Clinical correlates of sleep profiles in dementia with Lewy bodies: a latent class approach

Abstract

Background

Rapid eye movement behavioral disorder (RBD) is a well-studied feature of dementia with Lewy bodies (DLB). However, patients with DLB report a significantly wider range of sleep symptoms than those of RBD. Our objective was to characterize profiles of sleep disturbances in DLB.

Methods

Retrospective cross-sectional study including patients with DLB fulfilling McKeith et al. diagnostic criteria in 3 centers at AP-HP GHU Nord, Paris, France. Nine sleep symptoms were evaluated. Latent class analysis was performed on sleep symptoms to identify underlying patterns. Classes were compared on demographic and clinical characteristics.

Results

Among 359 patients (age 74.6 ± 7.5 years; 49% female, disease evolution: 4.6 ± 4.1 years), 95% had sleep disturbances. The most frequent symptoms included excessive daytime sleepiness (70.8%), probable RBD (67.7%), and insomnia/sleep fragmentation (60.9%). A motor disease onset was associated with RBD and nightmares, and a psychiatric onset with insomnia/sleep fragmentation. Latent class analysis identified three sleep profiles: a “Global sleep impairment” profile (n = 106) with the highest sleep-symptom burden, advanced cognitive impairment, and frequent hallucinations and fluctuations; a “RBD – Nightmares” profile (n = 149) associated with dysautonomia, clinical breathing disturbances and better-preserved cognition; and a “Paucisymptomatic” profile (n = 104) with the lowest sleep-symptom burden but advanced cognitive impairment.

Conclusion

We identified three distinct sleep profiles, which were associated with distinct patterns of core clinical features. These profiles may reflect specific underlying pathophysiological mechanisms. Better characterization of sleep symptoms may help guide therapeutic strategies and improve quality of life.

Supplementary Information

The online version contains supplementary material available at 10.1186/s13195-026-02059-8.

Background

Dementia with Lewy bodies (DLB) is characterized by substantial heterogeneity across clinical presentation, disease trajectories, and neuropathological substrates. Clinically, DLB presents with cognitive decline, associated with a variable combination of parkinsonism, attentional and cognitive fluctuations, visual hallucinations, and rapid eye movement sleep behavior disorder (RBD) [1]. This clinical heterogeneity is reflected in the variability in predominant symptoms at onset, which may include cognitive, psychiatric, or motor symptoms [2, 3]. These distinct patterns of onset may influence the time to diagnosis and the subsequent disease trajectory [4]. Different patterns of Lewy body pathology have been shown to influence the clinical phenotype [5]. In addition, the presence of copathologies may further contribute to clinical heterogeneity. Among these, Alzheimer’s disease–related neuropathological changes are frequently observed in DLB and represent an important source of pathological and clinical variability [6].

Among the clinical manifestations contributing to this heterogeneity, sleep disturbances represent a particularly prominent feature of DLB [7]. Indeed, sleep disturbances are more frequent and diverse in DLB than in other neurodegenerative diseases [8, 9]. While RBD, a parasomnia characterized by dream enactment behavior during REM sleep, is the most well-characterized sleep disturbance in DLB, reported in as many as 60% of cases, other sleep-related symptoms are also frequently observed [10, 11]. Excessive daytime sleepiness (EDS) has been reported in 11% to 100% of patients, and sleep apnea syndrome in 18% to 35% [10, 11]. Other frequent but less well-documented symptoms include nightmares, nocturnal visual hallucinations, nocturnal confusion, insomnia, restless legs syndrome (RLS), and circadian rhythm disturbances, including reversal of the sleep–wake cycle. Overall, an estimated 90% of patients with DLB report at least one sleep-related complaint during the course of the disease [10]. These symptoms are associated with poorer quality of life, increased caregiver burden, and may accelerate cognitive and physical decline [12, 13].

Variation in the sleep profiles in DLB and their contribution to disease heterogeneity has been rarely investigated [4, 14]. Sleep symptoms have frequently been examined in isolation; however, to disentangle their contribution to the clinical phenotype, it is essential to investigate the associations among co-occurrence patterns. Additionally, disease variability is more pronounced in the early stages, while clinical presentation becomes more homogeneous over time. In this regard, the relationship between sleep disturbances and the different onset phenotypes remains to be characterized.

This study aimed to (i) analyze and describe the frequency and characteristics of sleep disturbances in a cohort of patients diagnosed with probable DLB; (ii) examine associations between core clinical features of DLB and sleep disorders; and (iii) identify potential distinct patterns of sleep disturbance that may inform underlying pathophysiological mechanisms and therapeutic approaches.

