Phenotypic Variability in Slow‐Wave Sleep in Depression: Associations With Clinical Profiles and Disorder Severity

ABSTRACT

Slow‐wave sleep (SWS; stage N3) is often reported as reduced in depression, yet variability across depressive phenotypes remains poorly characterised. This study aimed to determine whether N3 architecture—proportion, duration, and latency—identifies clinically distinct profiles in major depressive episode (MDE). Ninety‐seven adults with MDE underwent overnight polysomnography. N3 metrics (latency, duration, % of total sleep period (TSP), % of total sleep time (TST)) were analysed both dimensionally (correlations with clinical measures) and categorically (reduced, normal, or increased N3). Backward stepwise logistic regression identified variables distinguishing N3 categories, and sensitivity analyses assessed psychotropic medication effects. Shorter N3 latency correlated with poorer subjective sleep quality and greater anxiety and depressive severity. Extremes of N3 (%TSP) carried clinical burden: increased N3% associated with earlier hospitalisation, younger age at first suicide attempt, and higher daytime sleepiness; reduced N3% related to more hospitalisations and depressive episodes. Categorical contrasts showed increased N3 duration linked to higher anxiety and depressive symptom severity, reduced N3 (%TST) predicted more hospitalisations, while increased N3 (%TST) aligned with greater seasonality and morning chronotype. Men exhibited higher %N3 TST than women; bipolar depression had lower N3 than unipolar depression. A five‐factor model (episode count, chronotype, seasonality, age, sex) distinguished increased %N3 TST from normal levels (AUC = 0.863; sensitivity = 0.957; specificity = 0.714), explaining 30.8% of variance. Medication analyses revealed no major effects. N3 sleep is not a uniform biomarker in depression. Its proportion and duration delineate distinct, clinically meaningful profiles. Sleep‐based stratification may guide precision care.

1. Introduction

Depressive disorders affect up to 20% of the general population and represent a critical public health concern (WHO ²⁰¹⁷). They significantly impair psychosocial functioning and quality of life, and are closely linked to an elevated suicide risk (Chesney et al. 2014). The World Health Organization (WHO) estimated that over 300 million people were living with depression in 2015, an 18% increase since 2005, and projects that by 2030, depressive disorders will become the leading cause of disability worldwide (Mathers and Loncar 2006). Despite international efforts to identify biomarkers of depression, none have been implemented in clinical practice for diagnostic purposes or for predicting disease progression or therapeutic response. This challenge is partly due to the highly heterogeneous nature of depressive disorders, leading to highly variable responses to treatment. Indeed, only about one in three patients achieve remission after a first SSRI trial (Trivedi et al. 2006), highlighting the critical need for research into the physiological disruptions underlying depressive disorders to better define and differentiate their subtypes.

Sleep disturbances, observed in over 90% of the individuals with major depressive episode (MDE) (Geoffroy et al. 2018), hold particular promise as potential biomarkers for refining the classification of depressive subtypes. Polysomnography (PSG) has consistently shown sleep‐architecture abnormalities in depression such as shortened REM latency and increased REM density (Leitner et al. 2025) and, in meta‐analysis, reduced slow‐wave sleep (SWS) relative to controls (Palagini et al. 2013; Wichniak et al. ²⁰¹⁷). SWS has emerged as a promising subject of investigation for biomarker discovery in mood disorders. SWS, characterised by the N3 stage, is the deepest, most restorative non‐REM stage and supports energy recovery, declarative memory consolidation, synaptic down‐selection, immune modulation, and glymphatic clearance of neuro‐metabolites (Besedovsky et al. 2019; Mander et al. ²⁰¹⁷; Xie et al. ²⁰¹³). Because SWS can be experimentally amplified, it is also viewed as a potential therapeutic target, particularly in late‐life or treatment‐resistant depression (Hejazi et al. 2024). Converging evidence shows that SWS is not uniformly reduced in depression but varies by clinical subtype. A large meta‐analysis reports an overall reduction in deep sleep (including SWS) in MDE compared with controls, but it does not specifically show that this effect is ‘driven’ by melancholic cases or that SWS is ‘preserved’ in atypical depression (Baglioni et al. 2016). Population‐based PSG data published later indicate reduced delta power specifically in current melancholic MDE, with no robust alterations in other subtypes (Solelhac et al. 2023). Another PSG meta‐analysis, focused on seasonal affective disorder, synthesised 7 studies (~183 participants) and primarily found REM abnormalities (increased REM amount, reduced REM latency). It did not demonstrate a reliable SWS difference versus controls, and it does not establish a reduced SWS in non‐seasonal MDE contrast (Bertrand et al. 2021). In bipolar depression (BD), findings for SWS activity are heterogeneous; several studies do not show an increase in SWS during the depressive phase. Comparative work tends to emphasise differences in sleep continuity and REM markers rather than a consistent SWS increase in BD (Leseur et al. 2025; Zangani et al. ²⁰²⁰). Recent EEG work in treatment‐resistant depression reports reduced NREM (Non–rapid eye movement stages (N1–N3), marked by slow‐wave brain activity) delta power compared with healthy controls, supporting a deficit in slow‐wave activity consistent with reduced deep‐sleep homeostasis (Hejazi et al. 2024). Together, these findings emphasise that SWS alteration could be a phenotype‐specific rather than a universal biomarker of mood disorders. Beyond its descriptive features, disturbed SWS has well‐documented clinical consequences: greater symptom severity in youth with depression (Santangeli et al. 2017), poor or delayed ketamine response (Duncan et al. 2013), higher risk of relapse/recurrence in depression (Paykel 2008), impaired declarative memory and executive control (Zhang and Gruber 2019), heightened next‐day emotional reactivity (Ben Simon et al. 2019), low‐grade systemic inflammation (Strumberger et al. 2024) and disrupted overnight metabolic clearance (Van Hattem et al. 2025), elevated long‐term cardiometabolic risk (Kianersi et al. 2023), suicidal ideation and behaviour (Bernert et al. 2015).

Given these considerations, alterations in SWS appear to be associated with more severe depression profiles and with specific depressive subtypes. However, findings regarding SWS remain heterogeneous, likely due to methodological variability (e.g., scoring criteria, sample sizes) and clinical heterogeneity (e.g., melancholic, atypical, seasonal, or treatment‐resistant depression) (Pillai et al. 2011). A focused investigation of SWS across the depressive spectrum is warranted. Determining whether reduced (or, in some phenotypes, increased) duration of N3 sleep in minutes, or proportion of N3 sleep (N3% of Total Sleep Time; TST or Total Sleep Period; TSP) are associated with symptom severity, cumulative illness burden, suicidality, pharmacological treatment class or specific depressive subtypes such as seasonality could provide clinically actionable insights.

This study aimed to quantify N3 duration, N3% TST and N3% TSP in a well‐characterised cohort drawn from routine care of adults experiencing an MDE and to compare SWS metrics across clinically relevant subgroups including male versus female patients, suffering from bipolar versus unipolar depressive disorder, with or without seasonal pattern, treatment resistance, and suicidal ideation. Additionally, this investigation sought to explore associations between SWS and key clinical outcomes. By focusing analyses on PSG‐derived SWS measures, this work intends to clarify the potential of SWS as a biomarker in MDE and to support the development of personalised, biomarker‐driven therapeutic strategies in depression.

