Clinical predictors of insomnia in borderline personality disorder: a polysomnographic and subjective examination

Abstract

Background

Sleep disturbances are common in patients with borderline personality disorder (BPD) and are associated with a poor prognosis and symptom severity. Research findings on sleep abnormalities in individuals with BPD have been inconsistent, with limited evidence linking subjective and objective measures.

Methods

We compared 44 women with BPD with 41 healthy controls. We examined differences (using ANCOVAs and ordinal logistic regression) and associations (using correlations) between objective sleep assessment (polysomnography) and subjective measures (Consensus Sleep Diary, Insomnia Severity Index, Pittsburgh Sleep Quality Index). We explored predictors of insomnia in BPD patients, including BPD severity, symptomatology, comorbid conditions, and medication use, via standard least squares regressions and ANOVAs.

Results

A total of 22% of patients with BPD had clinically significant insomnia (cut-off ≥ 15), 85% reported mild (subthreshold) insomnia (cut-off > 10) (Insomnia Severity Index; ISI), and 94% reported sleep quality disturbances (PSQI > 5). Compared with those in HC, PSG results in individuals with BPD revealed a longer duration in bed, longer sleep period, REM latency, wake after sleep onset latency, Stage N1 sleep duration, shorter N2 sleep duration, and, with age, more arousals and awakenings. The correlations between objective and subjective sleep measures were weak in both groups. In patients with BPD, a greater degree of insomnia predicted a reduction in total sleep time and increased awakenings and arousals on PSG. Clinical BPD severity, emotional reactivity and dysregulation, depression symptoms, posttraumatic stress symptoms, alexithymia, and presleep arousal were associated with greater insomnia in BPD patients.

Conclusions

Our study confirmed high rates of insomnia and sleep disturbances in individuals with BPD, which contrasted with relatively minor PSG alterations. Clinical BPD severity and symptomatology are associated with sleep abnormalities in individuals with BPD. Insomnia is a neglected yet important characteristic of the BPD phenotype, warranting more attention in future research and clinical guidelines.

Introduction

Borderline personality disorder (BPD) is a complex psychiatric disorder characterised by instability in affect regulation, impulse control, interpersonal relationships, and self-image [1, 2]. Patients with BPD significantly impact mental health care, accounting for approximately 10% of psychiatric outpatients and 20% of inpatients [3, 4]. Core symptoms include disturbed emotion processing and impaired emotion regulation, leading to high-risk behaviours such as substance abuse, binge eating, self-harm, and suicidal ideation [2, 5]. Comorbidities, such as major depressive disorder (MDD, 41–83%) and posttraumatic stress disorder (PTSD, 46–56%), are common [2, 6,7,8,9,10]. These conditions complicate the identification of sleep disturbances in patients with BPD, as both MDD and PTSD are also linked to sleep problems [11]. However, even without these comorbidities, there is a significant correlation between sleep disturbances and borderline symptoms [11].

Subjective sleep prevalence in individuals with BPD

Up to 95.5% of individuals with BPD report subjective sleep quality disturbances, such as difficulty falling asleep, interrupted sleep, daytime dysfunction, use of sleep medication, and reduced sleep duration compared to HCs [12,13,14]. Meta-analytic results confirmed significantly poorer sleep quality and longer sleep onset in self-reported data from individuals with BPD compared to HCs [15]. Additionally, maladaptive sleep-related cognitions were more commonly reported in nonrecovered patients with BPD (83.1%) than in recovered patients (54.3%) [16]. Further, it has been estimated that 63% of BPD patients experience at least one symptom of insomnia nearly every night (i.e., trouble falling asleep, staying asleep, or waking up too early) [17, 18]. Nevertheless, prevalence rates for clinically relevant insomnia are still lacking for population with BPD, underscoring the need for studies with larger samples, both with and without sleep quality and insomnia complaints, to provide a clearer understanding of the true prevalence of these disorders in this population [12, 19, 20].

Objective sleep assessments in individuals with BPD

Aside from inconsistencies in the subjective prevalence of sleep disorders in BPD, a further challenge in the field pertains to research on objective sleep abnormalities, which remains limited and inconsistent [21, 22]. A systematic review and meta-analysis by Winsper et al. (2017) examined the sleep characteristics of individuals with BPD. They analysed 32 studies and found significant differences between BPD patients and healthy controls (HC) in objective (i.e., polysomnography (PSG) and actigraphy) [15]. Specifically, compared with HC, BPD patients were associated with significantly poorer sleep continuity (e.g., longer sleep onset latency (SOL), reduced sleep efficiency (SE) and total sleep time (TST), increased time awake after sleep onset (WASO) and number of awakenings (NWAKE)), with significant moderate effect sizes for some of these sleep indices [15]. In contrast, few consistent alterations in sleep architecture have been reported (e.g., reduced REM latency and increased REM density) [15]. Contradictory architectural findings have been documented regarding deep sleep (i.e., stage 3/slow wave sleep (N3/SWS)) in BPD patients [15, 22, 23].

