Sleep spindle density increases in patients with severe obstructive sleep apnea after 6 months of treatment: a longitudinal study Authors: Ainhoa Álvarez-Ruiz-Larrinaga, Jorge Ullate, Angela L. D’Rozario, Jong-Won Kim, Anna E. Mullins, Ron R. Grunstein, Carla Pía, Carlos Egea-Santaolalla, David Gozal, Joaquín Durán-Cantolla, Maria Comas Published: 2026 ABSTRACT Obstructive sleep apnea (OSA) disrupts sleep structure and alters NREM sleep microarchitecture, including reductions in sleep spindle activity. Given the central role of sleep spindles in sleep continuity and neurophysiological processing, their reduction in patients with OSA suggests a potentially reversible sleep microstructural disruption. We evaluated sleep spindles across four EEG derivations (F3, F4, C3, and C4) during NREM sleep in patients with suspected OSA, assessed before and after 6 months of routine clinical practice. The 6-month polysomnography was performed with participants using their prescribed treatment during the recording. A total of 216 patients (mean age 52.0 ± 11.5 years; 32.9% women) completed the study, including 66 with no or mild OSA (AHI < 15 events/hour), 50 with moderate OSA (AHI 15–<30), and 100 with severe OSA (AHI ≥ 30). Pre- and post-treatment sleep study data were available for analysis. At baseline, patients with severe OSA exhibited significantly lower sleep spindle density (SSD) compared with patients with moderate OSA and those with mild or no OSA. After six months of routine clinical management, there was a significant increase in SSD, but only in the severe OSA group (e.g., N2 C4–M1: +0.47 events/min, 95% CI 0.30 to 0.65), with similar patterns observed across central and selected frontal derivations. After correction for multiple comparisons, subtype analyses supported a significant increase only for fast spindles at a central derivation. These findings suggest that severe OSA is associated with spindle deficits and that routine clinical management ameliorates altered sleep microarchitecture. SSD may serve as a sensitive marker of severity-related changes in sleep microstructure and could complement conventional sleep staging in clinical cohorts.Supplementary InformationThe online version contains supplementary material available at 10.1038/s41598-026-50660-6. Introduction Obstructive sleep apnea (OSA) is a highly prevalent sleep disorder worldwide and affects nearly a billion people1. In addition to its numerous comorbidities, OSA has been significantly associated with cognitive impairments2–4. These impairments are thought to be driven, at least in part, by intermittent oxyhemoglobin desaturation resulting from upper airway obstruction and by the disruptive effects of sleep fragmentation5. Nonetheless, the precise mechanisms underlying cognitive impairment in OSA remain incompletely understood6,7. Importantly, OSA also produces marked disruptions in sleep continuity and architecture, yet routine clinical metrics may not fully capture associated changes in sleep neurophysiology8,9. Overnight polysomnography (PSG) is recognized as the gold standard diagnostic method for identifying sleep-related breathing disorders because it provides comprehensive, simultaneous recordings of sleep architecture, respiratory events, and associated arousals10. Although PSG generates extensive neurophysiological information recorded on the electroencephalogram (EEG), much of these physiological data remain underexplored, primarily because clinical practice still relies mainly on conventional brain activity analyses, such as standard sleep staging and arousal scoring. Quantitative assessments of sleep microarchitecture derived from EEG may provide more sensitive indices of sleep disruption in individuals with OSA11,12. Among these quantitative EEG (qEEG) markers, sleep spindles have attracted particular interest as indicators of NREM sleep microarchitecture and sleep stability13,14. Sleep spindles are EEG oscillations in the 11–16 Hz frequency range, which wax and wane in amplitude, last at least 0.5 s and occur predominantly during non-rapid eye movement (NREM) sleep stages N2 and N315. Spindle oscillations are typically classified as slow (11–13 Hz, with an anterior cortical predominance) or fast (13–16 Hz, with a posterior cortical predominance)16. Sleep spindles originate from repetitive bursts of action potentials in GABAergic neurons of the thalamic reticular nucleus, which generate rhythmic inhibitory postsynaptic potentials that propagate through thalamocortical pathways13. They contribute to sleep continuity by protecting against external disturbances and are involved in synaptic plasticity and neural network reorganization17–19. Accordingly, spindle metrics have been proposed as sensitive markers of sleep microstructure and its disruption in clinical populations13,14,19,20, and emerging evidence suggests that they may also reflect sleep stability in patients with OSA21,22. Because OSA severity is also associated with greater sleep fragmentation and altered sleep–stage distribution, spindle measures may provide complementary information beyond conventional staging23. To date, however, no study has directly investigated whether routine clinical management of patients with suspected OSA differentially affects sleep spindle activity, as quantified by automated detection methods applied to multiple EEG derivations. Several studies have explored whether decreased spindle activity can be restored following OSA therapy, with evidence indicating that continuous positive airway pressure (CPAP) treatment may be associated with improvements in sleep microarchitecture compared with pre-treatment values24–26 and in some studies relative to control participants27. However, these studies are limited by small sample sizes and a predominance of male participants in some of the cohorts26,27. Furthermore, in several of these studies spindle activity was evaluated either during CPAP titration with PSG or in split-night studies26–28. Detection methods also vary widely, from manual scoring to automated algorithms applied to different EEG derivations. Together, these factors limit comparability across studies and leave uncertainty about how spindle activity changes over time in routine clinical cohorts undergoing standard management. In the present study, we aimed to evaluate changes in sleep microarchitecture, focusing on sleep spindle density, from baseline to a six months follow-up of standard clinical management in a clinical referral cohort of patients with suspected OSA. We examined whether these changes differed according to baseline OSA severity and described concurrent changes in PSG-derived sleep macrostructure to contextualize spindle findings. We hypothesized that spindle density would be lowest in severe OSA at baseline and would show the largest longitudinal change in this group over follow-up. Methods Participants This study is a secondary analysis of the APNiA prospective clinical referral cohort conducted at the Sleep Unit of Araba University Hospital (Vitoria-Gasteiz, Basque Country, Spain) between September 2015 and March 201829. The APNiA study included patients aged ≥ 18 years who had been referred for evaluation of suspected OSA because of snoring, witnessed apneas, excessive daytime sleepiness or non-restorative morning tiredness. Exclusion criteria were previous evaluation of OSA, presence of another sleep disorder, severe and unstable heart disease, or refusal to participate in the study. For this post hoc analysis, additional exclusions included neurological disorders known to affect sleep spindle activity, such as neurodegenerative diseases, and poor EEG recording quality precluding spindle measurement. As this was a post hoc secondary analysis, no a priori sample size calculation was performed. The analytic sample size was determined by the number of APNiA participants with complete baseline and 6-month PSG recordings and adequate EEG signal quality for spindle detection, as