2. Method

2.1. Study Design

Patients were drawn from a routine care cohort from the Psychiatry, Sleep and Chronobiology Reference Center (ChronoS Center), the Sleep Center of the Physiology‐Functional Exploration Department, or during hospitalisation for MDE at the Department of Psychiatry and Addictology, Bichat Claude Bernard Hospital in Paris, from September 2022 to July 2025. Patients' medical histories were collected (treatments, recurrences, progression) and, with questionnaires and current MDE characteristics, comorbidities following DSM‐5‐TR criteria.

Patients were 18–70 years old, had mild to severe MDE according to DSM‐5‐TR criteria (American Psychiatric Association ²⁰¹³), and were affiliated to a social security scheme. Patients who did not exhibit clinical symptoms of depression according to the DSM‐5‐TR diagnostic criteria were not considered.

Patients received an information note explaining that their clinical data could be used for research purposes, with the possibility to decline (non‐opposition procedure). This clinical research project entitled ‘Étude des biomarqueurs du sommeil et des rythmes circadiens dans les troubles psychiatriques « Som‐Psy »’ was approved (N°CER‐2020–56) by the Ethical Review Committee for Biomedical Research Projects ‘Comité d'Evaluation de l'Ethique des projets de Recherche Biomédicale (CEERB) Paris Nord’ (Institutional Review Board—IRB 00006477 of HUPNVS, Paris 7 University, AP‐HP).

2.2. Sociodemographic, Sleep and Psychiatric Clinical Parameters

Sociodemographic characteristics—gender, age and body mass index (BMI)—were collected. Clinical parameters assessed in routine care included the number of depressive episodes, hospitalisations, treatments, and severity of depressive and anxiety symptoms using the Hospital Anxiety and Depression Scale (HADS) (Zigmond and Snaith 1983) (HAD‐D for depression severity and HAD‐A for anxiety severity) and the Quick Inventory of Depressive Symptomatology—Self‐Report 16 items (QIDS‐SR16) (Reilly et al. 2015). The associated cut‐offs were: HAD ≥ 10 (clinically significant anxiety and/or depressive symptoms) and QIDS‐SR16: 6 to 10: mild depression, 11 to 15: moderate depression, 16 to 20: severe depression, 21 to 27: very severe depression.

Sleep and circadian patterns were assessed using self‐report questionnaires: Pittsburgh Sleep Quality Index (PSQI) (Buysse et al. 1989), which assesses sleep quality (PSQI; poor sleep quality > 5); the Epworth Sleepiness Scale (ESS, excessive sleepiness ≥ 11) (Johns 1991); the Insomnia Severity Index (ISI; mild insomnia ≥ 8) (Morin et al. 2011), the Hypersomnia Severity Index (HSI; mild hypersomnia ≥ 10) (Kaplan et al. 2019) and the Morningness‐Eveningness Questionnaire (MEQ, ≥ 70 definitely morning, 59–69 moderately morning, 42–58 neither, 31–41 moderately evening, ≤ 30 definitely evening chronotype) (Taillard et al. 2004). Seasonality was measured using the Global Seasonality Score (GSS) of the Seasonal Pattern Assessment Questionnaire (SPAQ) ranging from 0 (no seasonality) to 24 (extreme seasonality) (Reynaud et al. 2021).

2.3. Objective Sleep and Circadian Rhythms Measure Using Polysomnography

In the context of routine care, patients had one night of PSG using a standard system V‐PSG; Morpheus (Micromed, Treviso, Italy); Noxturnal (ResMed, Bella Vista, NSW, Australia); Alice 6 (Philips‐Respironics, Suresnes, France). PSG assessed 28 measures of sleep architecture, including duration, latency, and percentage of each sleep stage (N1, N2, N3, REM), total sleep time (TST), time in bed, and total sleep period (TSP, TST including micro‐awakenings). Latencies were defined as follows: N1 latency—from intended sleep to sleep onset; N2 and N3 latencies—from sleep onset to the onset of N2 and N3, respectively. Additional metrics included REM latency and duration, wake after sleep onset (WASO), sleep efficiency, nocturnal awakenings, total awakening index, apnea/hypopnea index (AHI), total arousal index, and the Kupfer Index (age + REM latency), a proposed depression biomarker reflecting sleep continuity and architecture disruption. A specific focus was placed on N3 (SWS), analysing N3 duration, N3 latency, and the duration of the first N3 episode.

Categorical analyses (low, normal, high) were based on established PSG cut‐offs (AASM, Rechtschaffen and Kales), widely used in sleep research, including psychiatric populations (Allan Rechtschaffen ¹⁹⁶⁸; Iber et al. ²⁰⁰⁷).

2.4. Statistics

Correlations between MDE and polysomnographic (PSG) variables were assessed using Pearson's r or Spearman's ρ. Categorical variables were reported as counts and percentages, continuous variables as mean ± SD. Normality was checked via Q–Q plots and the Shapiro–Wilk test (p > 0.05), and homoscedasticity via Levene's test (p > 0.05). N3 sleep metrics (latency, duration, first N3 duration, %TST, %TSP) were compared across sex, depression polarity (bipolar vs. unipolar according to the DSM‐5‐TR criteria), seasonal vs. non‐seasonal depression (based on a cut‐off score of 11 on the Global Seasonality Score from the SPAQ), and treatment‐resistant vs. non‐resistant depression using Student t‐tests or Mann–Whitney tests as appropriate. Clinical, sleep, and psychiatric variables were compared between patients with decreased (< 90 min), normal (90–120 min), or increased (> 120 min) N3 duration, and between decreased (< 15%), normal (15%–20%), or increased (> 20%) N3% TST (Ohayon et al. 2004; Redline et al. ²⁰⁰⁴). Secondly, variables differing between groups at p < 0.05 were entered into backward stepwise logistic regression (BSLR) to assess the combination of these selected variables that best classified individuals with increased, normal or reduced N3 stage. In addition, age and sex were included into the model as covariates because both exert robust, well‐described effects on slow‐wave sleep: SWS declines across the adult lifespan, and systematic sex differences in N3 have been reported in large cohorts showing that men tend to have less N3/SWS than women (Ohayon et al. 2004; Redline et al. ²⁰⁰⁴).

The sensitivity, specificity, test accuracy, area under the curve (AUC), and Youden's index were calculated to evaluate the model's predictive performance. The receiver operating characteristic (ROC) curve was used to visualise the model's discriminative ability. We also evaluated the significance of this global model and assessed its variance, including the R ² value.

Sensitivity analyses were conducted to test the robustness of our findings and account for potential medication effects on N3 sleep. For the five most commonly prescribed drug classes (antidepressants, benzodiazepine anxiolytics, hypnotics, mood stabilisers, and antipsychotics), participants were grouped by exposure (yes/no) on the night of polysomnography. Key N3 metrics (latency, absolute duration, %TSP, %TST) were compared between exposed and unexposed patients using the same statistical approach as for depression subtype comparisons.

All statistical analyses were performed using JAMOVI software (version 2.4.14), where a p‐value of < 0.05 was used as the significance threshold.

Based on the article of Solelhac et al. (2023), power analysis was conducted in G*Power (v.3.1) using the Exact test for bivariate correlations, with α = 0.05 (two‐tailed) and power = 0.80 (Solelhac et al. 2023). This resulted in a required sample size of N = 46 per correlation, which is consistent with our group sizes (n = 40) and supports that our study is adequately powered to detect associations of moderate to large magnitude.

