BACKGROUND:Differentiating between motor functional dissociative seizures (FDS) and motor epileptic seizures (ES) is a common diagnostic challenge, requiring video electroencephalography (vEEG) as gold standard. However, vEEG requires specialized technicians and clinical experts to set up and interpret and oftentimes fails to capture events. We sought to develop machine-learning (ML) tools to carry out this diagnostic task independently of vEEG or human review by a neurologist. METHODS:In this retrospective study, we developed two proof-of-concept ML models to differentiate motor ES from FDS based on video of FDS and ES events in patients who underwent inpatient vEEG monitoring at an academic medical center between 2012 and 2021. The first model employed a pose-estimation approach, using body landmark features that were labeled frame-by-frame by three neurologists. The second utilized an end-to-end 3D convolutional neural network (CNN), thereby learning directly from raw video frames. Using board-certified epileptologist review as a clinical gold standard, we measured model performance by area under the receiver-operating (AUROC) and precision-recall (AUPRC) curves, sensitivity, precision, and accuracy against a held-out test set of videos. RESULTS:We included 101 unique patients with 106 total event videos, comprising 61 (60.4%) ES and 45 (44.6%) FDS events. Both ML models distinguished both seizure types better than chance. The pose-estimation-based model achieved an AUROC of 0.71, AUPRC 0.53, sensitivity 0.90, precision 0.50, and accuracy 0.62. The CNN model exhibited superior overall performance, achieving an AUROC 0.78, AUPRC 0.84, balanced sensitivity and precision (both 0.82), and accuracy 0.80. CONCLUSION:Our findings demonstrate the superiority of CNN over pose-estimation models to differentiate between motor ES and FDS using video alone. Although future studies are needed, these models hold potential as adjunct diagnostic tools by enabling rapid, objective seizure evaluations without immediate neurologist involvement.

Background/Objectives: The aim of this pilot study was to evaluate the feasibility of developing individualized machine learning models using nocturnal wearable-derived autonomic nervous system (ANS) and sleep metrics to predict next-day headache risk in patients with migraine. We also examined the associations between nocturnal ANS and sleep measures and patient-reported outcome measures (PROMs) related to nociplastic pain, migraine burden, and non-restorative sleep (NRS). Methods: Adults with migraine wore the wrist-worn Empatica EmbracePlus® wearable during sleep and completed daily headache diaries for approximately 4 weeks (N = 10). Participants also completed daily headache diaries and PROMs assessing nociplastic pain, migraine burden, and non-restorative sleep. Personalized machine learning (ML) models were developed to predict next-day headache using nocturnal ANS activity (e.g., pulse rate variability (PRV), electrodermal activity (EDA), respiratory rate (RR)) and sleep metrics (e.g., interruptions, duration, awakenings). Model performance was evaluated using area under the receiver operating characteristic and precision-recall curves (AUROC, AUPRC), sensitivity, specificity, accuracy, and precision. Spearman correlations assessed the relationship between wearable-derived metrics and patient-reported outcome measurements of sleep quality (PROMIS-Fatigue, PROMIS-Sleep Disturbance) and a surrogate marker of nociplastic pain (Fibromyalgia (FM) Score). Results: 9 out of 10 participants wore the EmbracePlus device for at least the target duration of four weeks. For the next-day headache prediction, model performance varied between individuals; area under the ROC curve (AUROC) ranged from 28.2% to 81.2%. Nocturnal measures of EDA were strongly correlated with the FM score (Spearman's rho = 0.72-0.75, p < 0.05). Conclusions: Phasic EDA may warrant further investigation as a potential physiological indicator related to nociplastic pain mechanisms and next-day headache. However, these findings are preliminary, and larger multicenter trials are needed to confirm results of this pilot study.