documented in Fig. 1. Fig. 1Study flow diagram. Of 352 patients assessed for eligibility, 77 were excluded (15 did not meet inclusion criteria, 31 had baseline PSG not available and 31 had an ineligible baseline PSG), leaving 275 patients with an eligible baseline PSG. After standard clinical management and 6-month follow-up, 216 patients completed a second PSG. A further 59 were excluded at this stage (41 were lost to follow-up, 4 had PSG at 6-months not available and 14 had an ineligible PSG at 6 months). Study flow diagram. Of 352 patients assessed for eligibility, 77 were excluded (15 did not meet inclusion criteria, 31 had baseline PSG not available and 31 had an ineligible baseline PSG), leaving 275 patients with an eligible baseline PSG. After standard clinical management and 6-month follow-up, 216 patients completed a second PSG. A further 59 were excluded at this stage (41 were lost to follow-up, 4 had PSG at 6-months not available and 14 had an ineligible PSG at 6 months). All patients provided written informed consent. The study protocol was approved by the Clinical Research Ethics Committee (CEI) of Araba University Hospital in Vitoria-Gasteiz, Spain (BABTI-15-EC-01-APNIA, File:2015-026). The sleep spindle sub-study received additional approval from the Clinical Research Ethics Committee of the Araba Health Organization, Vitoria-Gasteiz, Spain, in 2021 (CEI OSI ARABA, File: 2021-027). All methods were performed in accordance with the relevant guidelines and regulations. Procedures Questionnaires At baseline, demographic and anthropometric characteristics were recorded. Information on comorbidities, medication use and relevant medical history was obtained from a structured, study-specific questionnaire. Standard clinical treatment In the APNiA study cohort, the treatment of OSA prescribed and in use at the time of the 6-month follow-up PSG reflected standard clinical practice, in which OSA management is individualized according to disease severity, symptom burden, and the presence of comorbidities30,31. In this context, mild OSA is typically managed with conservative interventions such as lifestyle modifications, positional therapy, or oral appliances, with CPAP reserved for symptomatic cases. Moderate OSA is commonly treated with CPAP as first-line therapy, whereas in severe OSA, CPAP constitutes the standard of care, with surgical options or advanced therapies considered when appropriate. Polysomnographic recordings Overnight polysomnography (PSG) was performed at baseline and at the 6-month follow-up, in accordance with the recommendations of the American Academy of Sleep Medicine (AASM)32. The 6-month follow-up PSG was performed with participants using their prescribed OSA treatment during the recording. As part of the APNiA research protocol, the 6-month in-lab PSG was scheduled for all enrolled participants, irrespective of baseline OSA severity, symptom evolution or type of treatment received. EEG activity was recorded following the international 10–20 system32 from six derivations, each referenced to the contralateral mastoid (F3–M2, F4–M1, C3–M2, C4–M1, O1–M2, and O2–M1). Additionally, recordings included two electrooculography channels, submental electromyography, electrocardiography, airflow pressure by nasal transducer, thermistor, thoraco-abdominal bands, blood oxygen saturation and body position. PSGs were acquired using Alice-6 (Philips Respironics, Murrysville, PA, USA), Compumedics (Melbourne, Australia) and SOMNOscreen plus (Somnomedics, Randersacker, Germany). Sleep scoring was performed manually using the standard criteria established by the AASM32, by two experienced scorers, including a trained sleep technician and a physician specialized in clinical neurophysiology.The Apnea–Hypopnea Index (AHI) was defined as the number of obstructive apneas and hypopneas, lasting at least 10 s, per hour of sleep. Obstructive apneas were defined as a ≥ 90% reduction in thermal signal amplitude, and hypopneas as a ≥ 30% decrease in the nasal pressure signal accompanied by ≥ 3% oxygen desaturation or an associated arousal31. OSA severity groups were defined as: mild or no OSA (AHI < 15), moderate OSA (AHI 15–<30) and severe OSA (AHI ≥ 30). All PSG assessments were performed using identical standardized protocols at baseline and follow-up and were applied uniformly across baseline OSA severity groups. Although recordings were obtained using different PSG systems, all studies were scored according to AASM criteria and subsequently exported to a common format for downstream processing, ensuring comparability of measurements across participants and severity groups. Hypoxic Burden (HB) was defined as the total area under the respiratory event–related desaturation curve. This metric was calculated according to the methodology described by Azarbarzin et al.33 by dividing the sum of all individual desaturation areas by total sleep time, yielding a value expressed as %·min/h. The HB index specifically quantifies the overall hypoxemic burden associated with respiratory events. The algorithm was implemented using a diagnostic sleep system (Alice 6 LDxS; Philips Respironics) and Sleepware software version 4.0.1.0 (Koninklijke Philips N.V.). Only patients who had undergone studies using this system, both before and after routine clinical practice, were included in the analysis. Hypoxic burden was available in a subset of 52 participants and analyses involving this metric should therefore be interpreted as exploratory. Sleep spindle detection We included all patients who underwent a baseline polysomnography (PSG) study and also a follow-up PSG at 6 months, provided they met the following conditions: time in bed > 6 h, total sleep time > 4 h and adequate electroencephalographic (EEG) signal quality. The study flow is summarised in Fig. 1. All polysomnographic recordings were exported to the standardized European Data Format (EDF) and EEG recordings were processed using a validated artefact-detection method34. Artefact-free all-night EEG data at F3-M2, F4-M1, C3–M2 and C4–M1 were subjected to an automated spindle detection algorithm validated in adults with OSA25 and applied identically across participants, irrespective of baseline OSA severity. Briefly, the algorithm applies band-pass filtering in the sigma range (11–16 Hz) and a Hilbert transform function to derive the amplitude envelope, and identifies spindle events using a relative amplitude threshold (median + 1 SD) and a duration criterion of 0.5–3.0 s. For each recording, we derived indices of sleep spindle events per minute of NREM sleep (N2 + N3) and of N2 alone for overall spindle density (11–16 Hz, events per min), slow spindle density (11 ≤ f ≤ 13 Hz; f: frequency), and fast spindle density (13 < f ≤ 16 Hz). Outcomes and variables The primary outcome was the within-participant change in overall sleep spindle density (events per minute) from baseline to 6 months, quantified for stage N2 and NREM sleep (N2 + N3), and additionally by spindle subtype (slow and fast). Secondary outcomes included changes in sleep macrostructure derived from PSG. Baseline OSA severity, defined by the Apnea–Hypopnea Index (AHI), was the main exposure of interest and was operationalized both continuously and categorically (mild/no OSA: AHI < 15; moderate: 15–<30; severe: ≥30). Prespecified covariates included age, sex and body mass index (BMI). In exploratory analyses, baseline AHI, baseline spindle density, age, sex and BMI were evaluated as predictors of 6-month changes in spindle density. Data analysis Quantitative variables were summarized as means (SD) and categorical variables as absolute frequencies (N) and percentages. Baseline comparisons across OSA severity groups were performed using one-way ANOVA or the Kruskal–Wallis test for continuous variables and Fisher’s exact test for categorical variables, as appropriate. Primary and secondary longitudinal outcomes were analysed using linear mixed-effects models (LMMs), fitted separately for each outcome. Each model included fixed effects for time (0 = baseline, 1 = 6 months), baseline OSA severity group (No/Mild, Moderate, Severe), the time × group interaction (primary test of interest), and prespecified baseline covariates age (years), sex (female vs. male), and BMI (kg/m²). A random intercept per participant was included to account for within-subject correlation across the two measurement occasions. Models were fitted using restricted maximum likelihood (REML) as implemented in the lme4 package in R. Given the number of outcomes examined, p-values for the time × group interaction were corrected for multiple testing using the Benjamini–Hochberg false discovery rate (FDR) procedure, applied separately within prespecified outcome families: (i) overall spindle density during NREM sleep (N2 + N3) and stage N2 (Table 2), (ii) slow and fast spindle density subtypes during stage N2 (Table 3), and (iii) PSG-derived macrostructure variables (Table 4). FDR-adjusted q-values < 0.05 were considered statistically significant and are reported alongside uncorrected p-values. For outcomes with a statistically significant time × group interaction after FDR correction, pairwise contrasts comparing changes between severity groups were computed using estimated marginal means from the LMMs (emmeans), with Šidák adjustment of difference-in-differences contrasts. No imputation of missing data was performed. Loss to follow-up was documented in Fig. 1. All tests were two-sided with α = 0.05. Analyses were performed in R version 4.5.1 using lme4 and emmeans; FDR correction used p.adjust(method = “BH”). Table 1Baseline demographic, anthropometric and clinical characteristics by OSA severity group. N OVERALLNO OSA-MILD OSA(AHI < 15)MODERATE OSA(AHI 15–<30)SEVERE OSA(AHI ≥ 30)P-value2166650100Male sex, n (%)145 (67.1)30 (45.5)36 (72.0)79 (79.0) < 0.001 1 Age (years)52.0 (11.5)47.5 (10.6)51.5 (11.1)55.2 (11.3) < 0.001 2 BMI (kg/m2)28.7 (5.0)26.7 (4.5)28.3 (4.7)30.3 (5.0) < 0.001 2 Neck circumference (cm)39.7 (4.2)37.1 (3.8)39.0 (3.3)41.7 (3.9) < 0.001 2 Waist circumference (cm)99.5 (12.9)92.3 (11.8)99.2 (11.6)104.5 (12.0) < 0.001 2 Systolic blood pressure (mmHg)125.1 (16.3)120.2 (16.1)120.5 (10.8)130.6 (17.0) < 0.001 2 Diastolic blood pressure (mmHg)78.1 (8.9)75.0 (9.4)76.7 (6.9)80.8 (8.6) < 0.001 2 Comorbidities, n (%)85 (39.4%)31 (47.0%)21 (42.0%)33 (33.0%)0.1811Cardiovascular75 (59.5%)17 (50.0%)11 (44.0%)47 (70.2%) 0.031 Dermatological3 (2.4%)1 (2.9%)1 (4.0%)1 (1.5%)0.769Gastrointestinal / Hepatic22 (17.5%)5 (14.7%)5 (20.0%)12 (17.9%)0.862Nephro-urological13 (10.3%)3 (8.8%)3 (12.0%)7 (10.5%)0.929Traumatological / Rheumatological23 (18.3%)3 (8.8%)3 (12.0%)17 (25.4%)0.099Hematological3 (2.4%)1 (2.9%)1 (4.0%)1 (1.5%)0.769Neurological10 (7.9%)4 (11.8%)2 (8.0%)4 (6.0%)0.559Ophthalmological1 (0.8%)1 (2.9%)0 (0.0%)0 (0.0%)0.468OSA severity was classified according to the Apnea–Hypopnea Index (AHI) from the baseline overnight polysomnography (PSG). Continuous variables are presented as mean (SD) and categorical variables as n (%). The P-value column compares the three severity groups using the test indicated by the superscript: ¹ Fisher’s exact test; ²ANOVA. All tests were two-sided (α = 0.05). Bold p-values indicate p < 0.05. Abbreviations: OSA, obstructive sleep apnea; AHI, Apnea–Hypopnea Index; PSG, polysomnography; BMI, body mass index. Table 1Baseline demographic, anthropometric and clinical characteristics by OSA severity group. N OVERALLNO OSA-MILD OSA(AHI < 15)MODERATE OSA(AHI 15–<30)SEVERE OSA(AHI ≥ 30)P-value2166650100Male sex, n (%)145 (67.1)30 (45.5)36 (72.0)79 (79.0) < 0.001 1 Age (years)52.0 (11.5)47.5 (10.6)51.5 (11.1)55.2 (11.3) < 0.001 2 BMI (kg/m2)28.7 (5.0)26.7 (4.5)28.3 (4.7)30.3 (5.0) < 0.001 2 Neck circumference (cm)39.7 (4.2)37.1 (3.8)39.0 (3.3)41.7 (3.9) < 0.001 2 Waist circumference (cm)99.5 (12.9)92.3 (11.8)99.2 (11.6)104.5 (12.0) < 0.001 2 Systolic blood pressure (mmHg)125.1 (16.3)120.2 (16.1)120.5 (10.8)130.6 (17.0) < 0.001 2 Diastolic blood pressure (mmHg)78.1 (8.9)75.0 (9.4)76.7 (6.9)80.8 (8.6) < 0.001 2 Comorbidities, n (%)85 (39.4%)31 (47.0%)21 (42.0%)33 (33.0%)0.1811Cardiovascular75 (59.5%)17 (50.0%)11 (44.0%)47 (70.2%) 0.031 Dermatological3 (2.4%)1 (2.9%)1 (4.0%)1 (1.5%)0.769Gastrointestinal / Hepatic22 (17.5%)5 (14.7%)5 (20.0%)12 (17.9%)0.862Nephro-urological13 (10.3%)3 (8.8%)3 (12.0%)7 (10.5%)0.929Traumatological / Rheumatological23 (18.3%)3 (8.8%)3 (12.0%)17 (25.4%)0.099Hematological3 (2.4%)1 (2.9%)1 (4.0%)1 (1.5%)0.769Neurological10 (7.9%)4 (11.8%)2 (8.0%)4 (6.0%)0.559Ophthalmological1 (0.8%)1 (2.9%)0 (0.0%)0 (0.0%)0.468OSA severity was classified according to the Apnea–Hypopnea Index (AHI) from the baseline overnight polysomnography (PSG). Continuous variables are presented as mean (SD) and categorical variables as n (%). The P-value column compares the three severity groups using the test indicated by the superscript: ¹ Fisher’s exact test; ²ANOVA. All tests were two-sided (α = 0.05). Bold p-values indicate p < 0.05. Abbreviations: OSA, obstructive sleep apnea; AHI, Apnea–Hypopnea Index; PSG, polysomnography; BMI, body mass index. Baseline demographic, anthropometric and clinical characteristics by OSA severity group. OSA severity was classified according to the Apnea–Hypopnea Index (AHI) from the baseline overnight polysomnography (PSG). Continuous variables are presented as mean (SD) and categorical variables as n (%). The P-value column compares the three severity groups using the test indicated by the superscript: ¹ Fisher’s exact test; ²ANOVA. All tests were two-sided (α = 0.05). Bold p-values indicate p < 0.05. Abbreviations: OSA, obstructive sleep apnea; AHI, Apnea–Hypopnea Index; PSG, polysomnography; BMI, body mass index. Table 2Total sleep spindle density (events/min) at baseline and 6-month follow-up, by OSA severity group.Derivation/ VariableGroupBaselinemean (95%CI)6 Monthsmean (95%CI)Change(95%CI)p-interaction(Time×Group)q-value(FDR)Post-hoc(Šidák-adj.)F3-M2NREMNo/Mild2.74 (2.45, 3.03)2.71 (2.42, 3.00)-0.03 (-0.28, 0.22)0.14940.1494–Moderate2.83 (2.52, 3.16)2.85 (2.52, 3.18)0.02 (-0.27, 0.30)Severe2.59 (2.34, 2.85)2.85 (2.60, 3.11)0.26 (0.06, 0.46)F4-M1NREMNo/Mild2.28 (2.00, 2.55)2.16 (1.89, 2.44)-0.11 (-0.33, 0.10) 0.0021 0.0034 L-M: 0.801L-S: 0.037*M-S: 0.005**Moderate2.49 (2.18, 2.80)2.24 (1.93, 2.55)-0.25 (-0.50, -0.00)Severe2.07 (1.83, 2.32)2.32 (2.07, 2.56)0.24 (0.07, 0.42)C3-M2NREMNo/Mild2.30 (2.03, 2.56)2.33 (2.07, 2.60)0.04 (-0.14, 0.22) 0.0283 0.0378 L-M: 0.997L-S: 0.085M-S: 0.082Moderate2.42 (2.13, 2.72)2.44 (2.14, 2.73)0.01 (-0.19, 0.22)Severe2.24 (2.01, 2.47)2.53 (2.30, 2.76)0.29 (0.15, 0.44)C4-M1NREMNo/Mild2.14 (1.88, 2.41)2.12 (1.86, 2.38)-0.02 (-0.22, 0.18) 0.0014 0.0028 L-M: 0.999L-S: 0.007**M-S: 0.013*Moderate2.23 (1.93, 2.52)2.19 (1.90, 2.49)-0.03 (-0.26, 0.20)Severe1.89 (1.66, 2.12)2.27 (2.03, 2.50)0.38 (0.22, 0.54)F3-M2N2No/Mild2.79 (2.49, 3.08)2.74 (2.45, 3.03)-0.05 (-0.30, 0.20)0.09010.1030–Moderate2.81 (2.49, 3.14)2.82 (2.50, 3.15)0.01 (-0.27, 0.29)Severe2.57 (2.32, 2.83)2.85 (2.60, 3.11)0.28 (0.08, 0.48)F4-M1N2No/Mild2.34 (2.07, 2.61)2.18 (1.91, 2.45)-0.16 (-0.38, 0.06) < 0.001 0.0025 L-M: 0.956L-S: 0.008**M-S: 0.004**Moderate2.47 (2.16, 2.77)2.23 (1.93, 2.54)-0.24 (-0.48, 0.01)Severe2.06 (1.82, 2.30)2.33 (2.10, 2.57)0.27 (0.10, 0.45)C3-M2N2No/Mild2.46 (2.19, 2.73)2.47 (2.20, 2.74)0.01 (-0.17, 0.20) 0.0011 0.0028 L-M: 1.000L-S: 0.005**M-S: 0.012*Moderate2.52 (2.21, 2.82)2.53 (2.22, 2.83)0.01 (-0.20, 0.23)Severe2.29 (2.05, 2.52)2.68 (2.45, 2.92)0.40 (0.25, 0.55)C4-M1N2No/Mild2.29 (2.01, 2.56)2.24 (1.97, 2.52)-0.04 (-0.26, 0.17) < 0.001 0.0011 L-M: 0.998L-S: <0.001***M-S: 0.004**Moderate2.33 (2.02, 2.64)2.31 (2.00, 2.62)-0.02 (-0.26, 0.23)Severe1.96 (1.72, 2.20)2.43 (2.19, 2.67)0.47 (0.30, 0.65)NREM (N2 + N3) and stage N2 are reported separately. Groups: No/Mild n = 66, Moderate n = 50, Severe n = 100. Sleep spindle density is reported across fronto-central derivations (F3–M2, F4–M1, C3–M2, C4–M1). Values are estimated marginal means (95% CI) from linear mixed-effects models including fixed effects for time (baseline vs. 6 months), OSA severity group (No/Mild, Moderate, Severe), the time × group interaction, and prespecified covariates age, sex, and BMI, with a random intercept for participant. Change represents the within-group estimated difference (6 months minus baseline). The primary test of interest is the time × group interaction; corresponding p-values and Benjamini–Hochberg false discovery rate q-values (FDR within the Table 2 outcome family) are shown. Post-hoc pairwise contrasts compare differences in change between severity groups (difference-in-differences) with Šidák adjustment. L, No/Mild; M, Moderate; S, Severe. Asterisks denote Šidák-adjusted post-hoc p-values (*<0.05, **<0.01, ***<0.001). Table 2Total sleep spindle density (events/min) at baseline and 6-month follow-up, by OSA severity group.Derivation/ VariableGroupBaselinemean (95%CI)6 Monthsmean (95%CI)Change(95%CI)p-interaction(Time×Group)q-value(FDR)Post-hoc(Šidák-adj.)F3-M2NREMNo/Mild2.74 (2.45, 3.03)2.71 (2.42, 3.00)-0.03 (-0.28, 0.22)0.14940.1494–Moderate2.83 (2.52, 3.16)2.85 (2.52, 3.18)0.02 (-0.27, 0.30)Severe2.59 (2.34, 2.85)2.85 (2.60, 3.11)0.26 (0.06, 0.46)F4-M1NREMNo/Mild2.28 (2.00, 2.55)2.16 (1.89, 2.44)-0.11 (-0.33, 0.10) 0.0021 0.0034 L-M: 0.801L-S: 0.037*M-S: 0.005**Moderate2.49 (2.18, 2.80)2.24 (1.93, 2.55)-0.25 (-0.50, -0.00)Severe2.07 (1.83, 2.32)2.32 (2.07, 2.56)0.24 (0.07, 0.42)C3-M2NREMNo/Mild2.30 (2.03, 2.56)2.33 (2.07, 2.60)0.04 (-0.14, 0.22) 0.0283 0.0378 L-M: 0.997L-S: 0.085M-S: 0.082Moderate2.42 (2.13, 2.72)2.44 (2.14, 2.73)0.01 (-0.19, 0.22)Severe2.24 (2.01, 2.47)2.53 (2.30, 2.76)0.29 (0.15, 0.44)C4-M1NREMNo/Mild2.14 (1.88, 2.41)2.12 (1.86, 2.38)-0.02 (-0.22, 0.18) 0.0014 0.0028 L-M: 0.999L-S: 0.007**M-S: 0.013*Moderate2.23 (1.93, 2.52)2.19 (1.90, 2.49)-0.03 (-0.26, 0.20)Severe1.89 (1.66, 2.12)2.27 (2.03, 2.50)0.38 (0.22, 0.54)F3-M2N2No/Mild2.79 (2.49, 3.08)2.74 (2.45, 3.03)-0.05 (-0.30, 0.20)0.09010.1030–Moderate2.81 (2.49, 3.14)2.82 (2.50, 3.15)0.01 (-0.27, 0.29)Severe2.57 (2.32, 2.83)2.85 (2.60, 3.11)0.28 (0.08, 0.48)F4-M1N2No/Mild2.34 (2.07, 2.61)2.18 (1.91, 2.45)-0.16 (-0.38, 0.06) < 0.001 0.0025 L-M: 0.956L-S: 0.008**M-S: 0.004**Moderate2.47 (2.16, 2.77)2.23 (1.93, 2.54)-0.24 (-0.48, 0.01)Severe2.06 (1.82, 2.30)2.33 (2.10, 2.57)0.27 (0.10, 0.45)C3-M2N2No/Mild2.46 (2.19, 2.73)2.47 (2.20, 2.74)0.01 (-0.17, 0.20) 0.0011 0.0028 L-M: 1.000L-S: 0.005**M-S: 0.012*Moderate2.52 (2.21, 2.82)2.53 (2.22, 2.83)0.01 (-0.20, 0.23)Severe2.29 (2.05, 2.52)2.68 (2.45, 2.92)0.40 (0.25, 0.55)C4-M1N2No/Mild2.29 (2.01, 2.56)2.24 (1.97, 2.52)-0.04 (-0.26, 0.17) < 0.001 0.0011 L-M: 0.998L-S: <0.001***M-S: 0.004**Moderate2.33 (2.02, 2.64)2.31 (2.00, 2.62)-0.02 (-0.26, 0.23)Severe1.96 (1.72, 2.20)2.43 (2.19, 2.67)0.47 (0.30, 0.65)NREM (N2 + N3) and stage N2 are reported separately. Groups: No/Mild n = 66, Moderate n = 50, Severe n = 100. Sleep spindle density is reported across fronto-central derivations (F3–M2, F4–M1, C3–M2, C4–M1). Values are estimated marginal means (95% CI) from linear mixed-effects models including fixed effects for time (baseline vs. 6 months), OSA severity group (No/Mild, Moderate, Severe), the time × group interaction, and prespecified covariates age, sex, and BMI, with a random intercept for participant. Change represents the within-group estimated difference (6 months minus baseline). The primary test of interest is the time × group interaction; corresponding p-values and Benjamini–Hochberg false discovery rate q-values (FDR within the Table 2 outcome family) are shown. Post-hoc pairwise contrasts compare differences in change between severity groups (difference-in-differences) with Šidák adjustment. L, No/Mild; M, Moderate; S, Severe. Asterisks denote Šidák-adjusted post-hoc p-values (*<0.05, **<0.01, ***<0.001). Total sleep spindle density (events/min) at baseline and 6-month follow-up, by OSA severity group. F3-M2 NREM F4-M1 NREM L-M: 0.801 L-S: 0.037* M-S: 0.005** C3-M2 NREM L-M: 0.997 L-S: 0.085 M-S: 0.082 C4-M1 NREM L-M: 0.999 L-S: 0.007** M-S: 0.013* F3-M2 N2 F4-M1 N2 L-M: 0.956 L-S: 0.008** M-S: 0.004** C3-M2 N2 L-M: 1.000 L-S: 0.005** M-S: 0.012* C4-M1 N2 L-M: 0.998 L-S: <0.001*** M-S: 0.004** NREM (N2 + N3) and stage N2 are reported separately. Groups: No/Mild n = 66, Moderate n = 50, Severe n = 100. Sleep spindle density is reported across fronto-central derivations (F3–M2, F4–M1, C3–M2, C4–M1). Values are estimated marginal means (95% CI) from linear mixed-effects models including fixed effects for time (baseline vs. 6 months), OSA severity group (No/Mild, Moderate, Severe), the time × group interaction, and prespecified covariates age, sex, and BMI, with a random intercept for participant. Change represents the within-group estimated difference (6 months minus baseline). The primary test of interest is the time × group interaction; corresponding p-values and Benjamini–Hochberg false discovery rate q-values (FDR within the Table 2 outcome family) are shown. Post-hoc pairwise contrasts compare differences in change between severity groups (difference-in-differences) with Šidák adjustment. L, No/Mild; M, Moderate; S, Severe. Asterisks denote Šidák-adjusted post-hoc p-values (*<0.05, **<0.01, ***<0.001). Table 3Slow and fast sleep spindle density (events/min) during stage N2 - Baseline, 6 months follow up and change by OSA severity group Stage N2 is reported only.Derivation/ VariableGroupBaselinemean (95%CI)6 Monthsmean (95%CI)Change(95%CI)p-interaction(Time×Group)q-value(FDR)Post-hoc(Šidák-adj.)F3-M2FastNo/Mild1.80 (1.53, 2.07)1.75 (1.48, 2.02)-0.05 (-0.33, 0.23)0.95130.9513–Moderate1.76 (1.46, 2.06)1.71 (1.41, 2.01)-0.05 (-0.37, 0.27)Severe1.64 (1.41, 1.87)1.64 (1.41, 1.87)0.00 (-0.23, 0.23)F4-M1FastNo/Mild1.14 (0.94, 1.34)0.91 (0.71, 1.11)-0.23 (-0.46, -0.00)0.27660.3161–Moderate1.19 (0.97, 1.42)0.91 (0.68, 1.14)-0.28 (-0.55, -0.02)Severe0.97 (0.79, 1.14)0.92 (0.74, 1.09)-0.05 (-0.24, 0.13)C3-M2FastNo/Mild1.78 (1.55, 2.01)1.81 (1.57, 2.04)0.03 (-0.14, 0.19) 0.0230 0.0573–Moderate1.80 (1.54, 2.06)1.78 (1.52, 2.04)-0.02 (-0.21, 0.18)Severe1.66 (1.46, 1.87)1.93 (1.72, 2.13)0.26 (0.13, 0.40)C4-M1FastNo/Mild1.55 (1.33, 1.77)1.54 (1.32, 1.76)-0.01 (-0.20, 0.18) 0.0049 0.0392 L-M: 0.999L-S: 0.020*M-S: 0.026*Moderate1.59 (1.34, 1.84)1.56 (1.31, 1.81)-0.03 (-0.25, 0.19)Severe1.32 (1.13, 1.52)1.65 (1.45, 1.84)0.33 (0.17, 0.48)F3-M2SlowNo/Mild0.99 (0.80, 1.18)0.99 (0.80, 1.18)0.00 (-0.17, 0.17) 0.0286 0.0573–Moderate1.06 (0.85, 1.27)1.11 (0.90, 1.32)0.06 (-0.14, 0.25)Severe0.93 (0.77, 1.10)1.21 (1.05, 1.38)0.28 (0.14, 0.42)F4-M1SlowNo/Mild1.20 (0.99, 1.40)1.27 (1.06, 1.47)0.07 (-0.09, 0.23) 0.0169 0.0573–Moderate1.27 (1.04, 1.50)1.32 (1.09, 1.55)0.05 (-0.14, 0.24)Severe1.10 (0.91, 1.28)1.42 (1.24, 1.60)0.32 (0.19, 0.46)C3-M2SlowNo/Mild0.68 (0.55, 0.80)0.66 (0.53, 0.79)-0.02 (-0.12, 0.09)0.06430.0858–Moderate0.72 (0.58, 0.86)0.74 (0.60, 0.89)0.03 (-0.09, 0.15)Severe0.62 (0.51, 0.73)0.76 (0.64, 0.87)0.14 (0.05, 0.22)C4-M1SlowNo/Mild0.73 (0.60, 0.87)0.70 (0.56, 0.84)-0.03 (-0.15, 0.08) 0.0435 0.0697–Moderate0.74 (0.59, 0.90)0.75 (0.60, 0.91)0.01 (-0.12, 0.15)Severe0.64 (0.51, 0.76)0.78 (0.66, 0.90)0.15 (0.05, 0.24)Fast spindles were defined as 13–16 Hz and slow spindles as 11–13 Hz. Groups: No/Mild n = 66, Moderate n = 50, Severe n = 100. Sleep spindle density is reported across fronto-central derivations (F3–M2, F4–M1, C3–M2, C4–M1). Values are estimated marginal means (95% CI) from linear mixed-effects models including fixed effects for time (baseline vs. 6 months), OSA severity group (No/Mild, Moderate, Severe), the time × group interaction, and prespecified covariates age, sex, and BMI, with a random intercept for participant. Change represents the within-group estimated difference (6 months minus baseline). The primary test of interest is the time × group interaction; corresponding p-values and Benjamini–Hochberg false discovery rate