3. Results

3.1. Clinical Sleep and Socio‐Demographic Characteristics

The study sample consisted of 97 participants (43 men, 54 women), with a mean age of 45.5 (±13.2) years and a mean BMI of 25.5 (±5.8 kg/m²). Clinical history showed a mean age of 26.0 (± 12.1) years at first depressive episode, 32.5 (± 12.7) years at first psychiatric hospitalisation, 4.5 (±5.8) depressive episodes, 1.9 (±2.7) hospitalisations, 24.9 (±17.1) years at first suicide attempt and 0.9 (±1.3) suicide attempts. Medication use is detailed in Table 1. Affective symptoms indicated clinically meaningful anxiety and depression: HAD‐A 12.4 (±3.7), HAD‐D 11.9 (±4.4), HADS total score 24.3 (±7.2); QIDS‐SR16 15.8 (±5.4). Sleep and circadian assessments showed disturbances: ISI 18.4 (±4.8), PSQI 12.7 (±3.3), ESS 8.9 (±5.3), HSI 17.8 (±4.1), and MEQ 47.7 (±10.8) (intermediate chronotype). Detailed results are in Table 1.

TABLE 1.

Socio‐demographic and clinical characteristics (N = 97) of patients with major depressive episode (MDE) drawn from routine care cohort from the project Sompsy.

Variables	Means (±SD) or n (%)
Gender
Men	43 (44.3%)
Woman	54 (55.7%)
Age (years)	45.5 (±13.2)
BMI (kg/m2)	25.5 (±5.8)
Age of first depressive episode	26 (±12.1)
Number of depressive episodes	4.5 (±5.8)
Age of first hospitalisations	32.5 (±12.7)
Number of hospitalisations	1.9 (±2.7)
Age of first suicidal attempt	24.9 (±17.1)
Number of suicidal attempts	0.9 (±1.27)
HAD‐A	12.4 (±3.7)
HAD‐D	11.9 (±4.4)
HADS	24.3 (±7.2)
QIDS—SR16	15.8 (±5.4)
MEQ	47.7 (±10.8)
SPAQ GSS (item 11)	10.9 (±6.22)
HSI	17.8 (±4.1)
ESS	8.9 (±5.3)
ISI	18.4 (±4.8)
PSQI	12.7 (±3.3)

Abbreviations: BMI, body mass index; ESS, Epworth sleepiness scale; HADS, hospital anxiety and depression scale (D: depression, A: anxiety); HSI, hypersomnia severity index; ISI, insomnia severity index; MEQ, morningness eveningness questionnaire; PSQI, Pittsburgh sleep quality index; QIDS, quick inventory depressive symptoms 16‐item scale; SPAQ GSS, seasonal pattern assessment questionnaire, global seasonality score.

3.2. Descriptive Statistics of Polysomnographic Data

Stage N1 lasted 38.3 ± 32.8 min (7.0% ± 6.1% of TSP; 10.5% ± 8.8% of TST) with a latency of 14.6 ± 27.6 min. Stage N2 was dominant, lasting 189.0 ± 73.9 min (45.8% ± 12.4% TSP; 49.8% ± 13.1% TST) with a latency of 8.5 ± 11.3 min. Stage N3 (SWS) accounted for 86.2 ± 39.5 min (19.7% ± 10.0% TSP; 23.9% ± 10.5% TST), with a first episode of 41.0 ± 30.3 min and a latency of 46.8 ± 69.1 min (latency of 45.8 ± 49.8 min, after exclusion of an outlier value (536.5 min)).

Non‐REM sleep totalled 311 ± 84.8 min. REM sleep appeared after 159 ± 116 min and lasted 65.4 ± 43.3 min (14.1% ± 8.4% TSP; 16.4% ± 9.0% TST). Kupfer Index averaged 199 ± 108. Sleep continuity showed 48.5 ± 51.8 min intra‐sleep wakefulness, 57.0 ± 55.1 min WASO, 25.3 ± 29.3 nocturnal awakenings, a total arousal index of 12.6 ± 4.4, and 81.5 ± 76.7 micro‐arousals. AHI was 14.4 ± 14.4. TST was 384 ± 104 min, TSP 437 ± 91 min, time in bed 759 ± 171 min, and sleep efficiency 77.8% ± 14.2%. Detailed results are in Table S1.

3.3. Correlations Between N3 Sleep and Sleep and Psychiatric Characteristics

We conducted correlation analyses between N3 sleep metrics and socio‐demographic, sleep and psychiatric data in the 97 complete sample of patients. N3 latency negatively correlated with anxiety (HAD‐A: Rho = −0.301, p = 0.027; HADS: Rho = −0.269, p = 0.049) and depressive symptoms (QIDS‐SR16: Rho = −0.329, p = 0.024), indicating faster transition into N3 with greater symptom severity. N3 duration (min) negatively correlated with the number of hospitalisations (Rho = −0.351, p = 0.009), which suggests that the higher the number of hospitalisations, the shorter the duration of N3 sleep. Regarding N3% TSP, a significant negative correlation was observed with age at first suicide attempt (Rho = −0.784, p < 0.001), indicating that a younger age at the time of the first attempt was associated with a higher N3% TSP. In addition, N3% TSP was negatively correlated with both the age at first hospitalisation (Rho = −0.396, p = 0.045) and the number of hospitalisations (Rho = −0.378, p = 0.018). These findings suggest that earlier psychiatric hospitalisation was linked to increased N3% TSP, and that a greater number of hospitalisations was associated with shorter N3% TSP.

Finally, when examining N3% TST, a significant negative correlation was identified with number of depressive episodes (r = −0.291, p = 0.048), suggesting reduced deep sleep proportion in individuals with more recurrent depression.

Regarding clinical subjective sleep data, a significant negative correlation was found between sleep quality (PSQI) and N3 latency (Rho = −0.363, p = 0.0009), indicating that poorer subjective sleep quality was associated with a shorter latency to N3 onset. For N3% TSP, excessive daytime sleepiness (ESS) was positively correlated (Rho = 0.292, p = 0.044), indicating that higher daytime sleepiness was associated with increased N3% TSP.

No statistically significant associations were found between first N3 episode duration and any of the clinical variables. Results are shown in Table 2 and Figure 1.

TABLE 2.

Correlation matrix showing the relationships between clinical and socio‐demographic characteristics and N3 variables measured by polysomnography.

Variables	Rho Spearman N3 latency	p	Rho Spearman N3 duration	p	Rho Spearman N3% TSP	p	r Pearson N3% TST	p	Rho Spearman 1st N3 length	p
Age	0.111	0.345	−0.171	0.108	−0.205	0.157	−0.163	0.127	0.074	0.646
BMI	0.117	0.409	0.117	0.365	0.172	0.253	0.148	0.252	−0.308	0.163
ISI	−0.062	0.612	−0.166	0.137	−0.109	0.469	−0.121	0.279	0.226	0.172
PSQI	−0.363	0.009	−0.072	0.579	−0.029	0.850	−0.002	0.986	0.266	0.266
ESS	−0.095	0.430	0.202	0.063	0.292	0.044	0.145	0.187	−0.075	0.650
HAD‐A	−0.301	0.027	0.107	0.401	0.132	0.369	0.019	0.882	0.115	0.592
HAD‐D	−0.714	0.208	0.067	0.598	−0.095	0.520	−0.112	0.376	0.025	0.908
HADS	−0.269	0.049	0.103	0.417	−0.001	0.997	−0.067	0.596	0.113	0.599
SPAQ GS	−0.177	0.143	0.091	0.413	−0.052	0.729	0.022	0.843	0.028	0.866
MEQ	0.111	0.446	0.917	0.536	0.729	0.400	0.073	0.591	−0.254	0.242
QIDS‐S16	−0.329	0.024	0.126	0.347	0.022	0.890	−0.031	0.820	0.286	0.221
HSI	−0.156	0.317	0.102	0.468	0.135	0.399	0.066	0.640	0.267	0.318
Age of first MDE	0.023	0.877	−0.075	0.577	−0.039	0.808	0.052	0.701	0.007	0.978
Number of MDE	0.081	0.625	−0.165	0.267	−0.282	0.101	−0.291	0.048	0.157	0.592
Age of first hospitalisation	0.355	0.070	−0.147	0.413	−0.396	0.045	−0.232	0.195	0.400	0.517
Number of hospitalisation	0.215	0.166	−0.351	0.009	−0.378	0.018	−0.216	0.117	−0.164	0.610
Age of first suicidal attempt	−0.079	0.764	−0.210	0.336	−0.784	0.001	−0.020	0.927	0.600	0.417
Number of suicidal attempts	−0.240	0.126	−0.148	0.307	−0.166	0.314	−0.185	0.198	0.201	0.409