Methods

Cohort

We conducted a retrospective multicenter cross-sectional observational study including patients with DLB evaluated at three centers: GHU AP-HP Nord (Cognitive Neurology Center, Lariboisière Fernand-Widal Hospital; Geriatric Day Hospital, Bretonneau Hospital); and Sainte-Périne Hospital Memory Clinic AP-HP, in the context of the BioCogBank protocol [15, 16]. All patients evaluated for a possible DLB between January 2009 and April 2025 were systematically screened. All patients underwent a detailed history and clinical assessment, including physical and cognitive examinations by behavioral neurologists or geriatricians trained in dementia and movement disorders, structural brain imaging, and any additional biomarkers deemed necessary by the treating physician to document the diagnosis. Motor symptoms were assessed using a standardized neurological examination based on the MDS-UPDRS framework, although the scale was not systematically administered. Orthostatic hypotension (OH) was defined as a decrease in systolic blood pressure ≥ 20 mmHg or diastolic blood pressure ≥ 10 mmHg within 3 min of standing; an attenuated heart rate response was considered suggestive of neurogenic etiology. Diagnoses were reviewed by a multidisciplinary team of neurologists, geriatricians, neuroradiologists, and biologists.

Patients were included if they met the criteria for “probable DLB” as defined by McKeith et al. (2017) and if they had available sleep data [1].

Core clinical features

Core clinical features, including cognitive impairment, cognitive fluctuations, motor symptoms, hallucinations, and OH, were coded as absent or present at diagnosis. Cognitive status (mild cognitive impairment vs. dementia) was determined by the treating clinician based on clinical evaluation, functional status, and cognitive testing [17, 18]. Mode of disease onset was classified as cognitive, motor, or psychiatric based on the predominant initial symptoms reported by the patient or caregivers and documented in clinical interviews and medical records [3, 4]. In patients with a predominance of motor symptoms at onset, differentiation from Parkinson's disease was made according to the "1-year rule", which supports a diagnosis of DLB when cognitive dysfunction precedes or occurs within 1 year of motor-symptoms onset [4]. The Mini-Mental State Examination (MMSE) score was used as a measure of global cognition. Disease- and sleep-related treatments were collected at the time of diagnosis (antidepressants, benzodiazepines, neuroleptics, antiepileptics, dopaminergic drugs, cholinesterase inhibitors, melatonin, stimulants, nocturnal ventilation, or light therapy).

Sleep variables

Sleep-related clinical data were collected at diagnosis by the clinician in charge, based on patient and/or caregiver reports and validated instruments when available. Probable RBD was defined by the presence of reported dream enactment behaviors, as assessed by the validated RBD Single-Question Screen [19]. Nightmares were defined as recurrent nightmares according to ICSD-3 [20]. Night-time hallucinations (of any sensory modality) and night-time confusion (including confusional arousals and nocturnal wandering) were recorded. EDS was evaluated through clinical interview alone or with the Epworth Sleepiness Scale (range: 0–24; higher scores indicate greater sleepiness; normal ≤ 10). Insomnia and/or reported sleep fragmentation were characterized by patient- or caregiver-reported difficulties initiating or maintaining sleep, frequent nocturnal awakenings, or early morning awakenings. Clinical breathing disturbances were assessed clinically, searching for snoring, witnessed apnea, gasping/choking, nocturia, and morning headaches or dry mouth upon awakening. Circadian rhythm sleep–wake disorders, including advanced or delayed sleep–wake phase types, were determined by the clinician based on reported bedtime and wake time patterns. Restless legs syndrome (RLS) was diagnosed according to the presence of an urge to move the legs, typically accompanied by unpleasant sensations in the legs, with partial or complete relief upon movement [21]. The total number of sleep-related symptoms was calculated (0–9). For less prevalent sleep symptoms (severe nocturnal hallucinations, confusion, circadian rhythm disorders, RLS), lack of documentation was interpreted as absence of the symptom, unless explicitly noted by the clinician. Conversely, for common symptoms (probable RBD, nightmares, EDS, and clinical breathing disturbances), their absence had to be explicitly stated; otherwise, data were considered missing. Patients with ≥ 5 sleep variables missing were excluded to limit bias from incomplete characterization.

Polysomnography

A subgroup of patients underwent video-polysomnography (PSG) ordered by their treating physician for clinical indications. All included sleep studies were performed in a sleep laboratory for one night, under the supervision of a sleep technician, acquired and scored according to established guidelines by a sleep medicine clinician [22]. The recorded parameters included an electroencephalogram with three to eight channels, a bilateral electrooculogram, chin and at least two limb surface electromyograms, and an electrocardiogram. Respiratory recording included oronasal airflow recording by nasal cannula and oral thermistor, thoracic and abdominal movements by inductance plethysmography, and oxygen saturation by finger pulse oximetry.

The following parameters were collected: standard sleep parameters (sleep latency, total sleep time, sleep efficiency, and sleep stage distribution); loss of REM sleep atonia and the presence of complex motor behaviors; respiratory events (obstructive and/or central sleep apnea, Cheyne–Stokes respiration, and respiratory effort-related arousals); and motor disturbances (periodic limb movement index and related arousals). The diagnosis of RBD was established according to ICSD-3 criteria based on the presence of REM sleep without atonia and dream-enactment behavior [20]. Disrupted sleep architecture, confusional arousals, and EEG abnormalities were noted.

Statistical analysis

Categorical variables were compared between groups using χ² or Fisher’s exact tests depending on the expected cell count. Continuous variables were compared using Student’s t-test for normally distributed data or the Mann–Whitney U test for non-normally distributed data. For categorical variables with more than two levels, post hoc pairwise comparisons with Bonferroni correction were computed.