q-values (FDR within the Table 3 outcome family) are shown. Post-hoc pairwise contrasts compare differences in change between severity groups (difference-in-differences) with Šidák adjustment. L, No/Mild; M, Moderate; S, Severe. Asterisks denote Šidák-adjusted post-hoc p-values (*<0.05, **<0.01, ***<0.001). Table 3Slow and fast sleep spindle density (events/min) during stage N2 - Baseline, 6 months follow up and change by OSA severity group Stage N2 is reported only.Derivation/ VariableGroupBaselinemean (95%CI)6 Monthsmean (95%CI)Change(95%CI)p-interaction(Time×Group)q-value(FDR)Post-hoc(Šidák-adj.)F3-M2FastNo/Mild1.80 (1.53, 2.07)1.75 (1.48, 2.02)-0.05 (-0.33, 0.23)0.95130.9513–Moderate1.76 (1.46, 2.06)1.71 (1.41, 2.01)-0.05 (-0.37, 0.27)Severe1.64 (1.41, 1.87)1.64 (1.41, 1.87)0.00 (-0.23, 0.23)F4-M1FastNo/Mild1.14 (0.94, 1.34)0.91 (0.71, 1.11)-0.23 (-0.46, -0.00)0.27660.3161–Moderate1.19 (0.97, 1.42)0.91 (0.68, 1.14)-0.28 (-0.55, -0.02)Severe0.97 (0.79, 1.14)0.92 (0.74, 1.09)-0.05 (-0.24, 0.13)C3-M2FastNo/Mild1.78 (1.55, 2.01)1.81 (1.57, 2.04)0.03 (-0.14, 0.19) 0.0230 0.0573–Moderate1.80 (1.54, 2.06)1.78 (1.52, 2.04)-0.02 (-0.21, 0.18)Severe1.66 (1.46, 1.87)1.93 (1.72, 2.13)0.26 (0.13, 0.40)C4-M1FastNo/Mild1.55 (1.33, 1.77)1.54 (1.32, 1.76)-0.01 (-0.20, 0.18) 0.0049 0.0392 L-M: 0.999L-S: 0.020*M-S: 0.026*Moderate1.59 (1.34, 1.84)1.56 (1.31, 1.81)-0.03 (-0.25, 0.19)Severe1.32 (1.13, 1.52)1.65 (1.45, 1.84)0.33 (0.17, 0.48)F3-M2SlowNo/Mild0.99 (0.80, 1.18)0.99 (0.80, 1.18)0.00 (-0.17, 0.17) 0.0286 0.0573–Moderate1.06 (0.85, 1.27)1.11 (0.90, 1.32)0.06 (-0.14, 0.25)Severe0.93 (0.77, 1.10)1.21 (1.05, 1.38)0.28 (0.14, 0.42)F4-M1SlowNo/Mild1.20 (0.99, 1.40)1.27 (1.06, 1.47)0.07 (-0.09, 0.23) 0.0169 0.0573–Moderate1.27 (1.04, 1.50)1.32 (1.09, 1.55)0.05 (-0.14, 0.24)Severe1.10 (0.91, 1.28)1.42 (1.24, 1.60)0.32 (0.19, 0.46)C3-M2SlowNo/Mild0.68 (0.55, 0.80)0.66 (0.53, 0.79)-0.02 (-0.12, 0.09)0.06430.0858–Moderate0.72 (0.58, 0.86)0.74 (0.60, 0.89)0.03 (-0.09, 0.15)Severe0.62 (0.51, 0.73)0.76 (0.64, 0.87)0.14 (0.05, 0.22)C4-M1SlowNo/Mild0.73 (0.60, 0.87)0.70 (0.56, 0.84)-0.03 (-0.15, 0.08) 0.0435 0.0697–Moderate0.74 (0.59, 0.90)0.75 (0.60, 0.91)0.01 (-0.12, 0.15)Severe0.64 (0.51, 0.76)0.78 (0.66, 0.90)0.15 (0.05, 0.24)Fast spindles were defined as 13–16 Hz and slow spindles as 11–13 Hz. Groups: No/Mild n = 66, Moderate n = 50, Severe n = 100. Sleep spindle density is reported across fronto-central derivations (F3–M2, F4–M1, C3–M2, C4–M1). Values are estimated marginal means (95% CI) from linear mixed-effects models including fixed effects for time (baseline vs. 6 months), OSA severity group (No/Mild, Moderate, Severe), the time × group interaction, and prespecified covariates age, sex, and BMI, with a random intercept for participant. Change represents the within-group estimated difference (6 months minus baseline). The primary test of interest is the time × group interaction; corresponding p-values and Benjamini–Hochberg false discovery rate q-values (FDR within the Table 3 outcome family) are shown. Post-hoc pairwise contrasts compare differences in change between severity groups (difference-in-differences) with Šidák adjustment. L, No/Mild; M, Moderate; S, Severe. Asterisks denote Šidák-adjusted post-hoc p-values (*<0.05, **<0.01, ***<0.001). Slow and fast sleep spindle density (events/min) during stage N2 - Baseline, 6 months follow up and change by OSA severity group Stage N2 is reported only. F3-M2 Fast F4-M1 Fast C3-M2 Fast C4-M1 Fast L-M: 0.999 L-S: 0.020* M-S: 0.026* F3-M2 Slow F4-M1 Slow C3-M2 Slow C4-M1 Slow Fast spindles were defined as 13–16 Hz and slow spindles as 11–13 Hz. Groups: No/Mild n = 66, Moderate n = 50, Severe n = 100. Sleep spindle density is reported across fronto-central derivations (F3–M2, F4–M1, C3–M2, C4–M1). Values are estimated marginal means (95% CI) from linear mixed-effects models including fixed effects for time (baseline vs. 6 months), OSA severity group (No/Mild, Moderate, Severe), the time × group interaction, and prespecified covariates age, sex, and BMI, with a random intercept for participant. Change represents the within-group estimated difference (6 months minus baseline). The primary test of interest is the time × group interaction; corresponding p-values and Benjamini–Hochberg false discovery rate q-values (FDR within the Table 3 outcome family) are shown. Post-hoc pairwise contrasts compare differences in change between severity groups (difference-in-differences) with Šidák adjustment. L, No/Mild; M, Moderate; S, Severe. Asterisks denote Šidák-adjusted post-hoc p-values (*<0.05, **<0.01, ***<0.001). Table 4PSG-derived sleep macrostructure - Baseline, 6 months and change by OSA severity group.Derivation/ VariableGroupBaselinemean (95%CI)6 Monthsmean (95%CI)Change(95%CI)p-interaction(Time×Group)q-value(FDR)Post-hoc(Šidák-adj.)TST (min)No/Mild360.02 (347.70, 372.35)367.67 (355.35, 380.00)7.65 (-5.70, 21.00)0.17580.2638–Moderate351.39 (337.54, 365.25)373.32 (359.46, 387.18)21.93 (6.58, 37.27)Severe371.77 (361.08, 382.46)376.19 (365.50, 386.88)4.42 (-6.43, 15.27)SE (%)No/Mild81.52 (78.97, 84.06)82.46 (79.92, 85.01)0.94 (-1.79, 3.68)0.53830.5873–Moderate81.69 (78.83, 84.55)84.28 (81.42, 87.14)2.59 (-0.55, 5.74)Severe82.09 (79.89, 84.30)82.52 (80.32, 84.73)0.43 (-1.79, 2.65)WASO (min)No/Mild64.08 (46.72, 81.45)50.92 (33.70, 68.14)-13.17 (-33.52, 7.19)0.26620.3195–Moderate82.01 (62.09, 101.92)49.88 (30.49, 69.28)-32.12 (-55.86, -8.39)Severe62.94 (48.00, 77.88)54.40 (39.52, 69.28)-8.54 (-25.05, 7.96)NREMsleep (min)No/Mild293.25 (282.40, 304.10)301.01 (290.16, 311.86)7.76 (-4.98, 20.50)0.15490.2638–Moderate289.57 (277.35, 301.79)303.73 (291.52, 315.95)14.16 (-0.47, 28.80)Severe311.49 (302.12, 320.86)308.98 (299.61, 318.35)-2.51 (-12.86, 7.84)N1 (% TST)No/Mild7.26 (5.44, 9.08)8.73 (6.91, 10.55)1.47 (-0.79, 3.72)0.88570.8857–Moderate7.04 (4.99, 9.10)9.29 (7.24, 11.35)2.25 (-0.34, 4.84)Severe10.48 (8.92, 12.05)12.54 (10.98, 14.11)2.06 (0.23, 3.89)N2 (% TST)No/Mild57.67 (55.08, 60.27)54.32 (51.73, 56.92)-3.35 (-6.66, -0.04) < 0.001 0.0012 L-M: 0.976L-S: 0.001**M-S: 0.013*Moderate57.93 (55.00, 60.86)53.63 (50.70, 56.56)-4.30 (-8.10, -0.50)Severe60.83 (58.60, 63.07)49.75 (47.52, 51.98)-11.09 (-13.77, -8.40)N3 (% TST)No/Mild16.79 (14.57, 19.02)18.98 (16.75, 21.20)2.18 (-0.45, 4.81) 0.0046 0.0110 L-M: 0.933L-S: 0.038*M-S: 0.013*Moderate17.65 (15.14, 20.16)18.75 (16.24, 21.25)1.10 (-1.92, 4.12)Severe13.11 (11.19, 15.03)19.61 (17.69, 21.53)6.50 (4.36, 8.63)REM (% TST)No/Mild18.30 (16.91, 19.70)18.01 (16.61, 19.40)-0.30 (-2.16, 1.57)0.09850.1970–Moderate17.41 (15.83, 18.99)18.32 (16.74, 19.90)0.91 (-1.23, 3.05)Severe15.78 (14.58, 16.97)18.08 (16.88, 19.27)2.30 (0.79, 3.82)Totalarousal (ev/h)No/Mild10.20 (5.59, 14.81)13.21 (8.61, 17.82)3.01 (-2.71, 8.74)0.20490.2732–Moderate15.08 (9.88, 20.27)17.34 (12.15, 22.53)2.26 (-4.31, 8.84)Severe27.98 (24.01, 31.94)24.96 (21.00, 28.92)-3.02 (-7.67, 1.63)Resp. arousal(ev/h)No/Mild4.77 (0.49, 9.04)5.84 (1.37, 10.31)1.07 (-4.64, 6.79) < 0.001 < 0.001 L-M: 0.996L-S: <0.001***M-S: <0.001***Moderate7.97 (3.43, 12.51)9.89 (4.99, 14.80)1.93 (-4.13, 7.99)Severe24.47 (21.08, 27.85)10.63 (7.04, 14.22)-13.84 (-17.84, -9.84)ODI (ev/h)No/Mild7.50 (3.87, 11.13)9.92 (6.32, 13.53)2.42 (-2.56, 7.40) < 0.001 < 0.001 L-M: 0.694L-S: <0.001***M-S: <0.001***Moderate14.37 (10.29, 18.44)13.01 (8.93, 17.08)-1.36 (-7.06, 4.34)Severe45.13 (42.05, 48.22)14.77 (11.68, 17.86)-30.36 (-34.41, -26.31)Hypoxic burden(%·min/h TST)No/Mild13.85 (-15.13, 42.82)16.94 (-11.34, 45.22)3.10 (-36.61, 42.80) < 0.001 < 0.001 L-M: 0.977L-S: <0.001***M-S: <0.001***Moderate33.78 (4.40, 63.17)26.55 (-6.61, 59.72)-7.23 (-47.06, 32.60)Severe143.16 (119.47, 166.86)30.18 (9.39, 50.97)-112.98 (-137.35, -88.62)Groups: No/Mild n = 66, Moderate n = 50, Severe n = 100. TST, total sleep time; SE, sleep efficiency; WASO, wake after sleep onset; ODI, oxygen desaturation index. Values are estimated marginal means (95% CI) from linear mixed-effects models including fixed