BACKGROUND:A significant proportion of stroke patients are lost to follow-up (LTFU) after discharge, which may increase risks of morbidity, mortality, and unnecessary hospitalization. We aimed to identify predictors of post-discharge LTFU in acute stroke patients from a large academic hospital system. METHODS:Using the American Heart Association's Get With the Guidelines registry, we conducted a retrospective analysis of acute stroke patients hospitalized at the Mount Sinai Hospital from January 2016 to December 2020. Our primary outcome was post-discharge LTFU (i.e. no ambulatory encounters within 12 months post-discharge). We used the least-absolute square and shrinkage operator (LASSO) for variable selection, then used multiple logistic regression to model the association of selected predictors with our primary outcome. RESULTS:We identified 2,597 patients, of whom 878 (33.8%) were LTFU. Patients LTFU were more likely to be male (52.9% vs. 47.4%, p = 0.0088), admitted for intracerebral hemorrhage (12.1% vs. 8.9%, p = 0.0047), discharged to skilled nursing facilities (19.8% vs. 17.0%, p < 0.001), and transferred from another hospital (48.0% vs. 40.7%, p = 0.0030). These patients were more likely to have a discharge modified Rankin Scale (mRS) score of 4-5 (35.2% vs. 30.2%, p = 0.0060) and had a higher mean discharge NIHSS score (6.1 vs. 4.9, p < 0.001). In the adjusted analysis, patients who self-paid (adjusted odds ratio (aOR) 3.8, 95%CI 1.3-11.4), were discharged to acute care facilities (aOR 5.3, 95%CI 1.5-18.4), or had a mRS of 5 (aOR 2.4, 95%CI 1.0-5.7) had significantly increased odds of LTFU. Patients with Medicare coverage (aOR 0.60, 95%CI 0.40-0.92), discharge to inpatient rehabilitation (aOR 0.54, 95%CI 0.34-0.86), family history of stroke (aOR 0.37, 95%CI 0.18-0.76) had significantly decreased odds of LTFU. CONCLUSIONS:In this study, we identified clinical characteristics and discharge dispositions associated with LTFU after stroke.

Prognostication after intracerebral hemorrhage (ICH) is prone to heterogeneity and bias, which represents a potential opportunity to integrate an artificial intelligence-based approach. We aimed to evaluate the feasibility of a commercially-available large language model in predicting functional outcomes after ICH using routine admission data. We used GPT-4o mini with standardized prompts providing clinical data from the first 24 h of admission to predict 3-month modified Rankin Scale (mRS) in a single-center cohort of 409 ICH patients. We then compared the accuracy of these GPT-predicted scores with patients' actual 3-month outcomes, as well as outcomes predicted by two board-certified neurologists. GPT-predicted scores showed moderate correlation with actual outcomes (r = 0.64 [95% CI: 0.58-0.70]), although they tended to underestimate 3-month mRS (mean difference -0.99 [95% CI: -1.16, -0.83; SD 1.68]). This was similar to the averaged clinician scores (r = 0.68 [95% CI: 0.62-0.73]; mean difference -0.35 [95% CI: -0.49, -0.20; SD 1.47]), although one clinician tended toward underestimation while the other tended toward overestimation. GPT's accuracy in predicting favorable 3-month outcomes (defined as mRS 0-3) was comparable to clinician-averaged scores (AUC 0.86 vs. 0.85), with high sensitivity (88%) and moderate specificity (67%). However, model performance was lower in older patients (r = 0.58 [95% CI: 0.46-0.67]; mean difference -1.44 [95% CI: -1.71, -1.16; SD 1.73]) and those with milder deficits (r = 0.22 [95% CI: 0.06-0.37]; mean difference -1.59 [95% CI: -1.90, -1.28; SD 1.89]). These findings suggest that, although it may underestimate 3-month disability, GPT-4o mini has comparable performance to clinicians and may be a feasible tool to aid prognostication after ICH.

Acute ischemic stroke (AIS) is a leading cause of death and long‐term disability worldwide, where rapid reperfusion remains critical for salvaging brain tissue. Although CT perfusion (CTP) imaging provides essential hemodynamic information, its limitations—including extended processing times, additional radiation exposure, and variable software outputs—can delay treatment. In contrast, non-contrast head CT (NCHCT) is ubiquitously available in acute stroke settings. This study explores a generative artificial intelligence approach to predict key perfusion parameters (relative cerebral blood flow [rCBF] and time-to-maximum [Tmax]) directly from NCHCT, potentially streamlining stroke imaging workflows and expanding access to critical perfusion data. We retrospectively identified patients evaluated for AIS who underwent NCHCT, CT angiography, and CTP. Ground truth perfusion maps (rCBF and Tmax) were extracted from VIZ.ai post-processed CTP studies. A modified pix2pix‐turbo generative adversarial network (GAN) was developed to translate co-registered NCHCT images into corresponding perfusion maps. The network was trained using paired NCHCT–CTP data, with training, validation, and testing splits of 80

