Abstract Introduction Urine drug testing (UDT) during MSLT is performed to rule out the presence of confounding substances. The 2021 AASM recommendations highlight cannabinoids as particularly relevant to MSLT interpretation, citing reports of shortened sleep latency or, after recent discontinuation, REM rebound. However, conflicting reports exist. Prevalence of cannabinoid use among patients with central disorders of hypersomnolence (CDH) is unknown. Methods Retrospective analysis was performed of patients with suspected CDH undergoing PSG/MSLT at Cleveland Clinic, 3/2009-2/2025, who had immunoassay UDT prior to MSLT. Included patients met ICSD-3 criteria for narcolepsy type 1/2 (NT1/2), idiopathic hypersomnia (IH) or undifferentiated hypersomnia (UH). For analyses involving MSLT parameters, the “cannabinoid positive” group (UDT+) was comprised of patients with UDT positive for cannabinoids alone. Results From 1,031 cases with verified ICSD-3 diagnoses, 91 (8.8%) had UDT positive for cannabinoids. Positivity did not differ between CDH subgroups (6/55 (10.9%) in NT1, 7/98 (7.1%) NT2, 21/258 (8.1%) IH, 57/620 (9.2%) UH; p=0.84). Overall N = 1,019 (54 NT1, 97 NT2, 256 IH, 612 UH) after excluding cases positive for >1 substance. UDT+ and UDT- groups did not differ in age (34.0±11.9 vs 35.0±14.5; p=0.89), female sex (61 (77.2%) vs 686 (73.0%); p=0.49), BMI (26.6[22.7, 31.5] vs 27.2[23.3, 32.3]; p=0.46), or race/ethnicity composition (58 (73.4%) Caucasian vs 734 (78.4%) Caucasian; p=0.49). However, UDT+ had higher Epworth Sleepiness Scale scores (15.2±5.1 vs 14.0±5.2; p=0.025) and lower AHI on PSG (1.00[0.30,3.4] vs 1.8[0.70,5.0]; p=0.011). For MSLT outcomes, UDT+ and UDT- did not differ by total SOREMPs (0.00[0.00,1.00] UDT+ vs 0.00[0.00,1.00] UDT-; p=0.45) or mean sleep latency (MSL) (10.6[5.3,13.5] UDT+ vs 10.0[5.9,14.6] UDT-; p=0.89). Likewise, UDT+ had similar likelihood as UDT- of reaching 2+ SOREMPs (13 (16.5%) UDT+ vs 184 (19.6%) UDT-; p=0.63) or MSL ≤ 8 (30 (30.8%) UDT+ vs 373 (39.7%) UDT-; p=0.75). Conclusion These findings demonstrate that a clinically important percentage of patients with suspected CDH use cannabinoids. Additionally, our results do not support ruling MSLTs as invalid solely due to cannabinoid use. Given the similar prevalence of MSLTs with MSL ≤ 8 and/or 2+ SOREMPs between groups, cannabinoids may not impact classification of CDH subtypes; further research is needed. Support (if any)

Abstract Introduction Polysomnography (PSG) reports rely on a handful of summary indices, such as the apnea-hypopnea index (AHI), which incompletely capture the multidimensional burden of sleep-disordered breathing and its downstream health risks. We developed a risk stratification pipeline that uses RoBERTa-based foundational model on raw PSG signals to generate patient-level embeddings and cluster individuals into risk-groups (RG1-RG5). Here, we perform external, independent validation of these clusters for predicting long-term cardiovascular outcomes using the publicly available community-based Sleep Heart Health Study (SHHS) dataset. Methods We fine-tuned a RoBERTa-based model on 10,000 clinical PSG recordings from Cleveland Clinic (Jan-2012 to Dec-2022) to match PSG scoring data (e.g, sleep stages and respiratory events). Latent embeddings from the model were aggregated at the patient-level and clustered using k-means with energy-distance criteria to define five risk-groups. The trained model and fixed cluster centroids were then applied without retraining to the raw PSG data in SHHS. Primary outcomes were all-cause mortality and incident heart failure. Kaplan-Meier analyses and Cox proportional hazards models, adjusted for age and sex, were performed in both datasets. Results In the clinical cohort, higher risk-groups showed significantly reduced survival for all-cause mortality (log-rank p< 0.001 for RG3-RG5 vs. RG1), with age/gender-adjusted hazards demonstrating a dose-response gradient: RG3 HR=1.51 (95% CI 1.17-1.96), RG4 HR=1.64 (95% CI 1.26-2.15), and RG5 HR=2.71 (95% CI 1.93-3.81), all versus RG1. For incident heart failure, RG5 had a significantly elevated risk (HR=1.67, 95% CI 1.12-2.48, p=0.011 vs. RG1). External validation in SHHS replicated these patterns despite differing PSG setups. For all-cause mortality, RG4 and RG5 remained associated with worse survival (log-rank p< 0.001; HR=1.31, 95% CI 1.07-1.62, and HR=1.58, 95% CI 1.07-2.32, respectively, vs. RG1). For incident heart failure, RG5 again emerged as a high-risk group (log-rank p=0.003; HR=2.13, 95% CI 1.11-4.09, p=0.023 vs. RG1). Conclusion Patient clusters derived from a fine-tuned PSG foundational model showed robust external validity for predicting all-cause mortality and incident heart failure in an independent cohort with substantially different PSG acquisition protocols. These findings support the generalizability and clinical utility of model-based PSG phenotyping for risk stratification. Support (if any) This project was partially supported by NIH 1R21HL170206-01.