effects for time (baseline vs. 6 months), OSA severity group (No/Mild, Moderate, Severe), the time × group interaction, and prespecified covariates age, sex, and BMI, with a random intercept for participant. Change represents the within-group estimated difference (6 months minus baseline). The primary test of interest is the time × group interaction; corresponding p-values and Benjamini–Hochberg false discovery rate q-values (FDR within the Table 4 outcome family) are shown. Post-hoc pairwise contrasts compare differences in change between severity groups (difference-in-differences) with Šidák adjustment. L, No/Mild; M, Moderate; S, Severe. Asterisks denote Šidák-adjusted post-hoc p-values (*<0.05, **<0.01, ***<0.001). Table 4PSG-derived sleep macrostructure - Baseline, 6 months and change by OSA severity group.Derivation/ VariableGroupBaselinemean (95%CI)6 Monthsmean (95%CI)Change(95%CI)p-interaction(Time×Group)q-value(FDR)Post-hoc(Šidák-adj.)TST (min)No/Mild360.02 (347.70, 372.35)367.67 (355.35, 380.00)7.65 (-5.70, 21.00)0.17580.2638–Moderate351.39 (337.54, 365.25)373.32 (359.46, 387.18)21.93 (6.58, 37.27)Severe371.77 (361.08, 382.46)376.19 (365.50, 386.88)4.42 (-6.43, 15.27)SE (%)No/Mild81.52 (78.97, 84.06)82.46 (79.92, 85.01)0.94 (-1.79, 3.68)0.53830.5873–Moderate81.69 (78.83, 84.55)84.28 (81.42, 87.14)2.59 (-0.55, 5.74)Severe82.09 (79.89, 84.30)82.52 (80.32, 84.73)0.43 (-1.79, 2.65)WASO (min)No/Mild64.08 (46.72, 81.45)50.92 (33.70, 68.14)-13.17 (-33.52, 7.19)0.26620.3195–Moderate82.01 (62.09, 101.92)49.88 (30.49, 69.28)-32.12 (-55.86, -8.39)Severe62.94 (48.00, 77.88)54.40 (39.52, 69.28)-8.54 (-25.05, 7.96)NREMsleep (min)No/Mild293.25 (282.40, 304.10)301.01 (290.16, 311.86)7.76 (-4.98, 20.50)0.15490.2638–Moderate289.57 (277.35, 301.79)303.73 (291.52, 315.95)14.16 (-0.47, 28.80)Severe311.49 (302.12, 320.86)308.98 (299.61, 318.35)-2.51 (-12.86, 7.84)N1 (% TST)No/Mild7.26 (5.44, 9.08)8.73 (6.91, 10.55)1.47 (-0.79, 3.72)0.88570.8857–Moderate7.04 (4.99, 9.10)9.29 (7.24, 11.35)2.25 (-0.34, 4.84)Severe10.48 (8.92, 12.05)12.54 (10.98, 14.11)2.06 (0.23, 3.89)N2 (% TST)No/Mild57.67 (55.08, 60.27)54.32 (51.73, 56.92)-3.35 (-6.66, -0.04) < 0.001 0.0012 L-M: 0.976L-S: 0.001**M-S: 0.013*Moderate57.93 (55.00, 60.86)53.63 (50.70, 56.56)-4.30 (-8.10, -0.50)Severe60.83 (58.60, 63.07)49.75 (47.52, 51.98)-11.09 (-13.77, -8.40)N3 (% TST)No/Mild16.79 (14.57, 19.02)18.98 (16.75, 21.20)2.18 (-0.45, 4.81) 0.0046 0.0110 L-M: 0.933L-S: 0.038*M-S: 0.013*Moderate17.65 (15.14, 20.16)18.75 (16.24, 21.25)1.10 (-1.92, 4.12)Severe13.11 (11.19, 15.03)19.61 (17.69, 21.53)6.50 (4.36, 8.63)REM (% TST)No/Mild18.30 (16.91, 19.70)18.01 (16.61, 19.40)-0.30 (-2.16, 1.57)0.09850.1970–Moderate17.41 (15.83, 18.99)18.32 (16.74, 19.90)0.91 (-1.23, 3.05)Severe15.78 (14.58, 16.97)18.08 (16.88, 19.27)2.30 (0.79, 3.82)Totalarousal (ev/h)No/Mild10.20 (5.59, 14.81)13.21 (8.61, 17.82)3.01 (-2.71, 8.74)0.20490.2732–Moderate15.08 (9.88, 20.27)17.34 (12.15, 22.53)2.26 (-4.31, 8.84)Severe27.98 (24.01, 31.94)24.96 (21.00, 28.92)-3.02 (-7.67, 1.63)Resp. arousal(ev/h)No/Mild4.77 (0.49, 9.04)5.84 (1.37, 10.31)1.07 (-4.64, 6.79) < 0.001 < 0.001 L-M: 0.996L-S: <0.001***M-S: <0.001***Moderate7.97 (3.43, 12.51)9.89 (4.99, 14.80)1.93 (-4.13, 7.99)Severe24.47 (21.08, 27.85)10.63 (7.04, 14.22)-13.84 (-17.84, -9.84)ODI (ev/h)No/Mild7.50 (3.87, 11.13)9.92 (6.32, 13.53)2.42 (-2.56, 7.40) < 0.001 < 0.001 L-M: 0.694L-S: <0.001***M-S: <0.001***Moderate14.37 (10.29, 18.44)13.01 (8.93, 17.08)-1.36 (-7.06, 4.34)Severe45.13 (42.05, 48.22)14.77 (11.68, 17.86)-30.36 (-34.41, -26.31)Hypoxic burden(%·min/h TST)No/Mild13.85 (-15.13, 42.82)16.94 (-11.34, 45.22)3.10 (-36.61, 42.80) < 0.001 < 0.001 L-M: 0.977L-S: <0.001***M-S: <0.001***Moderate33.78 (4.40, 63.17)26.55 (-6.61, 59.72)-7.23 (-47.06, 32.60)Severe143.16 (119.47, 166.86)30.18 (9.39, 50.97)-112.98 (-137.35, -88.62)Groups: No/Mild n = 66, Moderate n = 50, Severe n = 100. TST, total sleep time; SE, sleep efficiency; WASO, wake after sleep onset; ODI, oxygen desaturation index. Values are estimated marginal means (95% CI) from linear mixed-effects models including fixed effects for time (baseline vs. 6 months), OSA severity group (No/Mild, Moderate, Severe), the time × group interaction, and prespecified covariates age, sex, and BMI, with a random intercept for participant. Change represents the within-group estimated difference (6 months minus baseline). The primary test of interest is the time × group interaction; corresponding p-values and Benjamini–Hochberg false discovery rate q-values (FDR within the Table 4 outcome family) are shown. Post-hoc pairwise contrasts compare differences in change between severity groups (difference-in-differences) with Šidák adjustment. L, No/Mild; M, Moderate; S, Severe. Asterisks denote Šidák-adjusted post-hoc p-values (*<0.05, **<0.01, ***<0.001). PSG-derived sleep macrostructure - Baseline, 6 months and change by OSA severity group. NREM sleep (min) L-M: 0.976 L-S: 0.001** M-S: 0.013* L-M: 0.933 L-S: 0.038* M-S: 0.013* Total arousal (ev/h) Resp. arousal (ev/h) L-M: 0.996 L-S: <0.001*** M-S: <0.001*** L-M: 0.694 L-S: <0.001*** M-S: <0.001*** Hypoxic burden (%·min/h TST) L-M: 0.977 L-S: <0.001*** M-S: <0.001*** Groups: No/Mild n = 66, Moderate n = 50, Severe n = 100. TST, total sleep time; SE, sleep efficiency; WASO, wake after sleep onset; ODI, oxygen desaturation index. Values are estimated marginal means (95% CI) from linear mixed-effects models including fixed effects for time (baseline vs. 6 months), OSA severity group (No/Mild, Moderate, Severe), the time × group interaction, and prespecified covariates age, sex, and BMI, with a random intercept for participant. Change represents the within-group estimated difference (6 months minus baseline). The primary test of interest is the time × group interaction; corresponding p-values and Benjamini–Hochberg false discovery rate q-values (FDR within the Table 4 outcome family) are shown. Post-hoc pairwise contrasts compare differences in change between severity groups (difference-in-differences) with Šidák adjustment. L, No/Mild; M, Moderate; S, Severe. Asterisks denote Šidák-adjusted post-hoc p-values (*<0.05, **<0.01, ***<0.001). Results Participant characteristics Out of the 352 volunteers recruited for the APNiA study, 216 participants with PSG records available both at baseline and at six months were included in the present analyses (Fig. 1). Of these, 66 had mild to no OSA (AHI < 15 events/hour), 50 had moderate OSA (AHI 15–<30), and 100 had severe OSA (AHI ≥ 30) (Table 1). The proportion of males was higher with increasing OSA severity (45.5% in the AHI < 15 group vs. 72% and 79% in the moderate and severe groups, respectively; p < 0.001), and mean age was also higher (47.5 ± 10.6 vs. 51.5 ± 11.1 vs. 55.2 ± 11.3 years, p < 0.001). Markers of adiposity showed a similar gradient, with higher BMI, neck circumference and waist circumference detected in the more severe OSA group (all p < 0.001). Systolic and diastolic blood pressures were also higher in the severe OSA group compared with the milder categories (both p < 0.001). The prevalence of comorbidities (Table 1) did not differ significantly across severity groups (p = 0.181), except for cardiovascular comorbidity, which was more prevalent in patients with severe OSA (Table 1; p = 0.031). CPAP therapy was prescribed to 25.9% of patients with moderate OSA and 78.2% of patients with severe OSA. The mean prescribed CPAP duration (target use) was 6.4 ± 1.8 h per night, with no statistically significant differences between the two groups (p = 0.354). CPAP was not indicated for patients in the no OSA/mild OSA group. Among the remaining severe OSA cases, 17.2% were recommended lifestyle modifications (weight loss, avoidance of alcohol and sedatives, and lateral sleep positioning), and 4% received a prescription for a positional therapy device. The 6-month PSG was conducted with prescribed therapy in use when applicable, as detailed in the Methods. Sleep