PURPOSE:We aimed to assess the prevalence of financial distress, defined as the negative effects of the economic burden of medical care on patients' quality of life, in persons with epilepsy compared to those without epilepsy in a nationally representative sample of U.S. adults. We also aimed to identify associated factors of financial distress in persons with active epilepsy. METHODS:We pooled cross-sectional data from the 2021 and 2022 National Health Interview Survey. We divided individuals with epilepsy into active (medication use or seizure in the past year) and inactive cohorts. We analyzed measures of financial distress using survey responses from individuals who had self-reported epilepsy. Multivariable logistic regression was conducted to investigate the association between epilepsy status and financial distress, and to identify factors associated with financial distress measures among persons with epilepsy. RESULTS:In a multivariable logistic regression analysis, epilepsy status was not significantly associated with financial distress (OR=1.03, 95% CI: 0.87-1.22, p = 0.76). However, several factors were significantly associated with higher odds of financial distress among those with active epilepsy, including female sex (OR=1.38, 95% CI: 1.32-1.44), Hispanic ethnicity (OR=1.72, 95% CI 1.61-1.85), having a high school degree or less (OR=1.34, 95% CI: 1.26-1.41), one or more annual hospitalizations (OR=1.33, 95% CI: 1.27-1.40), and multiple comorbidities (OR=1.76, 95% CI: 1.60-1.93). CONCLUSION:A substantial proportion of individuals with active epilepsy experience financial distress, underscoring the need for system-level policies to expand access and patient-level strategies to improve financial navigation and promote affordable therapies.

Introduction: Interprofessional electronic consultations (eConsults) can reduce health care utilization and improve access to specialty care. However, health care utilization and access impacts of eConsults for headache disorders remain incompletely characterized. Methods: We conducted a retrospective, 1:3-matched cohort study comparing patients referred for in-person headache evaluations to patients who had a headache-related eConsult. The cohorts were propensity score-matched by age, sex, race, preferred language, provider specialty, insurance status, and medical comorbidities. Our primary outcome was the presence of one or more headache-related ambulatory encounters in the 12 months following the index referral date. We used univariable and conditional logistic regression models to ascertain the associations between referral type and outcome. Results: We identified 74 and 222 patients with eConsult and in-person referrals, respectively. Over the follow-up period, the proportion of patients with the primary outcome was significantly greater in the eConsult cohort than the in-person cohort (46.0% vs. 43.2%, p < 0.0001). A greater proportion of the in-person cohort had one or more ambulatory headache encounters in the 12 months preceding their referral than the eConsult cohort (10.8% vs. 5.4%, p < 0.0001). In the adjusted analysis, eConsult usage was not associated with significantly increased odds of the primary outcome (adjusted odds ratio [aOR] 1.1, 95% confidence interval [CI] 0.6-2.0, p = 0.71), although patients with one or more ambulatory neurology encounters in the preceding 12 months had significantly increased odds of the primary outcome (aOR 3.1, 95% CI 1.2-7.9, p = 0.015). Conclusion: Compared to in-person referrals, eConsult use for headache was not associated with significantly increased odds of having subsequent ambulatory headache-related encounters.

Clinical informatics (CI) is an emerging field within biomedical informatics that sits at the intersection of clinical care, health systems, and health information technology (IT). CI emphasizes how individuals (neurologists, patients, staff) interact with health IT (HIT), with a focus on designing systems that support optimal neurologic care. As neurology becomes more complex-expanding diagnostics, treatments, and subspecialties-there is a growing need for usable, efficient electronic health record systems that enhance, rather than burden, care delivery. This paper reviews roles CI neurologists can play, as translators, architects, advocates, and leaders across clinical, operational, and strategic domains. We highlight examples where CI expertise addresses challenges in neurology, including access to care, documentation burden, burnout, quality improvement, patient engagement, artificial intelligence, research registries, and precision health. With projected workforce shortages in neurology and CI, neurologists with CI expertise will ensure that HIT will effectively support high-quality neurologist-led care.