Note: Value in bold indicates a statistically significant difference with a p‐value of less than 0.05.

Abbreviations: BMI, body mass index; ESS, Epworth sleepiness scale; HADS, hospital anxiety and depression scale; HSI, hypersomnia severity index; ISI, insomnia severity index; MDE, major depressive episode; MEQ, morningness eveningness questionnaire; PSQI, Pittsburgh sleep quality index; QIDS, quick inventory depressive symptoms 16‐item scale; SPAQ GSS, seasonal pattern assessment questionnaire, global seasonality score.

FIGURE 1.

Scatter graph of associations between N3 sleep metrics and clinical variables in depression. (a) Number of hospitalisations across N3% TST. (b) Number of major depressive episode across N3% TST. (c) HAD‐A across N3 latency*. (d) HADS across N3 latency*. (e) PSQI across N3 latency*. (f) QIDS‐16 across N3 latency*. (g) Number of hospitalisations across N3 duration. (h) HAD‐A across N3 duration. (i) HADS across N3 duration. (j) QIDS‐SR16 across N3 duration. (k) Age of first hospitalisation across N3% TSP. (l) Number of hospitalisations across N3% TSP. (m) Age of first suicidal attempt across N3% TSP. (n) ESS across N3% TSP. *Note: Additional analyses were conducted after removal of the N3 latency outlier (536.5 min; see Table S2). ESS, Epworth Sleepiness Scale; HADS‐A/D, Hospital Anxiety and Depression Scale—Anxiety/Depression; PSQI, Pittsburgh Sleep Quality Index; QIDS‐SR16, Quick Inventory of Depressive Symptomatology—Self‐Report 16; TSP, total sleep period; TST, total sleep time.

Correlation analyses were also performed without the N3 latency outlier. The results are detailed in Table S2. Similar findings were observed, with N3 latency remaining negatively correlated with anxiety and depressive symptoms, indicating a faster transition into N3 sleep with greater symptom severity. In addition, poorer subjective sleep quality was associated with a shorter latency to N3 onset.

3.4. Comparison of Mean N3 Stage Data Between Various Depressive Subtypes

3.4.1. Study of the Differences in Mean N3 Metrics in Patients Diagnosed With MDE (N = 66) With Unipolar Depression (N = 43) and Bipolar Depression (N = 23)

Among 66 patients with polarity data (23 bipolar depression (BD), 43 unipolar depression (UD)), patients with BD had shorter N3 duration (70.66 ± 30.53 min vs. 92.91 ± 37.93 min; U = 318.5, p = 0.018) and lower N3 proportion for both TSP (14.55% ± 6.69% vs. 22.22% ± 10.38%; U = 131.5, p = 0.004) and TST (17.84% ± 7.71% vs. 26.72% ± 10.38%; t = 3.60, p < 0.001). No difference in N3 latency (Table 3; Tables S3 and S4, the latter excluding the N3 latency outlier and Figure S2).

TABLE 3.

Comparison of N3 sleep metrics in major depressive episode (MDE) by mood subtype (bipolar (N = 23) to unipolar (N = 43)), seasonality (Non seasonal (N = 37) to Seasonal (N = 46)), sex (Men (N = 41) to Women (N = 53)), treatment resistance (Resistant (N = 15) to Nonresistant (N = 51)) and suicidal ideation (Yes (N = 14) to no (N = 41)).

Variable	Unipolar (N = 43)	Bipolar (N = 23)	p	N‐Seasonal (N = 37)	Seasonal (N = 46)	p	Male (N = 41)	Women (N = 53)	p	Resistance (N = 15)	No resistance (N = 51)	p	No ideation (N = 41)	Ideation (N = 14)	p
N3 Latency (min)	32.5 ± 19.45	67.0 ± 127.69	0.964	42.5 ± 37.23	47.0 ± 84.23	0.369	41.2 ± 30.03	51.1 ± 89.90	0.249	36.3 ± 45.52	44.9 ± 78.57	0.448	38.4 ± 32.30	75.2 ± 162.18	0.585
N3 Length (min)	92.9 ± 37.93	70.6 ± 30.53	0.018	85.5 ± 37.10	88.8 ± 44.00	0.657	89.9 ± 34.23	83.3 ± 43.18	0.322	95.9 ± 41.58	81.8 ± 35.18	0.166	92.5 ± 35.87	71.8 ± 29.90	0.080
N3 TSP (%)	22.2 ± 10.38	14.5 ± 6.69	0.004	21.2 ± 9.83	18.1 ± 10.93	0.496	20.5 ± 9.06	19.1 ± 10.76	0.686	21.8 ± 12.82	19.3 ± 9.04	0.622	21.3 ± 10.31	16.9 ± 10.36	0.301
N3 TST (%)	26.7 ± 10.38	17.8 ± 7.71	< 0.001	24.3 ± 10.56	23.5 ± 10.96	0.750	26.4 ± 9.93	21.8 ± 10.59	0.036	25.5 ± 12.04	23.1 ± 9.92	0.425	25.1 ± 10.22	19.9 ± 10.20	0.108

Note: Value in bold indicates a statistically significant difference with a p‐value of less than 0.05.

Abbreviations: TSP, total sleep period; TST, total sleep time.

3.4.2. Comparison of N3 Metrics Between Patients Suffering From MDE (N = 83) With Seasonal Pattern (+SP, N = 46) and Patients Suffering From MDE Without Seasonal Pattern (−SP, N = 37)

Among 83 patients with SPAQ data (46 GSS ≥ 11 (+SP), a GSS < 11 (−SP)), no significant differences in N3 metrics were observed (Table 3; Tables S3 and S4, the latter excluding the N3 latency outlier).

3.4.3. Comparison of N3 Stage Metrics of Patients Suffering From MDE (N = 94) Between Men (N = 41) and Women (N = 53)

Among 94 patients with sex and N3 data (41 men, 53 women), women had lower N3% TST than men (21.89% ± 10.59% vs. 26.45% ± 9.93%, t = 2.13, p = 0.036); other N3 parameters were comparable (Table 3; Tables S3 and S4 the latter excluding the N3 latency outlier).

3.4.4. Comparison of N3 Stage Metrics Between Patients Suffering From MDE (N = 66) With Treatment Resistance (N = 15) and Without Treatment Resistance (N = 51)

Among 66 patients (15 with MDE with treatment resistance, 51 without), no significant differences in N3 sleep parameters were observed (Table 3; Tables S3 and S4 the latter excluding the N3 latency outlier).