Abstract Introduction Differentiation between narcolepsy type 2 (NT2) and idiopathic hypersomnia (IH) is often challenging due to overlapping features and limitations of diagnostic testing. We used machine learning on routine PSG features to directly predict expert clinical diagnosis of IH versus NT2. Methods We conducted a manual chart review of patients undergoing MSLT for suspected central disorders of hypersomnolence (CDH). Only individuals classified clinically as IH or NT2 were included. Forty-five PSG metrics from final reports, including demographics, sleep architecture, and respiratory indices, were extracted. Missing values were imputed using the median (for numeric features) or the mode (for categorical features), and features were standardized. Six machine-learning classifier techniques were evaluated using nested cross-validation (5-fold outer, 5-fold inner) with Optuna hyperparameter optimization (1,000 trials per inner fold). We used ANOVA with p-values< 0.05 within each fold for feature selection. SHAP values were used to quantify feature importance. T-test and chi-squared test were used to assess statistical significance between IH and NT2. Metrics are reported as mean (standard deviation). Results The cohort included 454 patients, N=147(32%) with a clinical diagnosis of IH and N=307 (68%) with NT2. Overall age was 34.0(13.1) years, BMI was 27.6(6.6) kg/m², 351(77.3%) female, and 327(72.0%) Caucasian. Sex distribution differed between IH and NT2 (female 81.4% vs 68.7%). Within the clinically defined IH and NT2 cohort, agreement with ICSD-3 diagnoses was 58% for IH, 67% for NT2, and 61% overall. The logistic regression classifier achieved the best performance, with an AUC-ROC of 66% (5%) and a balanced accuracy of 63% (4%). Using a 60% probability threshold for NT2, precision was 49% (8%), sensitivity was 36% (6%), and specificity was 82% (4%). The SHAP analysis indicated that the features most strongly associated with NT2 were shorter REM latency (p< 0.001), lower non-REM sleep time (p=0.003), higher sleep efficiency in the supine position (p=0.023), and male sex (p=0.014). Conclusion ICSD-3 (MSLT-based) diagnoses show poor agreement with expert clinical diagnoses of NT2/IH, highlighting the limitations of current diagnostic criteria and the need for alternative diagnostic modalities. Machine learning models applied to routine PSG features provide only moderate differentiation between NT2 and IH. Support (if any)

Abstract Introduction The 2021 AASM MSLT protocol recommendations do not provide a uniform requirement for urine drug testing (UDT) in patients undergoing evaluation for central disorders of hypersomnolence (CDH). Rather, UDT may be indicated “depending on clinical and community circumstances.” Selective performance of UDT may result in under-detection of recreational substances, including cannabinoids. Methods Retrospective analysis was performed of patients with suspected CDH undergoing PSG/MSLT at Cleveland Clinic from 3/2009 to 2/2025 who had immunoassay UDT the morning of MSLT. Included patients met ICSD-3 criteria for narcolepsy type 1/2 (NT1/2), idiopathic hypersomnia (IH) or undifferentiated hypersomnia (UH). Disclosure of cannabinoid use was determined by pre-test sleep provider office visit, during which patients were specifically asked about substance use, and from questionnaire responses about recreational drug use. Results Of 1,031 cases with verified ICSD-3 diagnoses (55 NT1, 98 NT2, 258 IH, 620 UH), 85 (8.24%) tested positive for cannabinoids and had documented disclosure data (63 (74.1%) Caucasian, 66 (77.6%) female, age 34.5 ± 11.8). Only 42 patients (49.4%) disclosed cannabinoid use. Differences in age and sex between disclosure and non-disclosure groups were not significant (age 32.7±9.6 vs. 36.2±13.5, p=0.16; female 31 (73.8%) vs. 35 (81.4%), p=0.40, respectively). Race/ethnicity composition was also similar between groups; disclosure group had 33 (78.6%) Caucasian, 7 (16.7%) Black, and 2 (4.8%) Multiracial patients vs. 30 (69.8%) Caucasian, 7 (16.3%) Black, and 6 (14.0%) Multiracial in non-disclosure group (p=0.39). Disclosure across CDH subtypes included 5/6 (83.3%) patients with NT1, 3/6 (50.0%) with NT2, 9/20 (45.0%) with IH, and 25/53 (47.2%) with UH (p = 0.48). Conclusion Among patients with suspected CDH testing cannabinoid-positive on UDT the day of MSLT, nearly half did not disclose their substance use beforehand. Moreover, age, sex, race/ethnicity, and ICSD-3 diagnosis did not correlate with disclosure status. These results suggest that UDT should be universally performed during MSLT, as neither patient report nor demographic factors accurately predict cannabinoid use. Support (if any)