To identify latent profiles of sleep disturbances within the cohort, a latent class analysis (LCA) was performed using the poLCA package in R [23]. All sleep-related symptom variables were included, and models with 2 to 5 classes were tested to determine the optimal number of classes. Each model was estimated with 50 random starts to reduce the risk of local maxima, with convergence set at 5,000 iterations and a tolerance of 1 × 10⁻⁶. Missing data were handled using the Full Information Maximum Likelihood (FIML) approach, allowing inclusion of incomplete cases without data imputation. Model selection was based on the Bayesian Information Criterion (BIC), balancing model fit and model complexity. Probabilities of each symptom were compared pairwise across classes with Wald tests, adjusting for multiple comparisons with Holm correction. Posterior class membership probabilities were calculated for all participants, who were then assigned to the latent class with the highest posterior probability. Between-class comparisons were performed for demographic and clinical variables, using χ² tests for categorical variables and analysis of variance (ANOVA) for continuous variables. To explore the effect of medication on sleep classes, treatments associated with the presence of sleep symptoms were included as covariates in the LCA model, and odds ratios (ORs) were calculated to quantify their association with class membership. As sensitivity analyses, the LCA model was re-run in separate analyses excluding patients receiving treatments associated with sleep symptoms.

To assess whether the PSG subgroup differed from the overall population, we performed a logistic regression with PSG availability as the dependent variable and all variables of interest as predictors. Concordance between clinical and PSG-based diagnoses was expressed as the percentage of overall agreement.

A two-sided significance level of p < 0.05 was used for all analyses. Analyses were performed using R software (version 4.4.2) and RStudio (version 2024.12.1 + 563, Posit, PBC).

Results

Cohort description

Of 416 evaluated patients, 359 who met criteria for probable DLB according to McKeith et al. criteria (2017) and with complete sleep data were included (Flow chart, Supp. Figure 1). Patients’ characteristics are shown in Table 1. The cohort included 49% female patients, with a mean age at diagnosis of 74.6 ± 7.5 years and disease duration of 4.6 ± 4.1 years. Most patients had a cognitive-predominant onset (59.3%), followed by a motor-predominant onset (24.8%) and a psychiatric-predominant onset (15.9%). The mean MMSE score was 20.6 ± 5.8. Antidepressants (40.6%) and dopaminergic drugs (23.8%) were the most frequently used treatments.

Table 1.

Demographics and clinical characteristics of the study population

	Overall	Missing data
	N = 359
Age at diagnosis, years, mean ± SD	74.6 ± 7.5	0 (0%)
Sex, female, n (%)	176 (49.0%)	0 (0%)
Disease duration, years, mean ± SD	4.6 ± 4.1	4 (1.1%)
Initial clinical phenotype, n (%)		0 (0%)
Cognitive	213 (59.3%)
Motor	89 (24.8%)
Psychiatric	57 (15.9%)
Symptoms at diagnosis, n (%)
Dementia	312 (89.7%)	11 (3.1%)
Motor symptoms	259 (74.9%)	13 (3.6%)
Visual hallucinations	205 (61.9%)	28 (7.8%)
Cognitive fluctuations	190 (68.1%)	80 (22.3%)
Orthostatic hypotension	130 (41.1%)	43 (12.0%)
MMSE,/30, mean ± SD	20.6 ± 5.8	119 (33.1%)
Treatments at diagnosis, n (%)
Antidepressant	132 (40.6%)	34 (9.5%)
Benzodiazepine	30 (9.3%)	35 (9.8%)
Neuroleptic	46 (14.2%)	35 (9.8%)
Dopaminergic therapy	78 (23.8%)	31 (8.6%)
Antiepileptic	15 (4.6%)	36 (10.0%)
Cholinesterase inhibitor	47 (14.5%)	34 (9.5%)

Quantitative variables are presented as mean ± standard deviation (SD). Categorical variables are expressed as the number of positive cases (percentage, %)

Sleep symptoms

Among the 359 patients in our cohort, the mean number of sleep-related symptoms was 3.73 ± 1.91. The most frequent symptoms were EDS (70.8%), probable RBD (67.7%), insomnia/clinical sleep fragmentation (60.9%), and nightmares (58.7%), followed by clinical breathing disturbances (37.9%), night-time confusion (34.2%), night-time hallucinations (33.8%), and more rarely, restless legs syndrome (9.2%) and circadian rhythm disorders (8.9%) (Fig. 1A). Pairwise associations among sleep symptoms (Fig. 1B, Supp. Figure 2) revealed strong interrelations between probable RBD and nightmares, night-time hallucinations, and clinical breathing disturbances (p < 0.0001), and between night-time confusion and hallucinations, insomnia/clinical sleep fragmentation, and EDS (p < 0.0001). No differences in sleep symptom frequency were found between men and women (all p > 0.05, Supp. Table 1, Supp. Figure 3).

Fig. 1.