Subgroup analysis of the meta-analytic data indicated that BPD patients had shorter REM sleep latencies than HC did, independent of depressive status [15]. Moreover, a longer REM latency in BPD patients than in depressed BPD controls has been documented [12], whereas other studies reported no differences [24]. All meta-analytic results were independent of concomitant psychotropic medication, and most studies reported a two-week washout period [15]; nevertheless, the effect of medication on sleep is vital for successful treatment of comorbid disorders, such as depression, and should therefore be taken into account [25].

Subjective and objective sleep assessment associations and discrepancies

Previous research has highlighted a significant discrepancy between subjective insomnia symptoms and objective sleep abnormalities, as measured by PSG, leaving the relationship between these two aspects unresolved [15, 16]. Meta-analytical data have revealed that while objective and subjective measures consistently show poorer sleep quality and longer sleep onset for individuals with BPD than HC do, subjective total sleep time and sleep efficiency do not significantly differ between groups, as objective data do [15]. These inconsistencies create a gap in understanding sleep disturbances in BPD, particularly since few studies have explored both subjective and objective measurements simultaneously [15]. Therefore, investigating whether validated subjective measures of sleep disturbances correlate with objective PSG sleep parameters is crucial for deepening our understanding of sleep problems in BPD and how they relate to insomnia constructs [20, 26].

Clinical predictors of insomnia and sleep disturbances (objective and subjective) in individuals with BPD

Spielman and Glovinsky’s model of insomnia highlights the interplay of predisposing factors (e.g., biological traits, personality traits, social factors), precipitating factors (e.g., medical and/or psychiatric illnesses, stressful life events), and perpetuating factors (e.g., excessive time in bed, napping, worry, rumination, and maladaptive perceptions about sleep) in its development and maintenance) [27, 28]. While insomnia is well-documented as a symptom or consequence of BPD, the reverse relationship—how clinical characteristics of BPD contribute to predisposing or precipitating factors of insomnia—remains underexplored [19, 20, 29, 30].

Most sleep studies on BPD have focused on examining the prevalence of sleep disturbances and their perpetuating factors [31,32,33,34]. For example, poor sleep quality is associated with greater functional impairment and increased emotional dysregulation [14, 17, 18], and maladaptive cognitions about sleep have been proposed as potential perpetuating factors impeding recovery [14, 15]. Nevertheless, the opposite direction has not yet been described in detail for the BPD population; in other words, how critical clinical variables in the psychopathology of BPD (i.e., emotion dysregulation, alexithymia, and arousal) are associated with precipitating or predisposing factors (i.e., trauma) of insomnia has not yet been explored [27, 35].

Emotion dysregulation and alexithymia, for example, may act as predisposing factors when they present as stable, enduring traits rooted in personality or developmental vulnerabilities [36]. Conversely, these features may serve as precipitating factors when triggered by acute stressors such as trauma, psychiatric comorbidities, or significant life events that intensify arousal and emotional instability [37]. This bidirectional relationship between psychopathological variables in BPD and the predictors of clinical insomnia remains insufficiently understood. This study, therefore, aims to explore this gap by investigating the association between clinical characteristics of BPD and insomnia. Such insights are crucial for advancing diagnostic and therapeutic approaches, especially given that untreated insomnia can exacerbate BPD symptoms by further impairing emotional regulation (15).

As summarised above, current state-of-the-art reports show a need for a clearer prevalence of subjective and objective sleep complaints and a mismatch between these two types of assessments, supporting the need for larger observational studies to be conducted. Furthermore, no previous report has comprehensively investigated how predisposing or precipitating clinical measures are associated with sleep disturbances in BPD patients. Bridging this gap may add to the literature and inform current clinical trials (i.e., cognitive behaviour therapy for insomnia, CBTi) on which factors are relevant and associated with insomnia in this population, which may lead to novel sleep-targeted (psychological or pharmacological) interventions that consider how clinical measures are associated with insomnia at the treatment baseline.

Methods

Aim and design

The aims of this cross-sectional case-control study are as follows:

First, to compare the prevalence rates between the two study groups regarding the Insomnia Severity Index (ISI) with two cut-offs, ≥ 15 (primary) and > 10 (secondary), the Pittsburgh Sleep Quality Index (PSQI), clinical sleepiness using the Epworth Sleepiness Scale (ESS), Presleep Arousal Scale (PSAS) and Morning-Eveningness Questionnaire (MEQ) were used (RQ1.1 Table 1). Second, to examine cross-sectional subjective (Consensus Sleep Diary; CSD) and objective (PSG: continuity and architecture) differences in sleep parameters between participants with BPD and HC assessed on the same night should be examined (RQ1.2 Tables 2 and 3). Our expected findings include higher insomnia scores, lower subjective sleep quality and higher presleep arousal scores in individuals with BPD than in HC. We expect to find worse sleep continuity measures and abnormalities in REM sleep architecture in the BPD cohort than in the HC [15, 21, 23, 38].