Abstract Introduction Associations between polysomnography (PSG) variables and next-day multiple sleep latency test (MSLT) outcomes across CNS disorders of hypersomnolence (CDH) are limited to observations in narcolepsy type 1 (NT1). We analyzed PSG–MSLT associations a U.S. cohort by diagnostic subgroups (NT1, narcolepsy type 2 (NT2) and idiopathic hypersomnia (IH)). Methods We analyzed valid paired PSG–MSLT studies from NT1, NT2, and IH patients diagnosed by clinical criteria. Analyses used Fisher’s exact tests, Kruskal–Wallis tests, ANOVA, and Spearman correlations. PSG variables were evaluated against MSLT mean sleep latency (MSL) and sleep onset REM period (SOREMP) number in the overall cohort and by subgroup. Results A total of 533 patients were included: mean age 33.9±13.4, 76.9% female. In NT1 (n=81), PSG–MSLT coupling was strongest. MSL showed the highest correlations with PSG sleep latency (ρ=0.59, p< 0.001), REM latency (REML) (ρ=0.34, p=0.002), and latencies to N1–N3. Individuals with a PSG SOREMP had shorter MSL (1.6[1.4, 3.5] vs 4.4[2.7, 7.1] min, p< 0.001), and more MSLT SOREMPs (5.0[5.0, 6.0] vs 2.0[2.0, 4.0], p< 0.001). PSG REML was negatively correlated with number of MSLT SOREMPs (ρ=-0.68, p< 0.001). In NT2 (n=149), observed associations persisted but were attenuated relative to NT1. MSL correlated with PSG sleep latency (ρ=0.26, p=0.001) and REML (ρ=0.20, p=0.017). Individuals with a PSG SOREMP had shorter MSL (3.3[2.2, 5.3] vs 5.1[3.7, 7.3] min, p=0.031) and more MSLT SOREMPs (5.0[4.0, 6.0] vs 2.0[1.0, 3.0], p< 0.001). Negative correlation between MSLT SOREMPs and PSG REML was observed (ρ=-0.41, p< 0.001). IH (n=303) displayed the weakest PSG–MSLT relationships, with patterns distinct from both NT1 and NT2. MSL correlated with PSG sleep latency (ρ=0.25, p< 0.001) and sleep stage latencies N1 and N2 (ρ=0.33 each, p< 0.001), but associations between PSG REML and MSL (ρ=0.10, p=0.098) and MSLT SOREMPS (ρ=-0.19, p>0.05) were weak. Conclusion Patients with clinical phenotypes of CDH demonstrate clear and distinct associations with next-day MSLT. NT1 shows the strongest PSG–MSLT coupling, NT2 moderate, and IH attenuated relationships. These findings underscore the potential value of PSG versus MSLT driven outcomes underscoring the absence of REM related relationships in IH that differentiate it from narcolepsy. Support (if any) None

Abstract Introduction Central disorders of hypersomnolence (CDH), including narcolepsy type 1 (NT1), narcolepsy type 2 (NT2), and idiopathic hypersomnia (IH), are associated with autonomic abnormalities. Animal studies show that orexin loss reduces sympathetic tone. However, clinical studies assessing autonomic function, including heart rate variability (HRV), in NT remain limited and inconsistent. Autonomic impairment is also reported in IH, but HRV is evaluated in only one study revealing increased parasympathetic tone during wake and sleep. The objective was to compare HRV in patients with NT1, NT2, IH and controls Methods Adults with NT1, NT2, IH according to ICSD-3-TR who underwent PSG followed by MSLT (PSG/MSLT) at Cleveland Clinic from 2012-2024 were included. Controls were individuals who completed PSG/MSLT in the same period with no CDH, with habitual sleep time > 7hours and mean sleep latency ≥12 minutes on MSLT. All participants had AHI< 10, PLMI < 10, and were free of medications affecting HRV. HRV indices- standard deviation of NN intervals (SDNN) and root mean square of successive differences (RMSSD) were derived from the ECG signal during PSG. Results A total of 213 adults were analyzed (18 NT1, 39 NT2, 103 IH, and 53 controls), with a predominance of females in all diagnostic groups (77.8% NT1, 69.2% NT2, 78.4% IH, and 71.7% controls). Mean age differed modestly across groups (overall 33.5±12.6 years; p=0.032). No significant difference in overall, NREM sleep, REM sleep, light sleep, deep sleep SDNN and RMSSD among NT1, NT2, IH, and control was observed. Overall SDNN was 68.3ms [52.9, 78.4] in NT1, 64.4ms [50.5, 100.9] in NT2, 61.5ms [46.0, 90.5] in IH, and 68.0ms [57.3, 90.4] in control (p-value 0.39). Overall RMSSD was 74.0ms [47.3, 85.6] in NT1, 67.0ms [49.4, 126.6] in NT2, 57.8ms [42.0, 103.9] in IH, and 66.3ms [53.9, 95.9] in control (p-value 0.52). Conclusion SDNN and RMSSD did not differ significantly among CDH subtypes or controls across sleep stages. Although these indices tended to be consistently lower in IH, differences were not statistically significant, suggesting limited utility of conventional HRV measures in detecting autonomic dysfunction in CDH. Support (if any) NIH grant 1R21HL170206-01

Abstract Introduction Sleep physiology and brain health are deeply interconnected. Fragmented sleep has been implicated in cognitive decline, neuroinflammation, and hippocampal vulnerability. However, studies linking quantitative structural MRI to polysomnography (PSG) metrics remain limited. We leveraged a PSG foundation model and a novel measure of spectral sleep fragmentation (SSF) to test whether PSG-derived measures are associated with MRI markers of neurodegeneration. Methods We fine-tuned a foundation model on 10,000 clinical PSG recordings from the Cleveland Clinic (Jan 2012 to Dec 2022). Latent embeddings were aggregated at the patient level and clustered using k-means to define five previously validated PSG risk-groups (RG1-RG5) associated with distinct clinical outcomes. The SSF metric, a frequency-domain metric computed from the spectral representation of the sleep hypnogram, emerged as the most discriminative feature for separating clusters after age. To investigate structural brain correlates, we identified patients with clinical brain MRI analyzed using the automatic software NeuroQuant, which provides structured volumetric measures of hippocampal and ventricular anatomy. We constructed a composite neurodegeneration index (CNI) by combining hippocampal and inferior-lateral ventricular percentiles into z-scores, with higher values indicating greater neurodegenerative change. After filtering for subjects with valid SSF, risk-group assignment, and MRI data, we computed Pearson correlations between SSF and the composite index overall and within each risk group. Results We analyzed 271 MRI reports (Age: 65.7±11.9, Male: 51.4%, BMI: 29.8±6.7). In the overall sample, higher SSF was modestly associated with greater neurodegenerative change (n=271, r=0.19, p=0.002). Stratified analyses per group revealed heterogeneous associations between SSF and CNI across risk-groups. Correlations were small and not significant in RG1-RG4 (RG1:n=89, r=0.17,p=0.10; RG2:n=40,r=0.15,p=0.35; RG3:n=89,r=0.12,p=0.24; RG4:n=39,r=0.02,p=0.90). In contrast, RG5, the highest-risk phenotype, showed a strong positive association between SSF and the neurodegeneration index (n=14, r=0.67, p=0.009), indicating that greater sleep fragmentation in this group tracked with more pronounced hippocampal-ventricular structural changes. Conclusion Spectral sleep fragmentation was associated with a composite index of hippocampal atrophy and ventricular enlargement, with the strongest relationship observed in the highest-risk PSG phenotype. These findings suggest that sleep fragmentation may serve as a relevant marker of neurodegenerative vulnerability, particularly among patients at the highest risk of adverse clinical outcomes. Support (if any)