Objective:To investigate whether nocturnal autonomic nervous system (ANS) activity and sleep metrics, as measured by a wearable device, can predict the occurrence of next-day migraine in patients with episodic and chronic migraine. Background:The unpredictable nature of migraine episodes contributes to disease burden and limits effective application of tailored preventive strategies. Small-molecule calcitonin-gene-related peptide (CGRP) receptor antagonists, available in limited monthly quantities, are now used for both acute and preventative treatment of migraine. Consequently, improving the ability to identify days with heightened migraine risk could significantly improve migraine management and treatment outcomes. Methods:In this prospective and observational study, adults with migraine (N = 10; 5 with chronic migraine and 5 with episodic migraine) wore the Empatica EmbracePlus®, smartwatch during sleep for a target duration of four weeks. Participants kept a headache diary recording days with no headache, non-migraine headache only, or migraine. First, group level analysis was performed using linear mixed-effects models (LMM). Next, personalized machine learning (ML) models were trained using nocturnal electrodermal activity (EDA), pulse rate variability (PRV), respiratory rate (RR), sleep duration, sleep interruptions, and awakenings to predict: (1) next-day migraine, and (2) next-day headache (both migraine and non-migraine). Performance was summarized using area under the receiver-operating and precision-recall curves (AUROC, AUPRC), sensitivity, specificity, accuracy, and precision. SHapley Additive exPlanation (SHAP) analyses identified the most influential predictors in highest performing next-day migraine and next-day headache models. Generalized Additive Models (GAM) explored nocturnal temporal dynamics of PRV and EDA. Results:Group level predictive performance assessed with LMMs did not reveal significant differences between ANS and sleep metrics on nights prior to no headache days, days with migraine, and days with non-migraine headache. However, individualized models using elastic-net regression, random forests, and gradient boosting machines showed modestly better-than-random AUROCs for next-day migraine prediction in 5/10 participants and next-day headache in 3/10 participants. For next-day migraine prediction models, four of five patients with episodic migraine showed better-than-random AUROCs; no patients with chronic migraine had better-than-random AUROCs. The highest-performing individualized models achieved moderate-to-good performance (AUROC 0.68 for next-day migraine and 0.81 for next-day headache). In highest performing models, SHAP analyses demonstrated sleep duration and a higher minimum PRV influenced next-day migraine and next-day headache probability, while EDA influenced next-day migraine, but not next-day headache. GAM analyses demonstrated that the first three hours after sleep onset and prior to awakening were time periods when PRV and EDA differed prior to a day with migraine or headache in these high performing models. Conclusions:Our findings indicate that applying individualized ML models to wearable-derived autonomic and sleep data may assist in the identification of heightened migraine risk and identified EDA, PRV, and sleep duration as important forecasting features. Our results provide a rationale for future studies that investigate how targeted medication and behavioral interventions on high-risk days may enhance therapeutic precision of migraine treatment and emphasize the importance of defining mechanistic subgroups of patients with migraine most likely to benefit from predictive modeling.

This manuscript examines the expanding role of population health strategies in neurology, emphasizing systemic approaches that address neurological health at a community-wide level. Key themes include interdisciplinary training in public health, policy reform, biomedical informatics, and the transformative potential of artificial intelligence (AI) and large language models (LLMs). In doing so, neurologists increasingly adopt a holistic perspective that targets the social determinants of health, integrates advanced data analytics, and fosters cross-sector collaborations-ensuring that prevention and early intervention are central to their efforts. Innovative applications, such as predictive analytics for identifying high-risk populations, digital twin technologies for simulating patient outcomes, and AI-enhanced diagnostic tools, illustrate the transition in neurology from reactive care to proactive, data-driven interventions. Examples of transformative practices include leveraging wearable health technologies, telemedicine, and mobile clinics to improve early detection and management of neurological conditions, particularly in underserved populations. These emerging methodologies expand access to care while offering nuanced insights into disease progression and community-specific risk factors. The manuscript emphasizes health disparities and ethical considerations in designing inclusive, data-driven interventions. By harnessing emerging technologies within frameworks that prioritize equity, neurologists can reduce the burden of neurological diseases, improve health outcomes, and establish a sustainable, patient-centered model of care benefiting both individuals and entire communities. This integration of technology, interdisciplinary expertise, and community engagement fosters a future where brain health is preventive, accessible, and equitable.