Table 1 Subjective sleep questionnaires at baseline

Table 2 Subjective sleep diary (SCD) with age as a covariate

Table 3 Objective sleep (PSG) continuity and architecture with age as covariate

Moreover, our study contains the following exploratory analyses:

To explore correlations between subjective and objective sleep measures in each group separately, we expect subjective and objective sleep quality and sleep onset to correlate (RQ2 Supplementary Table 1.1 & 1.2).

Further exploratory analysis pertaining only to our BPD cohort included the following: (RQ3) To explore (a) whether the ISI and PSQI are associated with PSG outcomes (RQ3.1); (b) whether BPD severity is associated with objective (PSG) outcomes (RQ3.2 Table 4); and (c) whether clinical measures are associated with insomnia (RQ3.3 Table 5). We hope to find the PQSI and ISI to be associated with objective measures such as TST, SOL or indices or arousals and awakenings that could be indicative of objective sleep disruptions in BPD patients. Furthermore, we hypothesise that BPD severity is associated with objective markers of sleep disturbances. Moreover, we hypothesise that depression, emotional dysregulation, presleep arousal, posttraumatic stress, alexithymia, emotional reactivity and BPD severity are associated with insomnia as potential precipitating factors and childhood trauma as a predisposing factor for insomnia.

Table 4 BPD clinical severity and its association to PSG outcomes

Table 5 Clinical measures association to insomnia in individuals with borderline personality disorder

Table 6 Demographic variables

Finally, we will conduct a subgroup analysis (RQ4) to explore whether PSG sleep measures differ between HC and participants with BPD, with or without comorbid MDD, PTSD, or past-month reports of suicidality (including impulses, thoughts, intentions, and/or actions). On the basis of the findings from the meta-analysis by Winsper et al. (2017), we do not expect to find significant differences between these subgroups [15].

Participants

Sixty-five women with BPD and 65 HC were recruited. All women on the waiting list for participation in the Dialectical Behaviour Therapy (DBT) program of the University Psychiatry Hospital Duffel, who were externally recruited via advertisements and referred via external sources, were invited to participate. The recruitment rate was 60% for inpatients, 14% for outpatients, and 26% for those on the waitlist. The inclusion criterion for participants with BPD was a baseline symptom severity of ≥ 20 on the BPD Severity Index (BPDSI-IV), which effectively distinguishes BPD from other personality disorders [39, 40]. The exclusion criteria were comorbid lifetime diagnosis of autism spectrum disorder, schizophrenia, schizoaffective disorder, schizophreniform disorder, bipolar disorder type I or delusional disorder (confirmed by DSM-5 criteria assessed via a semistructured clinical interview) and daily usage of more than one antidepressant or more than one antipsychotic drug at the time of PSG assessment, except for mirtazapine, agomelatine, and trazodone [41, 42].

The inclusion criteria for the control group cohort were as follows: no lifetime or current diagnosis of any psychiatric condition as confirmed by the Rapid Measurement Toolkit-20 [43], no use of any psychopharmacological medication, and no sleep disorders confirmed by cut-off scores on the ESS score < 10 [44], PSQI [45] score ≤ 5 [46] and ISI score ≤ 10 [47,48,49].

The exclusion criteria for all participants were the use of illicit drugs or alcohol < 48 h prior to the study evaluation and any use of benzodiazepines or GABA receptor-positive allosteric modulators within less than five times the half-life of the compound prior to the day of the study evaluations, current pregnancy, active shift work, current BMI < 17 or > 35 kg/m², history of seizure disorders or any other major neurological illness or any symptomatic/unstable/uncontrolled concomitant disease (e.g., renal failure, hepatic dysfunction, cardiovascular disease). Drug and pregnancy tests were conducted before testing.

Procedure and setting

Objective sleep was measured via polysomnography (PSG) assessment over two consecutive nights. The first night served to control first-night effects [50] and to rule out clinically relevant periodic limb movement disorder (PLMD) (PLM arousal index > 25/h) and/or sleep apnea (apnea-hypopnea index ≥ 15/hr). PSG took place at the sleep lab of the University Psychiatry Hospital Duffel. PSG recordings were recorded (30 min flexibility) between 23:00 (lights out) and 7:00 (lights on). Ratings on night segments were reviewed with an independent PSG scorer from the sleep laboratory at Antwerp University Hospital (UZA) at three different study time points. In cases of discrepancies, academic resources were consulted until an agreement was reached. Upon waking, the participants filled out the morning section of the sleep diary (CSD) [47, 48], offering detailed information about different aspects of their sleep from the previous night. Only the data from the second-night PSG were used for analysis. As part of the baseline procedures, all the participants completed a self-reported battery of clinical and sleep measures.