Abstract Introduction Sleep deprivation is a common seizure trigger in adults with epilepsy (AWE), but studies measuring the effects of total sleep duration on seizures are limited. We examined associations between habitual sleep duration (SD)/sleep disturbance and seizure frequency/severity. Methods We retrospectively analyzed seizure and sleep-related patient-reported outcomes in AWE who reported average nightly Sleep Duration (SD) over the past 4 weeks and completed the Liverpool Seizure Severity Scale (LSSS) during the same period. Seizure outcomes were derived from the LSSS 4-week seizure count and the LSSS total score. SD was modeled continuously and categorically (< 7h, 7-9h, >9h). Mixed-effects Poisson regression modeled seizure counts and log-linear mixed models with continuous scales, with adjustments for age, sex, and race. Results We analyzed 453 observations from 370 AWE (mean age 44.5±15.1 yr; 55.1% female; 80.3% Caucasian). Median SD was 6.0 (IQR 5-8), median LSSS 4-week seizure count was 2.0 (IQR 1-6), and median LSSS total severity score was 47.5 (IQR 30-62.5). SD groups were < 7h (n=269, 59.4%), 7-9h (n=147, 32.5%, and >9h (n=37, 8.2%); those sleeping >9h were younger than < 7h and 7-9h sleepers (38.8±13.7 vs 45.1±13.6 and 45.9±16.3; p=0.027). In the adjusted continuous model, SD was not associated with seizure frequency (RR 1.00, 95% CI 0.97-1.04; p=0.96). When analyzing seizure frequency categorically, adjusted mean seizure counts were 8.39 (4.04-17.46) for < 7h, 9.42 (95% CI 4.50-19.70) for 7-9h, and 13.30 (6.14-28.79) for >9h. Compared with < 7h, seizure frequency was similar in 7–9h (RR 1.12, 95% CI 0.91–1.38; p=0.27), but >9h sleepers had higher seizure frequency than < 7h (RR 1.58, 95% CI 1.14–2.20; p=0.006) and 7–9h (RR 1.41, 95% CI 1.03–1.93; p=0.031). LSSS total scores were not significantly associated with SD. Conclusion Relative to normal habitual SD, long sleep was associated with higher seizure frequency, while short sleepers had similar seizure counts. Seizure severity did not differ between SD groups. This is the largest investigation of SD and seizure outcomes in AWE. These results support expanding sleep education from avoiding sleep deprivation to also avoiding over-sleep as a modifiable strategy to improve seizure control. Support (if any) None

Despite the abundance of polysomnography (PSG) data, the limited summary metrics used in existing approaches may not provide the most informative insights for clinical decision-making. We hypothesized that a new data-driven clustering method using the entire multimodal raw PSG data could enable a precise risk stratification approach. We leveraged a new clinical data set to facilitate this data-driven approach and create a novel Foundation Model. We utilized 10,000 PSGs conducted at the Cleveland (1/2012-12/2022) and custom artificial intelligence techniques that incorporate time-series data, to develop a Foundation Model from raw PSG data. We optimized this new model to classify sleep stages, respiratory events, and oxygen desaturations. Resulting embeddings were used to cluster patients into distinct risk groups with a k-means algorithm. Baseline characteristics were compared between risk groups using chi-square tests for categorical variables and Welch’s ANOVA for continuous variables. Means and standard deviations are reported. Optimal stratification of embeddings was achieved with five clusters. Resulting risk groups (RG) had a graded increase in age, male predominance, cardiovascular risk factors, and sleep-disordered breathing(SDB) severity from RG1(n=3,357) to RG5(n=363). Males were more prevalent in RG4(n=1,144; 60.8%) and RG5(66.4%) and least in RG2(n=1,877; 37.6%). Body mass index was lowest in RG2(32.7±9.1kg/m2) and highest in RG5(35.4±10.4kg/m2). RG4 and RG5 were the oldest(58.6±16.1years) and RG2 the youngest(44.0±15.0years). RG1 had intermediate SDB with apnea hypopnea index (AHI:12.4±12.4), total sleep time (TST:328±61.1min), and risk (hypertension:59.8%,diabetes II:32.7%), and lowest cognitive impairment(14.1%). RG2 had the mildest SDB (AHI:5.4±6.4), longest TST (342±71.6min), and lowest cardiovascular risk (hypertension:47.5%,diabetes II:24.7%). RG3(n=2,867) had intermediate SDB and risk. RG4 had more abnormal PSG measures (AHI:22.7±13.2, TST:201±73.9min), more risk (hypertension:75.6%,diabetes II:65.2%), and highest major adverse cardiovascular events (43.6%) but lowest migraine (11.5%). RG5 had the most severe SDB (AHI:37.3±39.0), lowest TST (98.4±89.0min), and high risk (hypertension:77.4%,diabetes II:44.6%). All p-values were < 0.001. We created a Polysomnographic Foundational Model to stratify patients into risk groups characterized by different comorbidities not completely explained by traditional measures. RG4 and RG5 exhibited more severe SDB and unique clinical characteristics that prompt future investigation and warrant more attention from healthcare providers. IBM Discovery Accelerator, AIM Award, NIH 1R21HL170206-01