spindle density during NREM sleep and N2 After six months of standard clinical treatment, changes in sleep spindle density (SSD) increased in the severe OSA group, whereas no significant changes were observed in the non-OSA/mild OSA or moderate OSA groups (Table 2). In linear mixed-effects models adjusted for age, sex, and BMI, the time × group interaction was significant for NREM SSD at F4–M1, C3–M2, and C4–M1 (all qFDR < 0.05), whereas F3–M2 did not show a significant interaction after FDR correction (Table 2). Post-hoc contrasts indicated that the increase in SSD in the severe OSA group was greater than in the No/Mild and Moderate groups at F4–M1 and C4–M1 (Šidák-adjusted p < 0.05) (Table 2). A similar pattern was observed when the analysis was restricted to stage N2, with significant time × group interactions at F4–M1, C3–M2, and C4–M1 after FDR correction (Table 2). When N2 spindles were classified by frequency (fast > 13 Hz, slow ≤ 13 Hz), severe OSA patients showed increases in fast spindles at the central electrodes (C3, C4; +0.26 and + 0.33 events/min, respectively), whereas changes at the frontal electrodes were smaller or absent (Table 3). After correction for multiple comparisons, a significant time × group interaction was observed only for fast spindles at C4-M1 (qFDR < 0.05), with post-hoc contrasts indicating larger increases in severe OSA compared with No/Mild and Moderate OSA (Table 3). Analysis of slow spindles revealed nominally significant increases at the frontal electrodes F3-M2 and F4-M1 (+ 0.28 and + 0.32 events/min) and smaller non-significant increases at the central electrodes C3-M2 and C4-M1 (+ 0.14 and + 0.15 events/min). However, these slow-spindle effects did not survive FDR correction (Table 3). Together, these results indicate that changes were most robust for fast spindles at central sites, whereas slow-spindle changes were less consistent after multiple-comparison correction. Baseline spindle density did not differ by OSA severity but was influenced by demographics: age and higher BMI were associated with lower density, whereas women had higher spindle density (Table S1). The severe OSA group was older, had higher BMI, and was predominantly male (Table 1). Adjustment for age, sex, and BMI modestly attenuated effect estimates but did not alter the statistical significance of the Time × Group interaction for the primary SSD outcomes reported in Table 2 (q < 0.05), supporting a severity-specific pattern of change (Table S1). Additional fixed-effect estimates for N2 fast and slow spindle density models are reported in Table S2. Polysomnography-derived sleep macrostructure PSG-derived macroarchitecture measures by OSA severity are summarised in Table 4. From baseline to six months, significant between-group differences were observed in the distribution of sleep stages. The percentage of N2 sleep decreased more markedly in the severe OSA group (− 11.09%; 95% CI: −13.77 to − 8.40) compared with the No/Mild (− 3.35%; 95% CI: −6.66 to − 0.04) and Moderate groups (− 4.30%; 95% CI: −8.10 to − 0.50) (p-interaction < 0.001; qFDR = 0.0012; Table 4). Conversely, deep N3 sleep increased significantly in the severe group (+ 6.50%; 95% CI: 4.36 to 8.63), whereas the increases were smaller in the No/Mild (+ 2.18%; 95% CI: −0.45 to 4.81) and Moderate groups (+ 1.10%; 95% CI: −1.92 to 4.12) (p-interaction = 0.0046; qFDR = 0.0110; Table 4). The respiratory arousal index, ODI, and HB, all decreased markedly in the severe OSA group, whereas the No/Mild and Moderate groups showed only small, non-significant changes. The Time × Group interaction was significant for all three variables (all p-interaction < 0.001; all qFDR < 0.001; Table 4), and post-hoc comparisons confirmed significant differences between the severe group and the other two groups (all adjusted p < 0.001; Table 4). In mixed-effects models adjusted for age, sex, and BMI, β coefficients represent model-based estimates expressed in the natural units of each outcome. No significant changes over time were observed in TST, SE, or WASO (Table S3). However, age was significantly associated with reduced TST (β = −1.06 min/year; p < 0.001) and SE (β = −0.27%; p < 0.001), as well as increased WASO (β = 1.30 min/year; p = 0.0002). Regarding sleep architecture, N2 sleep decreased over time (β = −3.35%; p = 0.047), particularly in the severe OSA group (time × severe OSA interaction: β = −7.74; p = 0.0004), while deep N3 sleep increased in this group (interaction: β = 4.32; p = 0.0127). REM sleep was lower in severe OSA compared with the No/Mild group (β = −2.53; p = 0.009) but increased significantly over time in the severe OSA group (time × severe OSA interaction: β = 2.60; p = 0.034). Similarly, patients with severe OSA had higher baseline respiratory values than the No/Mild group for total arousals (β = 17.78; p < 0.001), respiratory arousals (β = 19.70; p < 0.001), ODI (β = 37.63; p < 0.001), and HB (β = 129.29; p < 0.001). These parameters decreased significantly over the follow-up in the severe OSA group, including respiratory arousals (time × severe OSA interaction: β = −14.92; p < 0.001), ODI (β = −32.79; p < 0.001), and HB (β = −116.08; p < 0.001). Discussion In this study, we quantified sleep spindle density in adults with suspected OSA and examined how spindle density changed after 6 months of standard clinical management, stratified by baseline OSA severity. At baseline, SSD was significantly lower in patients with severe OSA compared with those with mild or moderate disease. At follow-up, significant increases in SSD were observed exclusively in the severe OSA group, whereas patients with mild-no OSA or moderate OSA showed little or no consistent changes. Taken together, these findings indicate a partial recovery of sleep microarchitecture in patients with severe OSA following treatment and suggest that SSD may represent a sensitive marker of treatment-related changes in sleep microstructure in this subgroup. Several studies have reported an independent association between sleep spindle metrics and the AHI, indicating that greater OSA severity is associated with reduced SSD35,36. Recurrent apneic episodes could impair the interactions between the thalamic reticular nucleus and thalamocortical neurons, thereby disrupting spindle generation37,38. In addition to its association with the AHI, several studies have also suggested a relationship between sleep spindle measures and the ODI39. More recent evidence suggests that the recently developed hypoxic burden shows a stronger association with cognitive performance in OSA when compared to other traditional hypoxemia measures40,41. However, to date, no studies have directly examined the relationship between hypoxic burden and SSD. In our cohort, secondary exploratory analyses suggested that baseline and 6-month AHI, ODI and hypoxic burden were statistically significant yet weakly associated with SSD at a representative central derivation (all R² ≤ 0.15; Table S4), indicating that respiratory disturbance and nocturnal hypoxemia account for only a small proportion of the variability in SSD. This is consistent with our adjusted mixed-effects models (Tables 2 and 3 and Table S1), in which baseline AHI showed relatively modest positive coefficients for SSD change compared with the strong inverse effect of baseline SSD. Schönwald et al.42 examined the frequency and topographical distribution of sleep spindles during N2 sleep in patients with OSA. Using EEG recordings from frontal, central, and parietal regions, they compared patients with mild-to-moderate OSA with healthy controls. Their results showed that patients with OSA exhibited a higher proportion of slow spindles (< 13 Hz) and an overall slowing of spindle frequency, particularly in frontal regions. This baseline shift toward slower spindle activity may partly explain why, following treatment according to standard clinical practice, the increase in sleep spindle density observed in our study