each office visit to test for demographic outcome disparities, using univariable and multivariable analyses. RESULTS: We analyzed 84 675 clinic visits from 25 612 unique patients seen at our epilepsy center. We found little evidence of bias in the prediction accuracy or confidence of outcome classifications across demographic groups. Multivariable analysis indicated worse seizure outcomes for female patients (OR 1.33, P ≤ .001), those with public insurance (OR 1.53, P ≤ .001), and those from lower-income zip codes (OR ≥1.22, P ≤ .007). Black patients had worse outcomes than White patients in univariable but not multivariable analysis (OR 1.03, P = .66). CONCLUSION: We found little evidence that our LLM was intrinsically biased against any demographic group. Seizure freedom extracted by LLM revealed disparities in seizure outcomes across several demographic groups. These findings quantify the critical need to reduce disparities in the care of people with epilepsy.

Generalization of finetuned transformer language models to new clinical contexts.

Objective: We have previously developed a natural language processing pipeline using clinical notes written by epilepsy specialists to extract seizure freedom, seizure frequency text, and date of last seizure text for patients with epilepsy. It is important to understand how our methods generalize to new care contexts. Materials and methods: We evaluated our pipeline on unseen notes from nonepilepsy-specialist neurologists and non-neurologists without any additional algorithm training. We tested the pipeline out-of-institution using epilepsy specialist notes from an outside medical center with only minor preprocessing adaptations. We examined reasons for discrepancies in performance in new contexts by measuring physical and semantic similarities between documents. Results: Our ability to classify patient seizure freedom decreased by at least 0.12 agreement when moving from epilepsy specialists to nonspecialists or other institutions. On notes from our institution, textual overlap between the extracted outcomes and the gold standard annotations attained from manual chart review decreased by at least 0.11 F1 when an answer existed but did not change when no answer existed; here our models generalized on notes from the outside institution, losing at most 0.02 agreement. We analyzed textual differences and found that syntactic and semantic differences in both clinically relevant sentences and surrounding contexts significantly influenced model performance. Discussion and conclusion: Model generalization performance decreased on notes from nonspecialists; out-of-institution generalization on epilepsy specialist notes required small changes to preprocessing but was especially good for seizure frequency text and date of last seizure text, opening opportunities for multicenter collaborations using these outcomes.

Costs and Utilization of New-to-Market Neurologic Medications.

BACKGROUND AND OBJECTIVES: The objective of this study was to compare the utilization and costs (total and out-of-pocket) of new-to-market neurologic medications with existing guideline-supported neurologic medications over time. METHODS: We used a healthcare pharmaceutical claims database (from 2001 to 2019) to identify patients with both a diagnosis of 1 of 11 separate neurologic conditions and either a new-to-market medication or an existing guideline-supported medication for that condition. Neurologic conditions included orthostatic hypotension, spinal muscular atrophy, Duchenne disease, Parkinson disease, multiple sclerosis, amyotrophic lateral sclerosis, myasthenia gravis, Huntington disease, tardive dyskinesia, transthyretin amyloidosis, and migraine. New-to-market medications were defined as all neurologic medications approved by the US Food and Drug Administration (FDA) between 2014 and 2018. In each year, we determined the median out-of-pocket and standardized total costs for a 30-day supply of each medication. We also measured the proportion of patients receiving new-to-market medications compared with all medications specific for the relevant condition. RESULTS: We found that the utilization of most new-to-market medications was small (<20% in all but 1 condition), compared with existing, guideline-supported medications. The out-of-pocket and standardized total costs were substantially larger for new-to-market medications. The median (25th percentile, 75th percentile) out-of-pocket costs for a 30-day supply in 2019 were largest for edaravone ($712.8 [$59.8-$802.0]) and eculizumab ($91.1 [$3.0-$3,216.4]). For new-to-market medications, the distribution of out-of-pocket costs was highly variable and the trends over time were unpredictable compared with existing guideline-supported medications. DISCUSSION: Despite the increasing number of FDA-approved neurologic medications, utilization of newly approved medications in the privately insured population remains small. Given the high costs and similar efficacy for most of the new medications, limited utilization may be appropriate. However, for new medications with greater efficacy, future studies are needed to determine whether high costs are a barrier to utilization.

Prescription Opioid Initiation for Neuropathy, Headache, and Low Back Pain: A US Population-based Medicare Study.

Neuropathy, headache, and low back pain (LBP) are common conditions requiring pain management. Yet little is known regarding whether access to specialists impacts opioid prescribing. We aimed to identify factors associated with opioid initiation among opioid-naïve older adults and evaluate how access to particular specialists impacts prescribing. This retrospective cohort study used a 20% Medicare sample from 2010 to 2017. Opioid initiation was defined as a first opioid prescription filled within 12 months after a diagnosis encounter. Disease-related opioid initiation was defined as a first opioid prescription filled within 7 days following a disease-specific claim. Logistic regression using generalized estimating equations was used to determine the association of patient demographics, provider types, and regional physician specialty density with disease-related opioid initiation, accounting for within-region correlation. We found opioid initiation steadily declined from 2010 to 2017 (neuropathy: 26-19%, headache: 31-20%, LBP: 45-32%), as did disease-related opioid initiation (4-3%, 12-7%, 29-19%) and 5 to 10% of initial disease-related prescriptions resulted in chronic opioid use within 12 months of initiation. Certain specialist visits were associated with a lower likelihood of disease-related opioid initiation compared with primary care. Residence in high neurologist density regions had a lower likelihood of disease-related opioid initiation (headache odds ratio [OR] .76 [95% CI: .63-.92]) and LBP (OR .7 [95% CI: .61-.81]) and high podiatrist density regions for neuropathy (OR .56 [95% CI: .41-.78]). We found that specialist visits and greater access to specialists were associated with a lower likelihood of disease-related opioid initiation. These data could inform strategies to perpetuate reductions in opioid use for these common pain conditions. PERSPECTIVE: This article presents how opioid initiation for opioid-naïve patients with newly diagnosed neuropathy, headache, and LBP varies across providers. Greater access to certain specialists decreased the likelihood of opioid initiation. Future work may consider interventions to support alternative treatments and better access to specialists in low-density regions.