Mobile applications are widely used in epilepsy, although their impact on clinical effectiveness (CE) and their feasibility, acceptability, and usability (FAU) remain unclear. We conducted a systematic review investigating CE and FAU of epilepsy mobile applications using MEDLINE and Embase from database inception to June 21, 2024. We followed Preferred Reporting Items for Systematic Reviews and Meta-Analyses reporting standards. The protocol was registered on PROSPERO (CRD42019134848). In duplicate, we determined study quality using the Newcastle-Ottawa Quality Assessment Scale (NOQAS) and the Joanna Briggs Critical Appraisal Checklist (to determine eligibility for inclusion), risk of bias using the Cochrane Risk of Bias tool, and usability study quality using the 15-point Silva scale. We identified 8953 studies, of which 20 were included. Twelve (60.0%) addressed CE, nine (45.0%) acceptability, five (25.0%) usability, and eight (40.0%) feasibility. Five (25.0%) evaluated CE and FAU. Studies comprised prospective cohort (n = 9, 45.0%), pilot (n = 3, 15.0%), randomized controlled trial (n = 7, 35.0%), and pre/post (n = 1, 5.0%) designs. Most apps were used for self-management or to enhance education or communication between patients and providers. Cohort studies demonstrated fair quality (median NOQAS score = 5, interquartile range [IQR] = 5.0-5.8), whereas of seven randomized controlled trials, four (57.1%) had some concern for bias. Usability studies demonstrated high quality (median Silva score = 10, IQR = 10-11). Apps were predominantly intended for patient use (n = 9, 75.0%). Symptom reporting and medication management were the most common app targets in both CE and FAU studies (n = 8, 66.7%; n = 9, 69.2%), although FAU studies more frequently used monitoring or tracking (n = 10, 76.9%) and reminder setting (n = 10, 76.9%) than CE apps (n = 7, 58.3%). Investigations of application use most commonly studied CE and patient-facing apps. Additional high-quality evidence is necessary to evaluate the CE and FAU of app use in epilepsy to work toward the standardization of FAU metrics and development of implementation guidelines.

Acute stroke alerts are frequently triggered by conditions unrelated to cerebrovascular disease, resulting in false positives that burden clinical teams and contribute to diagnostic ambiguity. At a large academic center, we developed ScanNER v2, a machine learning (ML) model based on large-language models (LLMs) and structured clinical data to predict the presence of acute cerebrovascular disease (ACD) in approximately 16,000 stroke alerts occurring over 10 years with an area under the receiver-operating curve and F1 score of 0.72 and overall positive predictive value of 0.68. In this perspective, we outline a practical framework for operationalizing this model within hospital-based stroke systems. We first describe our health-system experience developing and validating an AI-enabled pipeline, named "ScanNER 2," then take the point of view of two implementation angles (high sensitivity and high specificity), outlining the operational and clinical tradeoffs for each approach. We also highlight challenges related to implementation, clinical governance, workflow integration, and equity, emphasizing guardrails required for responsible deployment. As stroke centers increasingly adopt AI-assisted tools, this type of thought experiment is essential to ensure that such ML-based innovations effectively enhance the core mission of delivering timely, high-quality acute stroke care.

BACKGROUND:The modified Rankin Scale (mRS) is the most common measure of post-stroke functional outcome. However, its impact on post-stroke care is limited by its subjectivity, impracticality in vulnerable populations, and susceptibility to cultural and language barriers. Artificial intelligence (AI) applied to triaxial wrist-worn accelerometry data may offer an objective way to characterize post-stroke functional status and related changes. METHODS:We analyzed data from the REACH Stroke-Sleep study at Columbia University Irving Medical Center. Using single-day epochs averaged over the first 30 days of recording, we trained logistic repression (LR), random forest (RF) and long short-term memory (LSTM) models to predict 1-month and 6-month mRS scores, as well as 1- to-6 month mRS changes. 5-fold cross validation was used, and model performance was evaluated using area under receiver-operating curve (AUROC) for binary exact-match predictions. RESULTS:We identified 362 patients, of whom 302 (83.4 %) had 1-month and 251 (69.3 %) had 6-month mRS scores. RF models (1-month AUROC 0.65, 95 %CI 0.57-0.73; 6-month 0.59, 95 %CI 0.51-0.66; ΔmRS 0.62, 95 %CI 0.53-0.71) outperformed LSTM (0.55, 95 %CI 0.48-0.63; 0.53, 95 %CI 0.45-0.60, 0.49; 95 %CI 0.39-0.59) and LR (0.54, 95 %CI 0.47-0.61; 0.58, 95 %CI 0.50-0.66; 0.50, 95 %CI 0.40- 0.61) models. CONCLUSIONS:AI applied to wearable accelerometry predicted short-term mRS score and functional change after stroke with accuracy above chance. Although predictive performance was modest, these results provide proof-of-concept that AI applied to passive wearable data can capture meaningful post-stroke functional variation and could inform future real-time monitoring frameworks. Integrating multimodal wearable and clinical data may further improve prediction of post-stroke functional outcomes.