Sleep-related symptoms in the cohort. A Frequency of each sleep-related symptom in the cohort. Missing data: excessive daytime sleepiness, n = 14 (3.9%); probable RBD, n = 1 (0.3%); insomnia/clinical sleep fragmentation: n = 2 (0.6%); nightmares, n = 4 (1.1%); clinical breathing disturbances, n = 42 (11.7%); night-time confusion, n = 2 (0.6%); night-time hallucinations, n = 1 (0.3%); restless legs syndrome, n = 2 (0.6%); circadian rhythm disorders, n = 1 (0.3%). B Chord diagram of sleep-related symptoms and their co-occurrence. Each color-coded segment represents one sleep-related symptom. The width of the connecting lines indicates the relative frequency of co-occurrence between symptom pairs

Regarding the association with the initial clinical presentation, patients with a psychiatric-predominant onset reported a higher number of sleep-related symptoms than those with a cognitive onset (4.2 ± 1.8 vs. 3.5 ± 1.8, p = 0.015). Motor onset was associated with a higher rate of probable RBD than cognitive onset (80.9% vs. 61.0%, OR = 2.70 [95% CI 1.45 to 5.23], p = 0.0023) and a higher rate of nightmares (71.6% vs. 50.7%, OR = 2.44 [95% CI 1.49 to 4.38], p = 0.0026; Fig. 2). Insomnia/clinical sleep fragmentation was more frequent in psychiatric onset than in cognitive onset (73.7% vs. 55.4%, OR = 2.25 [95% CI 1.14 to 4.64], p = 0.045), with a trend toward more frequent night-time confusion and hallucinations (p = 0.055 and p = 0.059, respectively).

Fig. 2.

Frequency of sleep-related symptoms by initial clinical phenotype. Comparisons across the three onset phenotypes (cognitive, motor, and psychiatric) were performed using χ² or Fisher’s exact tests, as appropriate. Post hoc pairwise comparisons were performed with adjustment for multiple comparisons using Bonferroni correction. ^*Probable RBD: initial cognitive phenotype (61.0%) vs. initial motor phenotype (80.9%), p = 0.0023; ^†Nightmares: initial cognitive phenotype (50.7%) vs. initial motor phenotype (71.6%), p = 0.0026; ^‡Insomnia/sleep fragmentation: initial cognitive phenotype (55.4%) vs. initial psychiatric phenotype (73.7%), p = 0.045

At the evaluation time point (Supp. Table 2), EDS was associated with advanced cognitive decline (p < 0.001) and fluctuations (p = 0.0034). Night-time hallucinations, confusion, and insomnia were associated with daytime hallucinations (p ≤ 0.03). Probable RBD was associated with a higher frequency of OH (p = 0.0090).

After adjustment for age and sex, the sleep symptom burden was not associated with MMSE score (β = −0.014, p = 0.67). The use of benzodiazepines was associated with a higher frequency of night-time confusion (56.7% vs. 32.4%, p = 0.0082) (Supp. Table 3). Dopaminergic therapy was associated with a higher frequency of restless legs syndrome (16.7% vs. 7.6%, p = 0.020).

Sleep profiles

To identify potential distinct patterns of sleep disturbances, we applied latent class analysis (LCA) using the nine sleep-related symptoms as indicator variables (Supp. Figure 4). The best-fitting model based on BIC supported a three-class solution (Fig. 3). Each class exhibited a distinct profile of sleep symptoms and was labeled according to its most salient features: “Global sleep impairment,” “RBD – Nightmares,” and “Paucisymptomatic” (Supp. Table 4). Proportion of missing data did not differ between classes (Supp. Table 5). The average posterior probabilities for the most likely class membership were high (“Global sleep impairment,” 0.91; “RBD – Nightmares,” 0.92; “Paucisymptomatic,” 0.96), indicating clear class separation. The “Global sleep impairment” class had the highest number of sleep-related symptoms (5.6 ± 1.1), followed by the “RBD – Nightmares” class (3.8 ± 1.3), while the “Paucisymptomatic” class showed the fewest (1.8 ± 1.4; p < 0.0001 for all comparisons). The “Global sleep impairment” class was characterized by a high burden of all sleep-related symptoms, with elevated rates of night-time hallucinations and night-time confusion (p < 0.001 compared with other classes). In contrast, the “RBD – Nightmares” class was systematically associated with the presence of probable RBD (100%), frequent nightmares and clinical breathing disturbances (p < 0.001 compared with other classes) and rare night-time hallucinations and confusion. EDS and insomnia/clinical sleep fragmentation were highly prevalent across all three classes, consistent with their overall frequency in the cohort, and were therefore less discriminative. In contrast, circadian rhythm disorders and RLS remained rare in all three classes.

Fig. 3.

Latent class analysis of sleep-related symptoms. A Distribution of sleep symptoms across identified classes using latent class analysis. B Estimated probability of each sleep symptom within each class. Comparisons of probabilities were performed by comparing median differences using Wald tests, adjusting for multiple comparisons using Holm correction. Significance is indicated as: * p < 0.05; ** p < 0.01; *** p < 0.0001