Apnea hypopnea index (AHI) is the prevailing measure of obstructive sleep apnea (OSA) presence and severity. Despite numerous consequences, AHI calculation measures only frequency of respiratory events. Measures which characterize OSA mechanistically may have better prognostic ability. Sleep apnea-specific hypoxic burden (SAHB) has been associated with incident heart failure (HF) in men. We hypothesized that SAHB predicts incident HF and offers superior predictive performance than traditional measures in this larger cohort of men and women. We retrospectively examined Cleveland Clinic patients (age≥18) who underwent polysomnography 2006-2017, excluding those with HF before or at the time of polysomnogram. Cox proportional hazards models of multiple-imputed datasets were constructed for incident HF by SAHB, AHI, and T90 in separate models adjusted for age, sex, race, body mass index, tobacco use, central sleep apnea, cardiovascular comorbidities, and positive airway pressure (time-varying). An interaction term was created between SAHB and sex. Non-nested partial likelihood ratio tests compared models. The sample included N=26,161 patients: age 51.1±14.1 years, 52% male, 75% White, 17% developed HF during follow-up (6.95 [5.12-8.93] years). A 1-SD increase in SAHB was associated with 5% higher HF incidence (HR=1.05, 95%CI=1.01-1.09); 1-SD increase in AHI was associated with 11% higher HF incidence (HR=1.11, 95%CI=1.07-1.16); 1-SD increase in T90 was associated with 19% higher HF incidence (HR=1.19, 95%CI=1.15-1.22). HF incidence in males with increasing SAHB was lower than in females (interaction p< 0.001). Both the AHI and T90 models fit better than the SAHB model (p=0.006, p< 0.001, respectively). Adding SAHB into the AHI or T90 models did not improve model fit (p=0.86, p=0.34, respectively). In this clinical cohort of men and women, while SAHB, AHI, and T90 were independently associated with incident HF, SAHB did not enhance predictive value. Traditional measures demonstrated superior model fit compared to SAHB in predicting HF incidence, and inclusion of SAHB did not improve model performance. Next steps include adding left ventricular ejection fraction data to assess differences across HF subtypes. More research is needed to enrich the knowledge that guides development of these novel diagnostic options. AASM Foundation Physician Scientist Training Grant, CCF Transformative Research Resource Development Award.

STUDY OBJECTIVES:To examine the unclear, inconsistent role of sleep architectural disruption in atrial fibrillation (AF) development. METHODS:Patients (age ≥ 18 years) who underwent in-laboratory sleep studies at Cleveland Clinic 2000-2015 were examined (follow-up: 7.8 ± 3.5 years). Primary predictors were arousal index and total sleep time. Secondary predictors included sleep efficiency, wakefulness after sleep onset, sleep and REM latency, and percentage of each sleep stage. Predictors were fit to Cox proportional hazard models predicting time from sleep study to AF by diagnosis code. Covariates included demographics, anthropometrics, tobacco use, sleepiness, apnea-hypopnea index, sleep apnea-specific hypoxic impact, cardiovascular risk factors and disease, mood disorders, medications, and positive airway pressure. RESULTS:In our cohort (n = 27 232, age: 49.4 ± 14.5 years, 43.7% male, 73.9% white), 2077 (7.6%) developed incident AF. Arousal index was not associated with AF incidence. For every hour of decreased total sleep time, AF incidence increased 8% (HR = 1.08, 95% CI = 1.04 to 1.11). For every 10-unit decrease in sleep efficiency, AF incidence increased 6% (HR = 1.06, 95% CI = 1.04 to 1.09). For every hour of increased wakefulness after sleep onset, AF incidence increased 11% (HR = 1.11, 95% CI = 1.05 to 1.16). For every 10-unit increase in percent N1, AF incidence increased 6% (HR = 1.06, 95% CI = 1.01 to 1.10). CONCLUSIONS:Less sleep time and greater sleep disruption were associated with increased incident AF in this large clinical cohort. These results suggest that sleep macro-architecture can influence AF development. Mechanistic and prospective studies are needed to verify whether sleep disruption is a novel target for AF prevention.

Patients with CNS disorders of hypersomnolence (CDH) are at increased risk of cardiovascular and other health comorbidities that require frequent monitoring, particularly in the setting of controlled substance use. A 2019-2021 market analysis of patients treated for CDH at Cleveland Clinic found only 27% of patient encounters were clinic visits; others, phone calls or electronic refill requests. Follow-up and safety assessments were not standardized. We piloted a Hypersomnia CarePath that standardizes access and quality and safety monitoring for a growing CDH population aligned with organizational and state regulations. Here, we report pilot phase results of adherence and safety assessments. CDH patients on stable pharmacotherapy were enrolled in a follow-up CarePath based on pharmacotherapy type (3-month visits for traditional stimulants; 6-month visits for all other medications). All required one in-person and one physician visit (vs. midlevel provider) per year. Visit compliance was defined as the percent of required visits completed. Annual EKG and UTOX were performed in the 3-month pathway and by physician request for others. A total of 121 patients were enrolled since July 2022, mean age 42.8 (SD 15.3) yr, 79.3% female, 78.3% Caucasian, 8.3% Black, 5.0% Asian, 8.3% Other. Of these, 81 (67%) were in Q3 and 40 (33%) in Q6 pathways. Age, gender, and race were similar between groups. Visit compliance was 91% overall, 97% for Q3 and 74% Q6. A total of 67 EKGs and 40 UTOX assessments were completed. Abnormal and unexpected UTOX results were found in 4 (10%) cases. Abnormal EKGs were observed in 24 (35.8%) cases including 6 (46.1%) males and 18 (33.3%) females. Patient and caregiver acceptance and satisfaction were highly favorable. Our novel Hypersomnia CarePath for CDH established standardized follow-up and safety assessments with high visit adherence and patient and caregiver satisfaction. Over one third of EKGs at enrollment (males more than females) and 10% of UTOX studies at the time of enrollment were abnormal. These findings support the need for routine quality and safety monitoring given the known risk of cardiovascular and psychiatric comorbidities in CDH.