Background:Acute stroke alerts are often activated for non-cerebrovascular conditions, leading to false positives that strain clinical resources and promote diagnostic uncertainty. We sought to develop machine learning (ML) models integrating large-language models (LLMs), structured electronic health record data, and clinical time series data to predict the presence of acute cerebrovascular disease (ACD) at stroke alert activation. Methods:We derived a series of ML models using retrospective data from stroke alerts activated at Mount Sinai Health System between 2011 and 2021. We extracted structured data (demographics, medical comorbidities, medications, and engineered time-series features from vital signs and lab results) as well as unstructured clinical notes available prior to the time of stroke alert. We processed clinical notes using three embedding approaches: word embeddngs, biomedical embeddings (BioWordVec), and LLMs. Using a radiographic gold standard for acute intracranial vascular event, we used an auto-ML approach to train one model based on unstructured data and five models based on different combinations of structured data. We evaluated models individually using the area under the receiver operating characteristic curve (AUROC), mean positive predictive value (PPV), sensitivity, and F1-score. We then combined the 6 model logits into a multimodal ensemble by weighting their logits based on F1-score, determining ensemble performance using the same metrics. Results:We identified 16,512 stroke alerts corresponding to 14,233 unique patients over the study period, of which 9,013 (54.6%) were due to ACD. The multi-modal model (AUROC 0.72, PPV 0.68, sensitivity 0.76, F1 0.72) outperformed all individual models by AUROC. One structured model based on demographics, comorbidities, and medications demonstrated the highest sensitivity (0.95). Conclusions:We developed a multi-modal ML model to predict ACD at stroke alert activation. This approach has promise to optimize stroke triage and reduce false-positive activations.

BACKGROUND:Interprofessional electronic consultations (eConsults) can improve access to specialty care and potentially reduce the need for in-person visits. However, their impact on health care utilization remains poorly characterized. METHODS:We conducted a retrospective, 1:1 propensity score-matched cohort study comparing patients referred for neurology evaluations via an eConsult or a traditional (in-person) referral at an urban academic medical center. Groups were matched by sociodemographic characteristics. The primary outcome was attendance of an outpatient neurology visit within 12 months of a referral or an eConsult. The secondary outcomes included the time to the first neurology visit within 6 and 12 months. Multivariable logistic regression and Cox proportional hazards models evaluated the outcomes. RESULTS:We identified 828 patients in each group between October 2019 and December 2023. At 12 months, eConsult patients were significantly less likely to attain the primary outcome compared with traditional referral patients (50.0% vs. 57.7%, p = 0.002), with multivariable analysis confirming significantly lower odds (odds ratios [OR]: 0.73, 95% CI: 0.60-0.89, p = 0.002). In the 6-month analysis, eConsults were associated with a significantly faster time to visit (45 vs. 59 days, p = 0.01), confirmed in adjusted Cox analyses (hazard ratio [HR]: 1.29, 95% CI: 1.12-1.49, p < 0.001). Over 12 months, the time to the neurology visit was shorter for eConsult patients by 19 days (p = 0.044), but these differences were not significant in adjusted Cox analyses (HR: 1.11, 95% CI: 0.98-1.27, p = 0.11). CONCLUSIONS:Neurology eConsults were associated with lower follow-up visit rates and faster access to care in the short term, supporting their potential role in accelerating access to neurological expertise and reducing the need for in-person visits.