Based on their individual symptom profiles, 106 patients were classified in the “Global sleep impairment” class, 104 in the “Paucisymptomatic” class, and 149 in the “RBD – Nightmares” class. No significant differences were observed across classes regarding age, sex, disease duration, or initial clinical presentation (Table 2; Supp. Table 6 providing detailed pairwise comparisons and effect sizes). Regarding clinical features, visual hallucinations were more frequent in the “Global sleep impairment” class than in both the “Paucisymptomatic” (84.2% vs 54.3%, OR = 4.35 [95% CI 2.17 to 9.09], p < 0.0001) and “RBD – Nightmares” (84.2% vs 50.7%, OR = 5.00 [95% CI 2.63 to 10.31], p < 0.0001) classes. Cognitive fluctuations were more frequent in the “Global sleep impairment” class than in the “Paucisymptomatic” class (80.0% vs. 61.4%, OR = 2.49 [95% CI 1.18 to 5.43], p = 0.031). Orthostatic hypotension was more common in the “RBD – Nightmares” class than in the “Paucisymptomatic” class (51.1% vs. 32.2%, OR = 2.20 [95% CI 1.21 to 4.04], p = 0.024). To address potential overlap with core DLB features, analyses adjusted for night-time hallucinations, confusion, or EDS showed that differences between classes in daytime hallucinations remained significant, whereas those in cognitive fluctuations were no longer significant (Supp. Table 7).

Table 2.

Sleep classes comparison

	Global sleep impairment	RBD – Nightmares	Paucisymptomatic	p-value
	n = 106	n = 149	n = 104
Age at diagnosis, years, mean ± SD	74.5 ± 7.8	73.9 ± 7.3	75.8 ± 7.5	0.071
Sex, female, n (%)	60 (56.6%)	70 (47.0%)	46 (44.2%)	0.16
Disease duration, years, mean ± SD	4.7 ± 4.6	4.5 ± 3.8	4.5 ± 4.0	0.83
Initial clinical phenotype, n (%)				0.035a
Cognitive	53 (50.0%)	86 (57.7%)	74 (71.2%)
Motor	31 (29.2%)	40 (26.8%)	18 (17.3%)
Psychiatric	22 (20.8%)	23 (15.4%)	12 (11.5%)
Symptoms at diagnosis, n (%)
Dementia	90 (88.2%)	130 (90.3%)	92 (90.2%)	0.85
Motor symptoms	73 (72.3%)	116 (79.5%)	70 (70.7%)	0.23
Visual hallucinations	85 (84.2%)	70 (50.7%)	50 (54.3%)	< 0.0001b
Cognitive fluctuations	64 (80.0%)	75 (64.7%)	51 (61.4%)	0.023c
Orthostatic hypotension	34 (35.4%)	68 (51.1%)	28 (32.2%)	0.0080d
MMSE,/30, mean ± SD	18.9 ± 6.0	22.1 ± 5.2	20.2 ± 5.9	0.0025e
Sleep-related symptoms at diagnosis,/9, mean ± SD	5.6 ± 1.1	3.8 ± 1.3	1.8 ± 1.4	< 0.0001f

Quantitative variables are presented as mean (standard deviation, SD). Categorical variables are presented as the number of positive cases (percentage, %). Comparisons were performed using χ² or Fisher’s exact tests, depending on expected cell counts. Post hoc pairwise comparisons were performed with Bonferroni correction

^aNo significant posthoc differences

^bGlobal sleep impairment vs. others, p < 0.0001

^cGlobal sleep impairment vs. Paucisymptomatic, p = 0.031

^dPaucisymptomatic vs. RBD – Nightmares, p = 0.024

^eRBD – Nightmares vs. Global sleep impairment: p = 0.0018

^fp < 0.0001 in-between all classes

Bold font indicates statistically significant omnibus test results

The mean MMSE score was lower in the “Global sleep impairment” (18.9 ± 6.0) than in the “RBD – Nightmares” class (22.1 ± 5.2; mean difference, −3.20 [95% CI −5.01 to −1.39], p = 0.0018). In multivariable regression analyses adjusting for age, sex, disease duration, and treatments, the associations between sleep classes and key outcomes (MMSE, hallucinations, fluctuations, and orthostatic hypotension) remained significant (Supp. Table 8).

Compared with the “RBD – Nightmares” class, benzodiazepine use was associated with higher odds of belonging to the “Paucisymptomatic” (OR = 7.64 [95% CI 1.37 to 42.4]) and “Global sleep impairment” classes (OR = 10.29 [95% CI 1.77 to 59.8]). In contrast, dopaminergic therapy was not significantly associated with class membership. In two separate sensitivity analyses excluding patients receiving benzodiazepines and dopaminergic therapy, respectively, the same three latent classes were identified (Supp. Figure 5).