3.4.5. Comparison of N3 Stage Metrics in Patients Diagnosed With MDE (N = 55) Between Those With Suicidal Ideation (N = 14) and Those Without Suicidal Ideation (N = 41)

Among the data from the 55 patients, (14 with suicidal ideation, 41 without), no significant differences in N3 sleep parameters were observed (Table 3; Tables S3 and S4 the latter excluding the N3 latency outlier).

3.5. Comparison of Clinical, Sleep and Psychiatrics Characteristics Between Patients Suffering From MDE (N = 97) With Increased N3 Duration (> 120 min, N = 15), Normal N3 Duration (90 < N3 Duration < 120 min, N = 21) and Reduced N3 Duration (< 90 min, N = 53)

3.5.1. Comparison of Clinical, Sleep and Psychiatrics Characteristics Between Patients Suffering From MDE (N = 36) With Increased N3 Duration (> 120 min, N = 15) and Patients With Normal N3 Duration (90 < N3 Duration < 120 min, N = 21)

Among the 36 patients for whom we had the corresponding data, 15 had N3 duration > 120 min and 21 had normal N3 duration (90–120 min). No significant differences were observed between the two groups regarding clinical, sleep, or psychiatric characteristics. Both groups showed moderate anxiety and depressive symptoms, mild seasonality, moderate insomnia with poor sleep quality, subjective hypersomnia, moderate daytime sleepiness, and a neutral chronotype (Tables 4 and 5).

TABLE 4.

Comparison of N3 stage metrics by N3 duration: Increased (> 120 min, N = 15) to normal (90–120 min, N = 21), normal (90–120 min, N = 21) to reduced (< 90 min, N = 53), and increased (> 120 min, N = 15) to reduced (< 90 min, N = 53).

Variable	Normal N3 (N = 21)	Increased N3 (N = 15)	p	Normal N3 (N = 21)	Reduced N3 (N = 53)	p	Reduced (N = 53)	Increased (N = 15)	p
AGE	45.8 ± 12.80	38.8 ± 14.95	0.162	45.8 ± 12.80	47.094 ± 12.66	0.792	47.094 ± 12.66	38.800 ± 14.953	0.056
BMI	27.68 ± 7.81	24.2 ± 4.59	0.307	27.68 ± 7.81	24.695 ± 4.64	0.215	24.695 ± 4.64	24.252 ± 4.597	0.607
ISI	18.30 ± 4.35	17.0 ± 4.15	0.431	18.30 ± 4.35	18.681 ± 5.15	0.630	18.681 ± 5.15	17.000 ± 4.158	0.125
PSQI	12.56 ± 3.43	12.0 ± 3.10	0.713	12.56 ± 3.43	12.811 ± 3.32	0.792	12.811 ± 3.32	12.000 ± 3.109	0.554
ESS	10.76 ± 5.04	9.2 ± 4.91	0.375	10.76 ± 5.04	8.260 ± 5.55	0.080	8.260 ± 5.55	9.214 ± 4.917	0.563
HAD‐A	12.24 ± 4.22	14.5 ± 3.50	0.201	12.24 ± 4.22	11.821 ± 3.39	0.698	11.821 ± 3.39	14.500 ± 3.505	0.049
HAD‐D	11.50 ± 4.36	14.3 ± 6.45	0.230	11.50 ± 4.36	11.385 ± 3.99	0.884	11.385 ± 3.99	14.375 ± 8.983	0.268
HADS	23.53 ± 7.77	28.8 ± 8.98	0.140	23.53 ± 7.77	23.205 ± 6.18	0.868	23.205 ± 6.18	28.875 ± 8.983	0.034
SPAQ GSS	11.57 ± 6.05	12.6 ± 7.09	0.635	11.57 ± 6.05	10.375 ± 5.90	0.445	10.375 ± 5.90	12.643 ± 7.099	0.232
MEQ	49.00 ± 8.64	47.3 ± 10.88	0.694	49.00 ± 8.64	47.697 ± 12.09	0.702	47.697 ± 12.09	47.375 ± 10.888	0.946
HSI	18.38 ± 4.06	17.6 ± 3.98	0.718	18.38 ± 4.06	17.484 ± 4.41	0.504	17.484 ± 4.41	17.667 ± 3.983	0.926
QIDS—SR16	16.69 ± 7.01	18.7 ± 3.99	0.451	16.69 ± 7.01	14.618 ± 4.71	0.223	14.618 ± 4.71	18.750 ± 3.991	0.027
Number of MDE	3.83 ± 5.34	2.1 ± 0.99	0.577	3.83 ± 5.34	5.519 ± 6.73	0.528	5.519 ± 6.73	2.125 ± 0.991	0.379
Number of hospitalisations	2.64 ± 4.50	0.6 ± 1.06	0.319	2.64 ± 4.50	1.875 ± 1.96	0.514	1.875 ± 1.96	0.625 ± 1.061	0.059
Number of suicidal attempts	1.06 ± 1.56	0.0 ± 0.0	0.167	1.06 ± 1.56	0.867 ± 1.17	0.807	0.867 ± 1.17	0.000 ± 0.000	0.175

Note: Value in bold indicates a statistically significant difference with a p‐value of less than 0.05.

Abbreviations: BMI, body mass index; ESS, Epworth sleepiness scale; HADS‐A/D, hospital anxiety and depression scale—anxiety/depression; HSI, hypersomnia severity index; Increase N3, Increased N3 duration = > 120 min; ISI, insomnia severity index; MEQ, morningness eveningness questionnaire; Normal N3, Normal N3 duration = 90–120 min; PSQI, Pittsburgh sleep quality index; QIDS‐SR16, quick inventory of depressive symptomatology—self‐report 16; SPAQ GSS, seasonal pattern assessment questionnaire—global seasonality score.

TABLE 5.

Mapping N3 sleep phenotypes to clinical burden and subgroup differences in MDE (N = 97).

N3 metric	Outcome/comparator	Phenotype
↓ N3 latency (min)	Subjective sleep quality (PSQI)	↓ Sleep quality
	Anxiety severity (HAD‐A)	↑ Anxiety
	Depressive severity (QIDS‐SR16)	↑ Depression severity
↑ N3 TSP (%)	Age at first psychiatric hospitalisation	Earlier hospitalisation
	Age at first suicide attempt	Younger age at attempt
	Daytime sleepiness (ESS)	Excessive daytime sleepiness
↓ N3 TSP (%)	Number of hospitalisations	↑ Number of hospitalisations
↓ N3 TST (%)	Number of depressive episodes	↑ Number of MDE
N3 duration (min): Increased (vs reduced)	Anxiety severity (HAD‐A)	↑ Anxiety severity
	Depression Severity (QIDS‐SR16)	↑ Depressive symptom severity
	HADS	↑ Anxiety and depressive symptom severity
N3 TST (%): Increased (vs normal)	Seasonality (SPAQ GSS)	↑ Seasonality
	Chronotype (MEQ)	↑ Morning Chronotype
	Number of depressive episodes	↓ Number of MDE
	Number of hospitalisations	↓ Number of hospitalisations
N3 TST (%): Increased (vs reduced)	Number of hospitalisations	↓ Number of hospitalisations

N3 metric across categories	Outcome/comparator	Phenotype
N3 TST (%)	Sex (men vs. women)	Male > Female
N3 duration (min)	Mood subtype (BD vs. UD)	BD < UD
N3 TST (%)	Mood subtype (BD vs. UD)	BD < UD

Abbreviations: BD, bipolar depression; ESS, Epworth sleepiness scale; HAD‐A/D, hospital anxiety and depression scale—anxiety/depression; MDE, major depressive episode; MEQ, morningness eveningness questionnaire; PSQI, Pittsburgh sleep quality index; QIDS‐SR16, quick inventory of depressive symptomatology—self‐report 16; SPAQ GSS, seasonal pattern assessment questionnaire—global seasonality score; TSP, total sleep period; TST, total sleep time; UD, unipolar depression.