was more robust for fast spindles at central derivations. Notably, fast spindles have been more consistently associated with sleep-dependent memory consolidation and cognitive processing, which may enhance the potential clinical relevance of these findings43. The reported macrostructural sleep changes after treatment provide contextual information for interpretation of the spindle findings. As expected, patients with severe OSA had the most altered sleep architecture at baseline, with reduced N3 and REM sleep and markedly elevated arousal indices, ODI and hypoxic burden compared with the other groups. After six months of standard clinical practice, only the severe group showed a clear redistribution of sleep stages, with lower N2% and parallel increases in N3 and REM sleep. Enhancing sleep architecture, such as increasing the stability of N2 or N3 sleep, may facilitate spindle generation and improve spindle quality, although direct experimental evidence to this effect remains limited35,44. Similar to other reports, treatment of severe OSA did not significantly impact NREM sleep quantity, but modifications in SSD reveal subtle changes in brain activity not captured by standard sleep staging25. This pattern mirrors the severity-specific treatment-related increase in SSD and supports the view that spindle recovery occurs primarily when respiratory events are effectively controlled and sleep becomes more continuous. Considering the established influence of sex and age on sleep spindle characteristics, post hoc analyses were undertaken to disentangle their potential contribution to the observed results. Women generally exhibit a higher spindle number and density than men, particularly during N2 sleep45. However, most studies evaluating changes in SSD following treatment for OSA have been conducted predominantly in men24–28. In our study, 67% of participants were men presenting with suspected OSA, and men were more likely to have moderate to severe OSA (p < 0.001) (Table 1). In our study (Table S1), after controlling for age, BMI, and OSA severity, female sex was consistently associated with higher spindle density. This effect was particularly pronounced for both fast and slow spindles in frontal derivations (F3, F4), whereas increases in central derivations (C3, C4) were more modest, both across NREM sleep and more specifically during N2. These findings align with previous studies reporting that women exhibit a higher proportion of frontal sleep spindles compared with men46. However, a recent meta-analysis rated the certainty of this evidence as low due to methodological heterogeneity47. Hormonal factors, such as estrogen and progesterone, as well as differences in thalamocortical connectivity, have been proposed as potential mechanisms underlying the increased sigma activity observed in women48–51. Nevertheless, the evidence remains inconclusive, and controlled experimental studies are needed to elucidate the precise mechanisms. With increasing age, a progressive decrease in the number, density, and duration of sleep spindles occurs, along with a gradual increase in intra-spindle frequency up to approximately 60 years of age52. In this study, patients with severe OSA were significantly older than the remaining participants, and we also observed an age-related reduction in SSD. However, when the effect of the AHI on SSD was assessed while controlling for age, a significant association between AHI and SSD persisted, independent of age (Table S2 and S3). This observation supports the view that reduced spindle activity in severe OSA is not solely attributable to age-related changes. Strengths and limitations A major strength of the present study relies on the prospective longitudinal design in a relatively large, mixed-sex clinical cohort with full-night PSG and repeated assessment after approximately six months of standard OSA clinical management, rather than during CPAP titration or split-night protocols. At follow up, the cohort included patients treated with CPAP, mandibular advancement devices, positional therapy, and non-device-based interventions, reflecting routine clinical practice. The quantification of spindles during NREM sleep and specifically during N2 allowed for comparisons of stage-specific and overall changes in spindle activity. Unlike previous studies, which often relied on small samples, predominantly male participants and heterogeneous spindle scoring methods21, we applied a validated automated detection algorithm25 to all recordings, across four frontal and central EEG derivations, and quantified both slow and fast spindles under strict EEG quality criteria. The stratification by OSA severity allowed us to uncover that treatment-related increases in SSD are primarily concentrated in the cluster of patients with severe OSA, whereas milder groups show relatively preserved spindle activity and limited changes over time. In addition, the parallel analysis of sleep macrostructure and respiratory disturbance indices provides an integrated view of how improvements in sleep continuity and oxygenation relate to changes in sleep macro- and microarchitecture in routine clinical practice. Several limitations should be acknowledged. First, this was an observational study in a routine clinical cohort, so treatment allocation was not randomised and included different therapeutic modalities. This precludes causal inference regarding the effect of any specific treatment on SSD and implies that residual confounding by indication and differences in baseline disease burden cannot be fully excluded despite adjustments. The sample size, although larger than in previous studies, may still be insufficient to robustly detect sex-specific or age-by-severity interactions, and participants were recruited from a single sleep unit, which may limit generalizability to other settings and ethnic backgrounds. All patients attended the sleep unit due to clinical symptoms suggestive of OSA, so we did not include a healthy control group or a group with untreated OSA followed longitudinally. Therefore, changes over time are interpreted in relation to milder categories of OSA rather than a true control group without OSA. An additional limitation is that hypoxic burden was available only in a subset of participants (n = 52), which limits power and warrants cautious interpretation of analyses involving this metric. Conclusion In summary, adult patients with severe OSA showed reduced SSD at baseline and significant increases in SSD after approximately six months of standard clinical treatment, in parallel with improvements in sleep architecture and respiratory indices. In contrast, SSD remained relatively stable in patients with mild or moderate OSA at both time points. These findings indicate that SSD is particularly sensitive to treatment-related changes in sleep microstructure in severe OSA. Future studies should extend beyond sleep spindle density and incorporate additional aspects of sleep microarchitecture, such as whole night sigma power and its dynamics, as well as the temporal coupling between spindles and cortical slow oscillations, which plays a key role in sleep-dependent memory processing, and spindle characteristics in parietal regions, as these metrics may provide additional mechanistic insights. More detailed measures of treatment exposure and broader cognitive batteries will also be important to clarify the mechanisms linking OSA treatment, sleep spindle dynamics and cognitive outcomes and assist in determining whether SSD or related qEEG measures can be used as markers of sleep microstructural recovery in clinical practice. Future randomized or carefully monitored treatment studies with objective adherence data will be needed to evaluate more precisely whether spindle recovery tracks the exposure and effectiveness of OSA therapies. Electronic Supplementary Material Below is the link to the electronic supplementary material. Supplementary Material 1 Supplementary Material 1