Polysomnography findings

PSG data were available for 69 patients in our cohort (Supp. Table 9). Most PSG recordings (n = 52) were obtained at the Bichat-Claude Bernard Sleep Disorders Center, with the remainder performed in other accredited centers. The number of sleep-related symptoms was significantly higher in the PSG subgroup compared with the remainder of the cohort (4.4 ± 1.6 vs. 3.6 ± 2.0, p < 0.05, Supp. Table 9). Analysis of the PSG data revealed impaired sleep efficiency (total sleep time/time in bed) in 42.6% (29/68) of patients, and impaired sleep quality ((slow-wave sleep + REM sleep)/total sleep time) in 46.9% (30/64) (Table 3). Sleep was frequently fragmented, with increased wake after sleep onset observed in 73.4% (47/64) of cases. Sleep architecture was overall altered, with over half of the patients showing abnormal sleep–wake patterns, characterized by changes in macroarchitecture (predominance of light non-REM with frequent arousals/stage transitions, profound reduction of N3 and REM) and microarchitecture (absence of sleep spindles, diffuse slow wave activity). SAS was confirmed by PSG in 79.7% of patients. Among those, 26.8% had mild SAS (5 to < 15 apneas/hour), 39.3% had moderate SAS (15 to < 30 apneas/hour), and 33.9% had severe SAS (≥ 30 apneas/hour). Most cases were obstructive sleep apnea (72.5%); central or mixed apnea was present in 27.5% of cases. There was a moderate overall agreement between clinical diagnosis of RBD and PSG-documented REM loss of atonia (72.1%) and between clinical and PSG-confirmed SAS (72.1%). Among the 10 patients with a clinical diagnosis of probable RBD not confirmed by PSG, 9 had SAS. The last patient had moderate periodic limb movements (27 per hour).

Table 3.

Polysomnography findings

	PSG data (n = 69)	Normal ranges (23)
Sleep architecture
Sleep efficiency, %, mean ± SD	74.8% ± 19.6%	≥ 80%
Wake after sleep onset, min., mean ± SD	84.6 ± 63.9	< 40 min
Microarousal index, n/hours, mean ± SD	22.3 ± 16.7	< 35/h
Sleep quality, %, mean ± SD	39.4% ± 18.0%	≥ 35%
Sleep stages, n (%)
N1 – Increased	35/64 (54.7%)	5–10% of TST
N1 – Decreased	15/64 (23.4%)
N2 – Increased	19/64 (29.7%)	45–55% of TST
N2 – Decreased	34/64 (53.1%)
N3 – Increased	24/67 (35.8%)	15–25% of TST
N3 – Decreased	20/67 (29.9%)
REM sleep – Increased	9/67 (13.4%)	20–25% of TST
REM sleep – Decreased	50/67 (74.6%)
Sleep-disordered breathing
Sleep apnea syndrome, n (%)	55/69 (79.7%)
Obstructive sleep apnea, n (%)	37/51 (72.5%)
Mixed sleep apnea, n (%)	14/51 (27.5%)
AHI, event/hour, mean ± SD	22.0 ± 18.0	< 5/h
REM sleep abnormalities
REM sleep without atonia, n (%)	46/60 (76.7%)
Complex motor behaviors, n (%)	18/60 (30.0%)
Periodic limb movements, n (%)	18/67 (26.9%)	< 15/h

Sleep stages data are presented as the number and ratio of individuals who presented with sleep stage duration outside the indicated normal range. Categorical variables are presented as the number of positive cases / number of available cases (percentage, %). All studies were acquired and scored according to the AASM Manual for the Scoring of Sleep and Associated Events (3rd edition, 2023)

TST Total sleep time, AHI Apnea–hypopnea index

Among the PSG subgroup, 46 patients (66.7%) belonged to the “RBD – Nightmares” sleep class, 13 (18.8%) to the “Global sleep impairment” class, and 10 (14.5%) to the “Paucisymptomatic” class (Table 4). No significant differences were found across the three classes in terms of sleep efficiency, microarousal index, loss of REM sleep atonia, night-time confusion, or the presence of SAS (regardless of severity).

Table 4.

Polysomnography findings across sleep classes

	Global sleep impairment n = 13	RBD – Nightmares n = 46	Paucisymptomatic n = 10	p-value
Sleep efficiency, %, mean ± SD	68.9 ± 25.1	76.2 ± 18.5	76.0 ± 16.4	0.59
Microarousal index, /hour, mean ± SD	21.0 ± 13.2	22.1 ± 14.6	24.8 ± 27.9	0.94
REM sleep without atonia, n (%)	8/11 (72.7%)	34/41 (82.9%)	4/8 (50.0%)	0.11
Night-time confusion, n (%)	3/10 (30.0%)	1/42 (2.4%)	1/9 (11.1%)	0.018a
Sleep-disordered breathing, n (%)	10/13 (76.9%)	38/46 (82.6%)	8/10 (80.0%)	0.90
Sleep apnea profile
Obstructive sleep apnea, n (%)	5/9 (55.6%)	27/35 (77.1%)	6/8 (75.0%)	0.41
Mixed sleep apnea, n (%)	4/9 (44.4%)	8/35 (22.9%)	2/8 (25.0%)
AHI, mean ± SD	21.6 ± 19.2	20.8 ± 15.6	28.3 ± 26.4	0.50

Quantitative variables are presented as mean (standard deviation, SD). Categorical variables are presented as the number of positive cases / number of available cases (percentage, %). Comparisons were performed using Fisher’s exact tests. Post hoc pairwise comparisons were performed with Bonferroni correction

AHI Apnea–hypopnea index

^aSignificance was not retained after Bonferroni correction in post hoc analysis

Discussion

Our results confirm that sleep symptoms in DLB are highly prevalent and often multiple, and can be identified clinically and confirmed by PSG. Using a data-driven approach, we identified three DLB sleep profiles, independent of disease duration, ranging from a low to a high burden of sleep symptoms, characterized by different clinical features. The “Global sleep impairment” profile had a high sleep symptom burden, more frequent daytime hallucinations and cognitive fluctuations, and poorer cognition. In contrast, the “Paucisymptomatic” class was characterized by a low burden of sleep symptoms but similar levels of cognitive decline. The “RBD – Nightmares” class, with a 100% prevalence of RBD, was characterized by frequent clinical breathing disturbances, OH, and more preserved cognition compared with the other two classes. These findings provide insight into the clinical phenotypes of DLB, which may have utility for patient stratification and prognostication but also for further mechanistic research.