3.5.2. Comparison of Clinical, Sleep and Psychiatrics Characteristics Between Patients Suffering From MDE (N = 74) With Reduced N3 Duration (< 90 min, N = 53) and Patients With Normal N3 Duration (90 < N3 Duration < 120 min, N = 21)

Among the 74 patients for whom we had the corresponding data, 53 had N3 duration < 90 min and 21 had normal N3 duration (90–120 min). No significant differences were observed between the two groups regarding clinical, sleep, or psychiatric characteristics. Both groups showed moderate anxiety and depressive symptoms, mild seasonality, moderate insomnia with poor perceived sleep quality, mild daytime sleepiness and a neutral chronotype (Tables 4 and 5).

3.5.3. Comparison of Clinical Variables (N = 68) Between Patients With Reduced N3 Duration (min) (N = 53) and Increased N3 Duration (min) (N = 15)

The N3 duration‐increased group (n = 15) showed a higher anxiety severity on HAD‐A (p = 0.049), higher HADS (p = 0.034), and higher depressive symptom severity on QIDS‐SR16 (p = 0.027) compared to the N3 duration‐reduced group (n = 53). No other significant difference was observed between the two groups regarding the other variables. Results are shown in Tables 4 and 5 and Figure 1h–j and in Figure S1.

3.6. Comparison of Clinical, Sleep and Psychiatrics Characteristics Between Patients Suffering From MDE (N = 89) With Increased N3% TST (> 20%, N = 53), Normal N3% TST (15% < N3 Duration < 20%, N = 19) and Reduced N3% TST (< 15%, N = 17)

3.6.1. Comparison of Clinical, Sleep and Psychiatrics Characteristics Between Patients Suffering From MDE (N = 72) With Increased N3% TST (> 20%, N = 53) and Patients With Normal N3% TST (15% < N3 Duration < 20%, N = 19)

Among the 72 patients for whom we had the corresponding data, those with N3% TST > 20% (n = 53) had a lower number of psychiatric hospitalisations (p = 0.021), a lower lifetime number of depressive episodes (p = 0.015), higher seasonality (p = 0.026), and a stronger morning chronotype (p = 0.019) compared to patients with N3% TST 15%–20% (n = 19). No other significant differences were observed (Tables 5 and 6 and Figures 1a,b and 2).

TABLE 6.

Comparison of clinical variables by N3% TST: Normal (N = 19) to increased (N = 53), normal (N = 19) to reduced (N = 17) and increased (N = 53) to reduced (N = 17).

Variable	Normal N3% TST (N = 19)	Increased N3% TST (N = 53)	p	Normal N3% TST (N = 19)	Reduced N3% TST (N = 17)	p	Reduced N3% TST (N = 17)	Increased N3% TST (N = 53)	p
AGE	46.00 ± 12.87	43.85 ± 13.37	0.569	46.00 ± 12.87	49.59 ± 13.33	0.418	49.59 ± 13.04	43.85 ± 13.37	0.108
BMI	25.44 ± 7.92	25.98 ± 6.02	0.68	25.44 ± 7.92	24.03 ± 2.42	0.916	24.03 ± 2.42	25.98 ± 6.02	0.407
ISI	19.40 ± 4.26	18.15 ± 4.59	0.427	19.40 ± 4.26	17.60 ± 5.97	0.603	17.60 ± 5.97	18.15 ± 4.59	0.922
PSQI	12.50 ± 3.12	12.74 ± 3.45	0.828	12.50 ± 3.12	12.45 ± 3.17	0.973	12.45 ± 3.17	12.74 ± 3.45	0.804
ESS	7.63 ± 5.02	10.08 ± 5.22	0.083	7.63 ± 5.02	7.44 ± 5.81	0.916	7.44 ± 5.73	10.08 ± 5.22	0.054
HAD‐A	11.71 ± 3.47	12.61 ± 4.04	0.462	11.71 ± 3.47	11.73 ± 2.53	0.992	11.73 ± 2.53	12.61 ± 4.04	0.494
HAD‐D	12.78 ± 3.75	11.70 ± 4.85	0.198	12.78 ± 3.75	10.82 ± 3.95	0.187	10.82 ± 3.95	11.70 ± 4.85	0.747
HADS	24.50 ± 6.89	24.23 ± 7.78	0.909	24.50 ± 6.89	22.55 ± 5.07	0.439	22.55 ± 5.07	24.23 ± 7.78	0.502
SPAQ GSS	8.05 ± 6.45	11.88 ± 5.84	0.026	8.05 ± 6.45	11.69 ± 6.12	0.108	11.69 ± 6.12	11.88 ± 5.84	0.910
MEQ	41.84 ± 13.05	50.08 ± 9.40	0.019	41.84 ± 13.05	48.89 ± 10.89	0.199	48.89 ± 10.89	50.08 ± 9.40	0.743
HSI	18.45 ± 3.14	18.32 ± 4.34	0.927	18.45 ± 3.14	15.55 ± 4.32	0.08	15.55 ± 4.32	18.32 ± 4.34	0.075
QIDS SR16	15.50 ± 5.93	16.14 ± 5.85	0.731	15.50 ± 5.93	14.67 ± 3.00	0.702	14.67 ± 3.00	16.14 ± 5.85	0.471
Number of MDE	8.70 ± 6.96	2.51 ± 2.11	0.015	8.70 ± 6.96	6.50 ± 9.96	0.394	6.50 ± 9.96	2.51 ± 2.11	0.407
Number of hospitalisations	2.54 ± 2.46	1.34 ± 2.96	0.021	2.54 ± 2.46	2.82 ± 2.36	0.764	2.82 ± 2.36	1.34 ± 2.96	0.005
Number of suicidal attempts	0.80 ± 1.48	0.78 ± 1.18	0.651	0.80 ± 1.48	1.38 ± 1.51	0.476	1.38 ± 1.51	0.78 ± 1.18	0.351

Note: Value in bold indicates a statistically significant difference with a p‐value of less than 0.05.

Abbreviations: BMI, body mass index; ESS, Epworth sleepiness scale; HADS‐A/D, hospital anxiety and depression scale—anxiety/depression; HSI, hypersomnia severity index; ISI, insomnia severity index; MEQ, morningness eveningness questionnaire; PSQI, Pittsburgh sleep quality index; QIDS‐SR16, quick inventory of depressive symptomatology—self‐report 16; SPAQ GSS, seasonal pattern assessment questionnaire—global seasonality score; TST, total sleep time.

FIGURE 2.

Comparison of clinical, sleep and psychiatric characteristics between patients with normal and increased N3% of Total Sleep Time (TST). BMI, body mass index; ESS, Epworth sleepiness scale; HAD, Hospital Anxiety and Depression scale. HSI, Hypersomnia Severity Index; ISI, Insomnia Severity Index; MDE, Major Depressiv Episode; MEQ, Morningness Eveningness Questionnaire; PSQI, Pittsburgh Sleep Quality Index; QIDS, Quick Inventory Depressive Symptoms 16‐item scale; SPAQ GSS, Seasonal Pattern Assessment Questionnaire, Global Seasonality Score.