Predictors of referral for long-term EEG monitoring for Medicare beneficiaries with drug-resistant epilepsy.

OBJECTIVE: For people with drug-resistant epilepsy, the use of epilepsy surgery is low despite favorable odds of seizure freedom. To better understand surgery utilization, we explored factors associated with inpatient long-term EEG monitoring (LTM), the first step of the presurgical pathway. METHODS: Using 2001-2018 Medicare files, we identified patients with incident drug-resistant epilepsy using validated criteria of ≥2 distinct antiseizure medication (ASM) prescriptions and ≥1 drug-resistant epilepsy encounter among patients with ≥2 years pre- and ≥1 year post-diagnosis Medicare enrollment. We used multilevel logistic regression to evaluate associations between LTM and patient, provider, and geographic factors. We then analyzed neurologist-diagnosed patients to further evaluate provider/environmental characteristics. RESULTS: Of 12 044 patients with incident drug-resistant epilepsy diagnosis identified, 2% underwent surgery. Most (68%) were diagnosed by a neurologist. In total, 19% underwent LTM near/after drug-resistant epilepsy diagnosis; another 4% only underwent LTM much prior to diagnosis. Patient factors most strongly predicting LTM were age <65 (adjusted odds ratio 1.5 [95% confidence interval 1.3-1.8]), focal epilepsy (1.6 [1.4-1.9]), psychogenic non-epileptic spells diagnosis (1.6 [1.1-2.5]) prior hospitalization (1.7, [1.5-2]), and epilepsy center proximity (1.6 [1.3-1.9]). Additional predictors included female gender, Medicare/Medicaid non-dual eligibility, certain comorbidities, physician specialties, regional neurologist density, and prior LTM. Among neurologist-diagnosed patients, neurologist <10 years from graduation, near an epilepsy center, or epilepsy-specialized increased LTM likelihood (1.5 [1.3-1.9], 2.1 [1.8-2.5], 2.6 [2.1-3.1], respectively). In this model, 37% of variation in LTM completion near/after diagnosis was explained by individual neurologist practice and/or environment rather than measurable patient factors (intraclass correlation coefficient 0.37). SIGNIFICANCE: A small proportion of Medicare beneficiaries with drug-resistant epilepsy completed LTM, a proxy for epilepsy surgery referral. While some patient factors and access measures predicted LTM, non-patient factors explained a sizable proportion of variance in LTM completion. To increase surgery utilization, these data suggest initiatives targeting better support of neurologist referral.

Patient Travel Distance to Neurologist Visits.

BACKGROUND AND OBJECTIVES: The density of neurologists within a given geographic region varies greatly across the United States. We aimed to measure patient travel distance and travel time to neurologist visits, across neurologic conditions and subspecialties. Our secondary goal was to identify factors associated with long-distance travel for neurologic care. METHODS: We performed a cross-sectional analysis using a 2018 Medicare sample of patients with at least 1 outpatient neurologist visit. Long-distance travel was defined as driving distance ≥50 miles 1-way to the visit. Travel time was measured as driving time in minutes. Multilevel generalized linear mixed models with logistic link function, which accounted for clustering of patients within hospital referral region and allowed modeling of region-specific random effects, were used to determine the association of patient and regional characteristics with long-distance travel. RESULTS: We identified 563,216 Medicare beneficiaries with a neurologist visit in 2018. Of them, 96,213 (17%) traveled long distance for care. The median driving distance and time were 81.3 (interquartile range [IQR]: 59.9-144.2) miles and 90 (IQR: 69-149) minutes for patients with long-distance travel compared with 13.2 (IQR: 6.5-23) miles and 22 (IQR: 14-33) minutes for patients without long-distance travel. Comparing across neurologic conditions, long-distance travel was most common for nervous system cancer care (39.6%), amyotrophic lateral sclerosis [ALS] (32.1%), and MS (22.8%). Many factors were associated with long-distance travel, most notably low neurologist density (first quintile: OR 3.04 [95% CI 2.41-3.83] vs fifth quintile), rural setting (4.89 [4.79-4.99]), long-distance travel to primary care physician visit (3.6 [3.51-3.69]), and visits for ALS and nervous system cancer care (3.41 [3.14-3.69] and 5.27 [4.72-5.89], respectively). Nearly one-third of patients bypassed the nearest neurologist by 20+ miles, and 7.3% of patients crossed state lines for neurologist care. DISCUSSION: We found that nearly 1 in 5 Medicare beneficiaries who saw a neurologist traveled ≥50 miles 1-way for care, and travel burden was most common for lower-prevalence neurologic conditions that required coordinated multidisciplinary care. Important potentially addressable predictors of long-distance travel were low neurologist density and rural location, suggesting interventions to improve access to care such as telemedicine or neurologic subspecialist support to local neurologists. Future work should evaluate differences in clinical outcomes between patients with long-distance travel and those without.