Background: Functional outcomes after stroke are commonly assessed via modified Rankin Scale (mRS). However, mRS is subject to patient and assessor biases and is impractical to collect in many cases, limiting its impact on post-stroke care. Artificial intelligence (AI) applied to wrist-worn triaxial accelerometry (WWTA) device data can objectively characterize post-stroke functional status and related changes. Methods: We used patient data from REACH Stroke-Sleep, a study investigating WWTA-derived measures of sleep, physical activity, and recurrent stroke risk among patients with acute stroke symptoms. We determined moving accelerometry averages and vector sums over four time windows (minute, hour, day, week). We trained a tree-based (random forest; RF) and deep learning (LSTM) model to predict individual 6-month mRS scores and differences between 1- and 6-month mRS scores. We used 5-fold cross validation, modeled each outcome as binary exact-match between actual-predicted values, and determined area under the receiver-operating curve (AUROC), sensitivity, precision, negative predictive value, and F1 scores for both models. For mRS score differences, we determined mean absolute error (MAE) and standard deviation (SD). Results: We identified 362 patients in REACH Stroke-Sleep, of whom 302 (83.4%) had a 1-month mRS score, 251 (69.3%) had a 6-month mRS score, and 191 (52.8%) had both. Patients wore devices for median 41.0 (IQR 34.4-44.0) days. For all outcomes, RF models (6-month AUROC 0.81, 95%CI 0.74-0.89; 1-6 month mRS AUROC 0.82, 95%CI 0.76-0.90) outperformed LSTM models (6-month AUROC 0.63, 95%CI 0.55?0.71; 1-6 month mRS AUROC 0.53, 95%CI 0.45-0.61). RF models (MAE 0.37, SD 0.12) outperformed LSTM (MAE 0.87, SD 0.48) for predicting 1-6 month mRS difference, modeled as a non-binarized outcome. Conclusions: We found that AI predicted short-term mRS and mRS changes after acute stroke symptoms from WWTA data with moderate performance. Future studies are warranted to investigate whether multimodal data can improve performance with the goal of developing objective, automatable functional status assessments ### Competing Interest Statement de minimis equity holding in Syntrillo (BK) ### Funding Statement Funding Support: CTSA grant UL1TR004419 (Kummer), NIH grants R01HL155915 and R01HL167050 (Nadkarni), R01NS121364 (Willey), and R01HL141494 (Shechter). ### Author Declarations I confirm all relevant ethical guidelines have been followed, and any necessary IRB and/or ethics committee approvals have been obtained. Yes The details of the IRB/oversight body that provided approval or exemption for the research described are given below: All procedures were approved by the IRB of Columbia University Irving Medical Center, and as per IRB approvals, de-identified data were shared with the Icahn School of Medicine at Mount Sinai. The IRB of Mount Sinai approved the use of the de-identified patient data shared by Columbia University Irving Medical Center for this research. I confirm that all necessary patient/participant consent has been obtained and the appropriate institutional forms have been archived, and that any patient/participant/sample identifiers included were not known to anyone (e.g., hospital staff, patients or participants themselves) outside the research group so cannot be used to identify individuals. Yes I understand that all clinical trials and any other prospective interventional studies must be registered with an ICMJE-approved registry, such as ClinicalTrials.gov. I confirm that any such study reported in the manuscript has been registered and the trial registration ID is provided (note: if posting a prospective study registered retrospectively, please provide a statement in the trial ID field explaining why the study was not registered in advance). Yes I have followed all appropriate research reporting guidelines, such as any relevant EQUATOR Network research reporting checklist(s) and other pertinent material, if applicable. Yes The data used for study is available after reasonable request to Columbia University Vagelos College of Physicians and Surgeons and is subject to the latter?s institutional board requirements and policies.

BACKGROUND:Frequent false-positive stroke alerts can strain health resources. Machine learning models can predict stroke alert accuracy and potentially reduce this strain, but these models require time-consuming labeling of large data sets. Weak labeling can accelerate machine learning model development by assigning annotations based on expert-defined heuristic rules rather than manual review. We sought to label a large, unlabeled sample of stroke alerts according to a binary outcome (presence/absence of acute cerebrovascular disease) using weak labeling. METHODS:We developed a weak labeling heuristic ensemble consisting of 4 hierarchical tiers, each of which generated a binary label using custom labeling algorithms. Tier 1 used rule-based named-entity recognition to generate binary labels from brain radiology reports. Tier 2 aggregated Tier 1 outputs over a 48-hour window. Tier 3 determined labels using diagnosis codes from stroke alert hospital encounters. Tier 4 generated a "final" encounter-level label based on label output combinations of Tiers 2 and 3. In 3 separate samples of stroke alerts, we determined sensitivity, specificity, and F1 scores of Tiers 1, 2, 3, and 4 by comparing Tier outputs to manual chart review. RESULTS:We identified 16 512 stroke alert activations between 2011 and 2021. For Tier 1, performance metrics were based on an initial manual review of 300 neuroimaging reports, achieving a sensitivity of 0.84, specificity of 0.96, and an F1 of 0.87. Tier 2 incorporated 716 neuroimaging reports with a sensitivity of 0.93, specificity of 0.90, and an F1 of 0.91. Tiers 3 and 4 were validated against 250 encounters. Tier 3 achieved a sensitivity of 0.77, specificity of 0.89, and an F1 of 0.80. Tier 4 achieved a sensitivity of 0.92, specificity of 0.86, and an F1 of 0.87. CONCLUSIONS:We successfully labeled a large registry of stroke alerts using weak labeling. This framework can potentially be extended to other clinical data sets.