The most frequent sleep-related symptoms were EDS, probable RBD, insomnia/sleep fragmentation, and nightmares, in over 50% of patients, in line with previous studies [24, 25].

EDS has been well documented in the literature, both in clinical cohorts and through objective measures [25, 26]. Notably, EDS was not associated with sleep symptom burden, including clinically reported sleep fragmentation, or with treatment use, suggesting a specific association with the underlying neurodegeneration process. Indeed, there is evidence for alterations of brainstem and hypothalamic wake-sleep systems, including loss of dopamine, noradrenaline, and hypocretin neurons. EDS has also been associated with cholinergic neuronal loss in the nucleus basalis of Meynert [27]. EDS frequently overlaps with cognitive fluctuations and is often regarded as one of their characteristic features, which may contribute to its higher rate of reporting. Those two phenomena are difficult to disentangle clinically, and objective measures are most likely required to distinguish between daytime sleep intrusions and attentional deficits.

It is established that RBD is directly linked to pathological lesions of alpha-synucleinopathies, and it is a cardinal criterion in the current diagnostic criteria for DLB. In autopsy-confirmed DLB cases, RBD has been observed in up to 76% of patients [28]. In a recent subtyping study, the “DLB with RBD” phenotype represented one of the core phenotypes in the alpha-synucleinopathy continuum, along with PD and PD with RBD [29].

PSG findings confirmed altered sleep architecture and a significant occurrence of sleep-disordered breathing. However, while studies focusing on the respiratory aspects of sleep disturbance in DLB are still scarce, there is no strong evidence that sleep apnea would be more prevalent in DLB than in the general older adult population [30]. Multiple factors may elevate the risk of SAS, including diminished upper airway muscle tone, instability in respiratory flow, sedative medications, and autonomic dysregulation of respiratory control [31, 32].

Our data also confirmed the importance of PSG to exclude RBD mimics, as all non-confirmed RBD cases exhibited sleep apnea or periodic limb movements. We observed no significant sex differences in the prevalence of the different symptoms, in contrast to previous studies that report more frequent RBD in males [33]. Sleep-clinic or PSG-based studies may be enriched in male patients, partly because men are more likely to be referred for vigorous or overt dream enactment behaviors [34].

RLS prevalence (9%) was relatively low. A higher prevalence might have been expected based on data from Parkinson’s disease, another synucleinopathy in which RLS is more frequently reported [35]. However, RLS may be underrecognized in this population, as diagnosis relies largely on subjective symptom reporting and may require adapted criteria in cognitively impaired patients [36].

We did not observe an independent association between overall sleep burden and MMSE after adjustment. This may reflect the multifactorial nature of cognitive impairment in this disease, which is related to the underlying disease process, but also to potential copathologies and other symptoms: dysautonomia, psychiatric manifestations, or treatment effects. This is consistent with previous studies highlighting the multifactorial nature of cognitive impairment in DLB [37, 38].

Sleep disturbances have been shown to reduce functional independence and increase caregiver burden, thereby impacting overall quality of life [39, 40]. Given the frequency, multiplicity, and clinical relevance of these symptoms, their systematic identification is essential.

Using latent class analysis, we identified three distinct sleep disturbance profiles: a “Paucisymptomatic” class, a “RBD – Nightmares” class, and a “Global sleep impairment” class. The burden of sleep disturbances was not associated with disease duration, which suggests that our profiles reflect different phenotypes rather than different disease stages. Additionally, the two extreme groups in terms of sleep symptom burden (“Global sleep impairment” and “Paucisymptomatic”) did not differ regarding MMSE, with worse cognitive performance than the “RBD – Nightmares” profile. This dissociation between sleep symptom severity, disease duration, and cognitive impairment supports the hypothesis that distinct sleep phenotypes may underlie different clinical phenotypes.

The “Global sleep impairment” profile was associated with a higher prevalence of daytime visual hallucinations and cognitive fluctuations. Prior studies have reported this association between sleep disturbances and neuropsychiatric symptoms. The presence of an instability of rest-activity rhythm measured with actigraphy has been correlated with hallucinations and cognitive fluctuations [41]. Similarly, in another actigraphy study, nocturnal activity was associated with delusions and reduced sleep time with hallucinations [42].