3.6.2. Comparison of Clinical, Sleep and Psychiatrics Characteristics Between Patients Suffering From MDE (N = 36) With Reduced wN3% TST (< 15%, N = 17) and Patients With Normal N3% TST (15% < N3% TST < 20%, N = 19)

Among the 36 patients for whom we had the corresponding data, those with N3% TST < 15% (n = 17) did not differ from patients with N3% TST 15%–20% (n = 19) in clinical, sleep, or psychiatric characteristics. Both groups showed moderate anxiety and depressive symptoms, mild seasonality, moderate insomnia with poor sleep quality, mild daytime sleepiness, and a neutral chronotype (Tables 5 and 6).

3.6.3. Comparison of Clinical Variables (N = 69) Between Patients With Reduced N3% TST (N = 17) and Patients With Increased N3% TST (N = 53)

Only the number of hospitalisations differed significantly between groups, being higher in the reduced N3% TST group than in the increased N3% TST group (p = 0.005). All other variables showed no statistically significant differences (p > 0.05), (Tables 5 and 6 and Figure 1a).

3.7. Analysis of Predictive Factors for Patient Distribution Based on Increased and Normal N3% TST: Binomial Regression

We next used BSLR to define the best model to predict depression with increased N3% of TST.

Several subjective measures and clinical variables showed significant differences only between the increased N3% TST and the normal N3% TST groups in the univariate analysis and were entered into a BSLR along with age and sex. A combination of five variables predicted depression with N3% TST impairment well among 30 patients for whom we had all the data: the number of MDE, the GSS score, the MEQ score, age and sex (Table 7). The overall model showed a statistical trend (p = 0.074) with a McFadden's R ² of 0.308, indicating a moderate fit.

TABLE 7.

Analysis of predictive factors for patient distribution based on increased and normal N3% TST: Binomial regression.

Predictor	Estimation	Standard error	Z	p	Odds ratio	95% confidence interval
						Inferior	Superior
Intercept	−0.51501	2.7930	−0.1844	0.854	0.597	0.00251	142.469
MDE Number	−0.31423	0.1370	−2.2935	0.022	0.730	0.55835	0.955
SPAQ GSS	0.00732	0.0904	0.0809	0.936	1.007	0.84374	1.203
MEQ	0.10720	0.0643	1.6666	0.096	1.113	0.98130	1.263
AGE	−0.05937	0.0499	−1.1886	0.235	0.942	0.85448	1.039
Sex
Female–Male	1.02426	1.2481	0.8206	0.412	2.785	0.24122	32.155

Model	Deviance	AIC akaike information criterion	R 2 McF	General model test
				χ 2	ddl	p
1	22.6	34.6	0.308	10.0	5	0.074

Precision	Specificity	Sensibility	Area under the curve	Youden index
0.900	0.714	0.957	0.863	0.671

Note: Value in bold indicates a statistically significant difference with a p‐value of less than 0.05.

Abbreviations: MDE, major depressive episode; MEQ, morningness eveningness questionnaire; SPAQ GSS, seasonal pattern assessment questionnaire—global seasonality score; TST, total sleep time.

The model performed strongly with a sensitivity of 0.957, specificity of 0.714, precision of 0.900, and an AUC of 0.863, reflecting a good ability to distinguish between the groups. The Youden index was 0.671, showing a strong balance between sensitivity and specificity. Results are shown in Table 7. Additionally, the predictive values were calculated. The model yielded a Positive Predictive Value (PPV) of 91.67% and a Negative Predictive Value (NPV) of 83.33%, suggesting high reliability in identifying individuals with increased N3% TST, as well as reasonable accuracy in excluding those without it. Interestingly, the results showed that only the number of depressive episodes was significantly associated with normal N3% TST.

Importantly, the model was not adjusted for disorder polarity. Although patients with bipolar depression showed a significantly lower N3% TST compared to unipolar patients, the proportion of individuals with N3% TST > 20% did not differ between the two groups. Therefore, polarity was not retained as an adjustment variable in the regression model.

3.8. Sensitivity Analyses: Medication Effects

3.8.1. Comparison of N3 Stage Metrics in Patients Suffering From MDE (N = 64) Between Those Using Antidepressant (N = 35) and Those Who Didn't Use Antidepressant (N = 29)

Among the 64 patients for whom we had the data, 35 used antidepressants and 29 did not. No significant differences in N3 sleep parameters were observed (Table S5).

3.8.2. Comparison of N3 Stage Metrics in Patients Suffering From MDE (N = 63) Between Patients Using Benzodiazepine‐Class Anxiolytic (BZD, N = 27) and Patients Who Are Not Taking Benzodiazepine‐Class Anxiolytic (Non‐BZD, N = 36)

Among the 63 patients for whom we had the data, 27 used benzodiazepines and 36 did not. N3% TST was higher in the BZD group (26.0% ± 9.66%) than in the non‐BZD group (20.6% ± 11.1%; t = 2.06, p = 0.044) (Table S5). No other N3 parameters differed.

To explore the potential impact of anxiolytic use on our outcome of interest, a second BSLR model was tested including anxiolytic use as an additional predictor (Table S6). Twenty nine patients for whom we had all the data were included in the test. The overall model was not statistically significant (p = 0.092). Anxiolytic use was not a significant predictor of increased N3% TST (p = 0.293). Model performance was also lower compared to the initial model. The AUC was 0.864, with a sensitivity of 0.909 and a specificity of 0.429 at the default cutoff of 0.5. The Youden Index was 0.338. These results suggest that while anxiolytic use showed some association with the outcome in exploratory analysis, its inclusion did not meaningfully improve predictive accuracy or model fit in the regression model.

3.8.3. Comparison of N3 Stage Metrics in Patients Suffering From MDE (N = 65) Between Patients Using Mood Stabiliser (N = 21) and Patients Without Mood Stabiliser (N = 44)

Among the 61 patients for whom we had the data, 21 used mood stabilisers and 44 did not; no significant differences in N3 sleep parameters were observed (Table S5).

3.8.4. Comparison of N3 Stage Metrics in Patients Suffering From MDE (N = 63) Between Patients Using Hypnotic (N = 23) and Patients Without Hypnotic (N = 40)

Among the 63 patients for whom we had the data, 23 used hypnotics and 40 did not; no significant differences in N3 sleep parameters were observed (Table S5).

3.8.5. Comparison of N3 Stage Metrics in Patients Suffering From MDE (N = 64) Between Patients Using Antipsychotic (N = 8) and Patients Without Antipsychotic (N = 56)

Among the 64 patients for whom we had the data, 8 used hypnotics and 56 did not; no significant differences in N3 sleep parameters were observed (Table S5).