Introduction: The use of remote patient monitoring (RPM) services for neurological disorders remains understudied, particularly in the context of newer billing codes introduced before the COVID-19 pandemic. Methods: This retrospective cohort study utilized data from commercial and Medicare employer-sponsored administrative claims between January 1, 2019, to December 31, 2021. The study population included all patients with at least one qualifying RPM-related Current Procedural Terminology (CPT) code for a neurological disorder, separated into first-generation (CPT 99091) codes and second-generation (CPT 99453, 99454, 99457, 99458) code cohorts. We compared patient and encounter characteristics between both cohorts. Results: We identified 27,756 encounters attributable to 11,326 patients who received RPM services for neurological disorders, of whom 5,785 (51.1%) received RPM via second-generation billing codes, 3,941 (34.8%) were female, 6,712 (59.3%) were between 45 and 64 years old, and 10,488 (92.6%) had a primary diagnosis of sleep-wake disorder. The second-generation cohort was significantly more likely to be female (41.5% vs. 27.8%, p < 0.001), be of age 65 or older (15.7% vs. 7.1%, p < 0.001), and reside in urban areas (93.4% vs. 87.6%, p < 0.001) than the first-generation cohort. Patients in the second-generation cohort were more likely to receive RPM in office settings (86.3% vs. 62.5%, p < 0.001), by physicians (77.0% vs. 40.3%, p < 0.001), and less likely for sleep-wake disorders (87.9% vs. 97.5%, p < 0.001) than the first-generation cohort. Patients who received RPM from physicians were most often evaluated by pulmonologists (31.4%). Discussion: In this commercially insured patient population receiving RPM for neurological disorders, we found that sleep-wake disorders and non-neurologists were over-represented.

Background Natural language processing (NLP), a branch of artificial intelligence that analyzes unstructured language, is being increasingly used in health care. However, the extent to which NLP has been formally studied in neurological disorders remains unclear. Objective We sought to characterize studies that applied NLP to the diagnosis, prediction, or treatment of common neurological disorders. Methods This review followed the PRISMA-ScR (Preferred Reporting Items for Systematic Reviews and Meta-Analyses Extension for Scoping Reviews) standards. The search was conducted using MEDLINE and Embase on May 11, 2022. Studies of NLP use in migraine, Parkinson disease, Alzheimer disease, stroke and transient ischemic attack, epilepsy, or multiple sclerosis were included. We excluded conference abstracts, review papers, as well as studies involving heterogeneous clinical populations or indirect clinical uses of NLP. Study characteristics were extracted and analyzed using descriptive statistics. We did not aggregate measurements of performance in our review due to the high variability in study outcomes, which is the main limitation of the study. Results In total, 916 studies were identified, of which 41 (4.5%) met all eligibility criteria and were included in the final review. Of the 41 included studies, the most frequently represented disorders were stroke and transient ischemic attack (n=20, 49%), followed by epilepsy (n=10, 24%), Alzheimer disease (n=6, 15%), and multiple sclerosis (n=5, 12%). We found no studies of NLP use in migraine or Parkinson disease that met our eligibility criteria. The main objective of NLP was diagnosis (n=20, 49%), followed by disease phenotyping (n=17, 41%), prognostication (n=9, 22%), and treatment (n=4, 10%). In total, 18 (44%) studies used only machine learning approaches, 6 (15%) used only rule-based methods, and 17 (41%) used both. Conclusions We found that NLP was most commonly applied for diagnosis, implying a potential role for NLP in augmenting diagnostic accuracy in settings with limited access to neurological expertise. We also found several gaps in neurological NLP research, with few to no studies addressing certain disorders, which may suggest additional areas of inquiry. Trial Registration Prospective Register of Systematic Reviews (PROSPERO) CRD42021228703; https://www.crd.york.ac.uk/PROSPERO/display_record.php?RecordID=228703