The heterogeneity of sleep profiles observed in this study may reflect differences in the distribution of underlying neuropathology in DLB. Current models propose distinct pathways of α-synuclein propagation, including “brain-first” and “body-first” forms, which differ in the relative involvement of cortical, brainstem, and peripheral autonomic structures [43, 44]. These patterns of pathological spread have been proposed to account for part of the clinical variability observed across patients with DLB. Sleep disturbances in DLB involve neural circuits distributed across multiple levels of the nervous system, including brainstem structures regulating REM sleep and autonomic functions, as well as cortical networks involved in perception and cognition [45–47]. Accordingly, different sleep symptoms may reflect dysfunction at different anatomical levels and could therefore be compatible with variability in the distribution of pathology. In addition, concomitant pathologies such as Alzheimer’s disease–related changes are common in DLB and may contribute to clinical variability, including sleep manifestations [48, 49]. Some neuropathological studies have also reported an inverse relationship between RBD and the burden of AD copathology [50].

In addition to pathophysiological differences, other factors may contribute to individual vulnerability to sleep disturbances, including chronotype, premorbid sleep history, environmental factors, and genetic predispositions.

Limitations

First, this study has several limitations related to its retrospective design. Sleep disturbances were evaluated during routine clinical consultations and documented by the treating physician, based primarily on clinical interviews and, when appropriate, specific questionnaires. This variability in assessment methods may have introduced heterogeneity in reporting and interpretation. In particular, probable RBD was defined clinically and therefore may have limited specificity, as other nocturnal phenomena may mimic dream-enactment behaviors. Because sleep variables were collected retrospectively from medical records, absence of documentation may not always reflect the true absence of symptoms. The differential handling of missing data for common versus less prevalent symptoms may therefore have introduced some degree of bias. Underrecognition of sleep symptoms cannot be excluded, particularly in patients with more advanced cognitive impairment. In addition, sleep symptoms included in the LCA were defined as binary variables (presence or absence) and therefore did not capture the severity of the underlying sleep disorders. However, this approach enabled the analysis of a large real-world clinical cohort and the assessment of a broad spectrum of sleep symptoms.

Pharmacological treatments were heterogeneous, and detailed dosage information was not consistently available, which limited our ability to precisely explore the relationship between medication exposure, sleep symptoms, and latent sleep profiles. Cognition was assessed using the MMSE, a screening tool that does not capture precisely the cognitive impairment profile in DLB.

Second, diagnoses were based on clinical assessment in the absence of neuropathological confirmation. Nevertheless, all included patients fulfilled the most recent diagnostic criteria for probable DLB as defined by McKeith et al. (2017), which have demonstrated a diagnostic accuracy of 79% in a neuropathologically confirmed cohort [1, 51]. We did not have systematically available data on AD copathology, which has been shown to influence sleep symptoms and rest-activity patterns. The implementation of specific biological markers, including seed amplification assays that are now becoming available, will be essential to improve diagnostic accuracy. In this context, our work provides a strong foundation for future prospective studies integrating biological confirmation. Future studies integrating sleep phenotyping with biological and imaging biomarkers will be important to explore the underlying mechanisms associated with the identified sleep profiles. Due to limitations in the available DLB cohort with sleep data, we could not validate our identified sleep profiles using an independent replication cohort. Finally, only a subgroup of patients had available PSG data, likely reflecting a selection bias toward those with more pronounced sleep disturbances and without severe dementia. Systematic PSG evaluations in larger cohorts would provide objective measures of sleep and enhance understanding of sleep architecture and underlying pathophysiological mechanisms; however, this may not be feasible in patients with advanced cognitive decline. Alternatively, home-based longitudinal sleep assessments may offer additional insights into the influence of sleep on DLB symptomatology and disease progression [52, 53].

Conclusion

Our study provides new insights into the clinical heterogeneity of DLB and highlights the complex interplay between sleep symptoms and the core clinical features of the disease. We identified three distinct sleep profiles, each defined by specific sleep symptoms and different levels of sleep symptom burden, which were associated with distinct patterns of core clinical features. These profiles may correspond to previously proposed disease models or may reflect modifying factors influencing DLB phenotypes. From a clinical perspective, our findings emphasize the importance of systematically assessing sleep symptoms in patients with DLB. These manifestations may provide meaningful information about disease presentation and clinical heterogeneity and help identify potentially modifiable contributors to patient and caregiver burden. Further biomarker, neuropathological, and molecular studies are warranted to clarify the underlying mechanisms linking these sleep profiles to disease expression.

Authors’ contributions

LR, MPD, CP, and AV designed the study. LR, CA, KG, OV, AD, CH, JD, EC, CD, and AV performed the acquisition of the data. LR, AV, MPD, and CP analyzed and interpreted the data. LR and AV wrote the manuscript. All authors reviewed the manuscript.

Funding

This study was funded through a Société Française de Recherche et Médecine du Sommeil Grant 2022. Agathe Vrillon is funded through the Foundation Alzheimer Young Researcher Program (Grant 1310194). The funding sources had no role in the study design; in the collection, analysis, and interpretation of data; in the writing of the report; and in the decision to submit the article for publication.

Data availability

The dataset used during the current study is available from the corresponding author on reasonable request.

Declarations

Ethics approval and consent to participate

All patients provided written informed consent for data use within the BioCogBank protocol, a prospective clinical and biological cohort study on neurocognitive disorders (sponsor: AP-HP #APHP23099, NCT06244875), following the principles of the Declaration of Helsinki. The protocol was approved by the French ethics committee CPP Est III (23–10-02, 2023-A01413-42).

Competing interests

The authors declare no competing interests.