Autonomic dysfunction (AD) is implicated in sleep apnea and atrial fibrillation (AF), however age- and sex-specific interactions are unclear. We hypothesize age- and sex-specific differences in sleep apnea-specific heart rate response (ΔHR), an AD biomarker, on AF susceptibility. Cleveland Clinic patients (age≥18) who underwent polysomnography 1/2/2000-12/30/2017 were retrospectively examined. Cox proportional hazards models were constructed for time from sleep study to AF by diagnosis code. ΔHR (respiratory event-related electrocardiogram-based heart rate difference; signal processing via Python®) was adjusted for demographics, body mass index (BMI), tobacco use, cardiopulmonary disease, anti-arrhythmic medications, positive airway pressure, and minimum heart rate during events. Statistical analyses were conducted using R and SAS. The sample included n=23,419 patients (age 51[41-61] years, 51% male, 74% White, BMI 34[29-40] kg/m2, apnea hypopnea index 18.3[7.5-42.6]) over7±3-years. High vs. mid-range ΔHR was associated with 22% increased AF incidence (HR=1.22, 95%CI=1.09-1.35); low vs. mid-range was not. An interaction of ΔHR by sex (p=0.031) for low vs. mid-range ΔHR showed AF incidence was lower in males than females (HR=0.77, 95%CI=0.61-0.97). In sex-stratified models, high vs. mid-range ΔHR in females showed the highest AF incidence (HR=1.27, 95%CI=1.06-1.52), followed by low vs. mid-range in females (HR=1.22, 95%CI=1.03-1.45), high vs. mid-range in males (HR=1.18, 95%CI=1.03-1.35), and low vs. mid-range in males though non-significant (HR=0.93, 95%CI=0.79-1.09). An interaction of ΔHR by age (p< 0.001) showed older patients (age>51) had 51% increased AF incidence relative to younger patients for high vs. mid-range ΔHR (HR=1.51, 95%CI=1.18-1.94); among younger patients, low vs. mid-range ΔHR had 42% increased AF incidence (HR=1.42, 95%CI=1.08-1.86). A ΔHR by race interaction was non-significant (p=0.905). A polysomnography-derived sleep apnea-specific AD biomarker was associated with AF with important age- and sex-specific interactions. There was a U-shaped trend in females and older patients, a positive linear trend in males, and a negative linear trend in younger patients. Younger females with low ΔHR had highest AF incidence. Findings suggest sleep apnea-specific AD plays a role in AF development, highlighting the need for tailored management of sleep apnea-related AF risk based on age and sex. AASM Foundation Physician Scientist Training Grant, Cleveland Clinic Neuroscience Transformative Research Resource Development Award

Central Disorders of Hypersomnolence (CDH) are classified primarily by polysomnography-multiple sleep latency test (PSG-MSLT). Lack of biomarkers and variable MSLT results in disorders other than narcolepsy type 1 (NT1) contribute to under- and mis-diagnosis. Limited studies explore PSG characteristics that may differentiate CDH subtypes using large datasets. This retrospective cohort included 1,330 patients who underwent PSG-MSLT for hypersomnolence (January 2003-August 2024) at Cleveland Clinic. Cases were classified as NT1, narcolepsy type 2 (NT2), idiopathic hypersomnia (IH) and undifferentiated hypersomnia (UH; not meeting CDH criteria) based on physician clinical diagnosis. 557 patients with disorders associated with hypersomnia (i.e. OSA) were excluded. Demographic and PSG characteristics were compared with ANOVA, Kruskal-Wallis, and Pearson chi-square tests. Of 773 patients (33.7±14.0 yr, 79.4% female), 72(9.3%) had NT1,121(15.7%) NT2, 296(38.3%) IH, and 284(36.7%) UH. While groups did not differ in age or gender, more Caucasians had IH and UH than NT1(78.6, 80.8 vs. 62.5%, p=0.001). Epworth Sleepiness Scale scores were lower in UH than CDH groups (12.6±5.0 vs NT1-15.9±5.6,NT2-16.1±5.1,IH-14.2±5.2, p< 0.001). Self-reported sleep time was longer for UH than NT2 without other group differences (UH-9.1±2.1,NT2-8.4±1.7,NT1-8.6±2.1,IH-8.8±1.9 hours, p=0.003). PSG sleep onset REM periods (SOREMPs) were more common in NT1 and NT2 than IH and UH (NT1-27.8%,NT2-9.1%,IH-1.01%,UH-1.06%, p< 0.001). NT1 and NT2 had shorter REM latency than IH and UH(NT1-73.5[7.0,123.0],NT2-74.0[53.0,112.0],IH-103.0[71.0,161.5],UH-105.0[74.5,183.0] min, p< 0.001). NT1 and UH showed more wakefulness after sleep onset than NT2 and IH (NT1-43.5[26.0,97.0],UH-47.5[25.5, 85.0],NT2-32.5[17.3,61.5],IH-30.5[17.5, 62.5] min, p< 0.001). Sleep latency was shorter in CDH than UH (NT1-10.8[4.5,26.0],NT2-14.0[6.0,24.5],IH-17.5[7.8,30.5],UH-24.5[12.5,39.0] min, p< 0.001), and sleep efficiency higher in NT2 and IH than UH (NT2-87.4±7.3,IH-86.3±8.8,UH-82.0±10.0, p< 0.001). NT1 had greater, though non-significant, arousal index, stage shifts, and N1 percentage, along with lower sleep efficiency than NT2 and IH. No significant differences were observed between NT2 and IH in sleep latency, sleep efficiency, arousal index, WASO, stage shifts, or stage percentages. We found significant differences in PSG variables between CDH that confirm and extend prior observations. Minimal differences between NT2 and IH support a common pathophysiology. Recognizing the larger group of UH with distinct PSG features from CDH is important in clinical practice.