4. Discussion

This study is the first, to our knowledge, to show that SWS delineates clinically distinct depressive profiles based on its own phenotype—its proportion (TSP, TST), duration, and latency. Across dimensional analyses, shorter N3 latency was consistently associated with poorer subjective sleep quality and greater anxiety and depressive severity. Both increased and reduced N3 proportions were linked to markers of more severe depression. Increased %N3 (TSP) was associated with earlier hospitalisation, younger age at first suicide attempt, and greater daytime sleepiness, whereas reduced %N3 was linked to more hospitalisations and more recurrent depressive episodes. Categorical analyses further highlight that both increased and reduced either N3 duration in minutes or proportions of N3 sleep (%N3 TST) were linked to clinical indicators of severe depression: increased N3 duration in minutes compared to reduced one exhibited significantly higher anxiety and depressive symptom severity, indicating a greater depressive symptom burden and overall distress. In addition, reduced %N3 TST was linked to a higher number of psychiatric hospitalisations compared to increased %N3 TST, whereas increased %N3 was associated with greater seasonal mood variation, a stronger morning chronotype but lower lifetime number of depressive episodes and hospitalisations compared to normal %N3 TST. Interestingly, a robust five‐factor model (number of depressive episodes, chronotype, seasonality, age, sex) distinguished increased from normal %N3 TST (AUC = 0.863), with excellent sensitivity and high positive predictive value. This combination of clinical and chronobiological features including a higher sensibility to season, a tendency towards morningness, younger age and being a man, strongly predicted the subgroup with increased N3% TST and effectively ruled out its absence.

All these results challenge the long‐held assumption that SWS is uniformly diminished in depression and align with our hypothesis regarding the heterogeneity of mood disorders. Notably, the originality of the present findings lies in the observation that MDE with high N3 duration or N3% index is a marker of clinical severity across the life course. Specifically, from a dimensional perspective, higher N3% TSP subgroup is characterised by an earlier age at first hospitalisation, a younger age at first suicide attempt, and greater daytime sleepiness. From a categorical perspective, subjects with increased N3% TST also show greater sensitivity to seasonal variations, and a morning‐type chronotype. This is coherent with literature describing a hypersomnolent depressive phenotype that carries higher suicidal vulnerability, larger seasonal impacts on mood, sleep, and circadian dimensions throughout the year (Maruani et al. 2023). Our study reflected the broad heterogeneity previously reported in depression (Baglioni et al. 2016). Notably, we identified—for the first time—a reduction in N3 sleep in bipolar depression, both in duration and proportion of TST (Sarzetto et al. 2022). Men also showed higher N3% TST than women, in line with prior findings (Patel et al. 2004). No differences in N3 were found between patients with and without a seasonal MDE pattern; however, dimensional analysis revealed that individuals with increased N3% TST displayed greater sensitivity to seasonal variations. Likewise, N3 metrics did not differ according to treatment resistance or suicidal ideation. These observations highlight the need for broader phenotyping. A next step would involve cluster analyses integrating N3 measures (level, latency, first‐cycle metrics), global sleep architecture, and key clinical variables (episode count, hospitalisations, suicidality, chronotype, seasonality, daytime sleepiness). Based on our results, distinct clinical clusters are expected. One profile may present reduced N3%, particularly in bipolar depression and in women with recurrent hospitalisations. This relapsing trajectory may benefit from close monitoring, proactive relapse prevention, mood‐stabilising strategies, and N3‐enhancing interventions, including chronotherapeutic (Gottlieb et al. ^2019a) such as light therapy (Geoffroy et al. 2025), and cognitive behavioural therapy adapted for mood disorders (Riemann et al. 2023). Another profile, defined by increased N3%, morning chronotype, marked seasonality, daytime sleepiness, earlier onset suicidal behaviour and psychiatric hospitalisation, predominantly among individuals with unipolar depression, suggests a hypersomnolent seasonally sensitive phenotype with a morning chronotype. This group may benefit from circadian‐modulating treatments, including morning bright‐light therapy, phase‐shifting interventions, and wake‐promoting strategies aimed at reducing N3 proportion (Gottlieb et al. ^2019b). This combined approach may also help address the increased vulnerability to suicidal ideation often associated with hypersomnolence. Very interestingly, our study revealed that the duration of N3 sleep and its proportion relative to total sleep time (TST) are highly distinct variables and are not associated with the same characteristics of depression. Patients with increased N3 duration showed greater severity of anxiety and depressive symptoms, an association not observed in patients with an increased N3% of TST. This suggests that N3 sleep duration, and its proportion may reflect different underlying pathophysiological mechanisms of depression. These patterns suggest that N3 sleep should not be considered as a single, universal “sleep biomarker”. Rather, its full‐night distribution—encompassing latency and duration, and cycle—represents a dynamic and informative dimension for identifying biologically and clinically distinct subtypes of depression. The physiological heterogeneity captured by N3 architecture appears to mirror key differences in illness trajectory, relapse vulnerability, and care needs, offering a foundation for precision psychiatry anchored in objective sleep parameters.

Several limitations should be noted. Our findings suggest that N3 proportion and duration may contribute to clinically relevant stratification within depressive episodes. However, this study was exploratory and had limited statistical power. Accordingly, it should be considered a first step in the development of a predictive model. A second phase, involving a larger and independent sample with alpha risk correction for multiple tests, is planned to formally evaluate model performance, generalisability, and robustness using appropriate validation procedures. More broadly, evaluation of N3 measures as potential biomarkers will require controlled, longitudinal studies with independent samples. In addition, our data were based on single‐night PSG recordings, which may be affected by first‐night effects. The absence of randomised psychotropic exposure introduces possible confounding. Nevertheless, sensitivity analyses showed that only anxiolytics increased N3 proportion, and anxiolytic use did not remain in the final regression model predicting elevated N3% TST, suggesting that this phenotype is not medication‐driven. Finally, variability in PSG systems and clinical settings may have introduced measurement variance.

5. Conclusion

Our results identify distinct N3 phenotypes linked to specific clinical profiles and illness severity, supporting individualised depression care guided by sleep architecture. N3 metrics may stratify relapse risk, chronicity, and healthcare use beyond symptom scales. Future studies should use multi‐night PSG, clustering, and longitudinal follow‐up to validate these phenotypes and guide targeted interventions.

Author Contributions

Julia Maruani, Antoine Salmeron, Pierre A. Geoffroy: conceptualization, investigation, methodology, validation, visualization, writing – review and editing, supervision, writing – original draft, data curation. All authors contributed to the article and approved the submitted version.

Funding

The authors have nothing to report.

Conflicts of Interest

The authors declare no conflicts of interest.

Supporting information

Table S1: Polysomnographic results (N = 97) of patients with MDE.

Table S2: Correlations between N3 latency and clinical/sleep variables.

Table S3: Comparison of overall PSG metrics across clinical subgroups: bipolar and unipolar depression (N = 66); MDE with and without seasonal pattern (N = 83); men and women with MDE (N = 94); treatment‐resistant and non–treatment‐resistant MDE (N = 66); suicidal ideation and none (N = 55).

Table S4: Comparison of N3 latency between clinical subgroups without outlier.

Table S5: Comparison of N3 sleep metrics in user's vs. non‐users of antidepressants (N = 64), anxiolytics (N = 63), mood stabilisers (N = 65), hypnotics (N = 63), and antipsychotics (N = 64).

Figure S1: Clinical variables differing between patients with reduced and increased N3 sleep duration. HADS, Hospital Anxiety and Depression scale; QIDS, Quick Inventory Depressive Symptoms 16‐item scale. 1. N3 Increased Duration; 2. N3 Reduced Duration.

Figure S2: N3 sleep differences between unipolar and bipolar depression. TSP, total sleep period; TST, total sleep time; 1. Unipolar; 2. Bipolar.

Data Availability Statement

The data that support the findings of this study are available on request from the corresponding author. The data are not publicly available due to privacy or ethical restrictions.