Central Disorders of Hypersomnolence (CDH) are classified primarily by polysomnography-multiple sleep latency test (PSG-MSLT). Lack of biomarkers, variable MSLT results in disorders other than narcolepsy type 1 (NT1) and inaccessibility of 24-hr PSG and actigraphy in most U.S. sleep centers contribute to misdiagnoses. We compared CDH diagnoses by sleep expert clinical phenotyping and ICSD-3 criteria to estimate the magnitude of potential misdiagnosis in a U.S. quaternary care sleep center. This retrospective study included patients evaluated for hypersomnolence at Cleveland Clinic (January 2003 - August 2024). Patient with a clinical diagnosis of NT1, NT2, and IH based on sleep physician clinical history were included. ICSD-3 diagnoses were based on PSG/MSLT SOREMPs and MSLT mean sleep latency (MSL): narcolepsy: MSL< 8 min + >2SOREMPs; IH: MSL< 8 min + < 2SOREMPs. Those not meeting ICSD-3 criteria were considered undiagnosed. ICSD-3 diagnoses were analyzed between clinical groups using ANOVA, Kruskal-Wallis, and Pearson chi-square tests, followed by pairwise comparisons. Of 489 patients (33.6±13.6 yrs, 78.7% female), 72(14.7%) had NT1, 121(24.7%) NT2, and 296(60.5%) IH. More NT1 and NT2 patients had >2 SOREMPs than IH (NT1-81.9%, NT2-71.1%, IH-12.4%, p< 0.001). MSL was longer in IH than narcolepsy (IH-8.1±4., NT1-4.9±3.8, NT2-5.4±3.2 min, p< 0.001). Percentage agreement between ICSD-3 and clinical diagnoses was 100% for NT1, but lower for NT2 and IH (NT2-61.2%,IH-58.1%, p< 0.001). Among clinical NT1, 26.4% did not meet PSG-MSLT criteria. Among clinical NT2, 38.9% had different ICSD-3 diagnoses: 24% reclassified as IH and 14.9% were undiagnosed. Among clinical IH, 41.9% had different ICSD-3 diagnosis: 0.6% reclassified as NT2 and 41.2% were undiagnosed. Overall, 28.6% of CDH patients diagnosed based on clinical criteria were not diagnosed as CDH by ICSD-3 criteria due to lack of sleep laboratory findings. This study showed diagnostic discrepancies between clinical phenotyping and ICSD-3 criteria for NT2 and IH, while NT1 diagnoses aligned. Over a quarter of clinically diagnosed patients remained undiagnosed by ICSD-3 criteria. These findings illustrate limitations of current diagnostic criteria and the need for to improve diagnostic accuracy in the evaluation of CDH.

REM Sleep without Atonia (RSWA) is the polysomnographic biomarker of REM Sleep Behavior Disorder (RBD). RSWA is associated with serotonergic (SSRIs/SNRIs) antidepressants (ADs). The effects of other ADs and depression on RSWA are unknown. We analyzed the relationship between AD use and RSWA by depression diagnosis status aiming to determine whether RSWA is influenced by the diagnosis of depression, serotonergic agents, or ADs in general. Polysomnograms at Cleveland Clinic (September 2018-October 2024) with >10% REM sleep and RSWA scoring by chin and flexor digitorum superficialis signals were included. Cases were categorized by depression status (yes=923; no=564), then stratified by AD use including any AD (SSRIs-386, SNRIs-181, bupropion-145, trazodone-117, tricyclic AD-60, mirtazapine-29), serotonergic (SSRI/SNRI), non-serotonergic (bupropion-n=145, mirtazapine-n=), and AD-free (n=653), and finally by gender. RSWA was quantified by AASM V2.6 criteria. Wilcoxon rank sum tests compared RSWA% between depression and medication status groups. Results are shown as median (IQR). A total of 1,487 were included: mean age 52.9±16.7, 47.2% female. Depression (vs no depression) was predominantly present in females (56.7% vs. 31.7%, p< 0.001) but had similar RSWA% (3.9[0.5, 14.6] vs 4.2[0.0, 18.0], p=0.72). RSWA% was 6.3[1.1,19.2], 6.7[1.3, 18.3], 0.7[0.0, 6.3], and 2.9[0.0, 11.4] in any AD, serotonergic, non-serotonergic, and drug free AD groups, respectively (p< 0.001). Serotonergic and AD-free groups without depression had higher RSWA% than those with depression (11.8[3.4, 31.5]/3.2[0.0, 14] vs 7.4[1.5, 20]/2.5[0.0, 9.4], p=0.04/p=0.038). For serotonergic ADs, this relationship was present only in women. RSWA did not differ in the any AD and non-serotonergic group by depression status (p=0.56). RSWA is influenced by AD use rather than depression itself. Serotonergic agents are associated with higher RSWA% than non-serotonergic ADs, particularly in patients without depression. A possible explanation is that RSWA may be mediated by serotonin, as patients without depression, and thus without serotonergic depletion, appear more sensitive to the effects of serotonergic ADs. Our findings support the use of non-serotonergic agents over serotonergic agents, if clinically appropriate in populations at risk for RBD and ⍺-synucleinopathies. This work has important implications on quantification of RSWA for the diagnosis of RBD.

Traditional sleep measures are not consistently associated with incident cardiovascular events in observational studies. We hypothesized deep learning approaches could uniquely identify distinct patient groups with varying cardiovascular disease (CVD) incidence, independent of traditional measures. We used artificial intelligence to analyze N=10,000 polysomnograms completed at Cleveland Clinic (1/2012-12/2022) enriched with underrepresented populations. We created a time-series foundational model guided by sleep macro-architecture and respiratory annotations, generating embeddings to cluster patients into risk groups using k-means. Follow-up time was from polysomnogram until death or last follow-up. Propensity scores were estimated using multinomial logistic regression on age, sex, body mass index (BMI), and years of available data, then inverse probability weighted, stabilized, and trimmed at 1st and 99th percentiles to minimize bias introduced by extreme values. Cox proportional hazards regression examined risk groups as predictors adjusted for age, sex, BMI, and apnea hypopnea index (AHI) with comorbidities included per outcome. Patients with baseline disease were excluded. The cohort [age 50.4±24.7 years, 50% male, 44% White, 34% Black, 5.3% Asian, 14.8% multiracial] had follow-up of 4.4[2.0-7.5] years. A 5-cluster solution provided the best stratification: Risk Group 1 (RG1) was reference. RG5 (highest AHI and arousal index, lowest mean and minimum SaO2 and total sleep time(TST)) had the highest CVD incidence (major adverse cardiovascular events(MACE): HR=1.60, 95%CI=1.13-2.28, myocardial infarction: HR=1.75, 95%CI=1.07-2.88, atrial fibrillation: HR=1.83, 95%CI=1.15-2.93) and all-cause mortality (HR=2.76, 95%CI=1.98-3.84). RG4 (highest %time SaO2< 90%) had elevated incidence to a lesser degree than RG5. RG3 (intermediate) had the highest stroke (HR=1.38, 95%CI=1.13-1.69) and ischemic heart disease (HR=2.01, 95%CI=1.37-2.94) incidence, second highest MACE incidence (HR=1.34, 95%CI=1.12-1.60) and lowest all-cause mortality (HR=1.47, 95%CI=1.14-1.90). RG2 (lowest AHI and %time SaO2< 90%, highest mean and minimum SaO2 and TST) had the lowest CVD incidence but not all-cause mortality (HR=1.47, 95%CI=1.14-1.90). We created a large language model of raw polysomnogram data which identified groups that differed in CVD incidence after propensity score analysis. Risk groups were associated with adverse cardiovascular outcomes, thus supporting enhanced utility of a novel deep learning model over traditional approaches for CVD risk stratification. Future directions include validation with external data sources. IBM Discovery Accelerator, AIM Award

Abstract Introduction Continuous positive airway pressure (CPAP) adherence represents a pervasive, multifactorial challenge. Hypoglossal nerve stimulation (HNS) is an alternative treatment for moderate-severe obstructive sleep apnea (OSA) for those who are CPAP-intolerant. We hypothesized differential predictors of therapy adherence and greater adherence with HNS versus CPAP. Methods We compared adherence and its predictors in patients from the Cleveland Clinic HNS registry (cloud-based data monitoring) versus those on CPAP from the HomePAP study at the 1st and 3rd month timepoints. Linear models assessed age, sex and body mass index (BMI) as 1- and 3-month HNS and CPAP adherence predictors. Propensity score matching (1:1) without replacement and t–tests were used to compare adherence across the groups. Propensity scores were estimated with support vector machine using age, sex, BMI and Epworth Sleepiness Scale (ESS) at baseline. Results We analyzed data from n=63 HNS patients (age:57.9±10.0 years, 60% male, body mass index (BMI):28.8 ±3.3 kg/m2), ESS score:8.83±4.69) and n=126 CPAP patients from the HomePAP study (age:49.8±11.9 years, 63% male, BMI:39.1±9.43 kg/m2, ESS score:14.3±3.8). At 1-month, female sex was associated with increased HNS therapy usage (1.20, SE:0.47, p< 0.01); a finding not observed at 3-months. Propensity score-matched analysis included 28 HNS patients (age: 53.6± 9.9 years, 67% male, BMI:30.7± 2.6 kg/m2, ESS:11.4± 3.7) and 28 CPAP patients (age:58.5± 9.6 years, 85% male, BMI:31.2± 4.0 kg/m2, ESS:12.2± 3.6). At 1-month (HNS n=28, CPAP n=27), usage was 3.90±1.0 hours for CPAP and 6.97±1.09 hours for HNS (p< 0.01). At 3-months (HNS N=27, CPAP N=24), usage was 4.38±1.87 (SD) hours for CPAP and 6.44±1.96 hours for HNS (p< 0.01). Conclusion In this comparative analysis, we identify female sex as a factor associated with early adherence to HNS, findings not observed with CPAP. Using a rigorous approach to address baseline differences, we identify a substantial increased level of adherence in HNS versus CPAP therapy of approximately 2-3 hours which persisted until 3 months. Future investigation should focus on assessing HNS versus CPAP adherence-related improvement in clinical outcomes. Support (if any) NIH grant R21HL170206.