Measures

Subjective sleep assessments

The sleep assessment battery used to assess the prevalence of sleep disturbances in both groups (RQ1.1) included the following: the ISI, PSQI, ESS, PSAS and MEQ.

The ISI is a self-report questionnaire used to assess the nature, severity, and impact of insomnia [48, 49] over the past two weeks, with seven items assessing difficulty falling asleep, staying asleep, and early morning awakenings, as well as dissatisfaction with sleep, daytime impairment, perception by others, and distress. The respondents rated each item on a 5-point scale from 0 to 28. A cut-off score of ≥ 15 (moderate to severe insomnia) was chosen as the primary outcome to rule out the presence of clinically significant insomnia, as previously suggested in clinical populations and to prevent false-positive rates in primary care populations [51]. Further, a cut-off of > 10 as a secondary outcome to detect mild (subthreshold) insomnia was used as a secondary measure, in line with other BPD studies [30, 38, 49]. We examined both cut-offs in our sample to examine the percentage in our sample that would endorse mild to moderate insomnia.

The PSQI [45] is a 19-item self-report questionnaire assessing sleep quality and disturbances experienced within the past month. The questionnaire includes a mix of open-ended and Likert scale items. Seven components are derived from individual item scores, culminating in a total score ranging from 0 to 21. The PSQI was chosen for its ability to measure aspects of sleep beyond insomnia severity. A score exceeding 5 indicates suboptimal sleep quality and is chosen as a cut-off [49].

The PSAS comprises 16 items and was used to assess presleep arousal [52]. Responses were rated on a scale from 1 (‘not at all’) to 5 (‘extremely’), with total scores ranging from 8 to 40.

The MEQ [53] is a 19-item tool that evaluates a person’s tendency toward being in the morning or evening (α = 0.87). The total score is calculated by summing the item responses, with higher scores reflecting a morning preference and lower scores indicating an evening preference.

To assess RQ1.2, the sleep diary (CSD) [54, 55] was administered the morning following PSG assessment. Upon waking, the participants completed the morning portion of the diary, providing detailed information about various aspects of their previous night’s sleep. Sleep quality and other open-ended questions related to pre- and post-sleep symptomatology, perceptions and circumstances were rated on a scale ranging from 0 (very poor/not rested) to 5 (very good/very well rested).

Battery of clinical assessments

The clinical assessments used in RQ3.3 include the Beck Depression Inventory (BDI) [56], a 21-question questionnaire, each offering four statements and asking respondents to choose the one that best represents them. These statements range in severity, with higher values reflecting more severe depressive symptoms.

The Difficulties in Emotion Regulation Scale (DERS) [57] is a 36-item scale with a 5-point rating ranging from almost never to almost always. It evaluates emotion regulation problems, with higher scores indicating greater problems with emotion regulation. The DERS contains six subscales that reflect that multifaceted definition of emotion regulation; these are (1) nonacceptance of emotional responses (NONACCEPTANCE), (2) difficulties engaging in goal-directed behaviour (GOALS), (3) impulse control difficulties (GOALS), (4) lack of emotional awareness (AWARENESS), (5) limited access to emotion regulation strategies (STRATEGIES), and (6) lack of emotional clarity (CLARITY).

The Emotion Reactivity Scale (ERS) is a 21-item self-report questionnaire designed to assess three key aspects of emotional reactivity: sensitivity, intensity, and persistence of emotions [58].

The Child Trauma Questionnaire (CTQ) [59] is a 28-item self-report tool that retrospectively assesses childhood abuse and neglect across five subscales: emotional, physical, and sexual abuse, along with emotional and physical neglect. The responses are rated on a 5-point scale, with higher scores indicating greater trauma. The tool is reliable and valid.

The Self-Rating Inventory for PTSD (ZIL) [60] is a 22-item questionnaire that is assessed on a 4-point scale ranging from ‘not at all’ to ‘very much’ to what extent the person experienced PTSD complaints in the past 4 weeks. A cut-off score of 51 is a good predictor for a clinical PTSD diagnosis.

The Toronto Alexithymia Scale (TAS) [61] is a 20-item questionnaire consisting of a 5-point scale ranging from “strongly disagree” to “strongly agree”. Higher scores indicate problems identifying and describing emotions, with a cut-off score of 61 indicating alexithymia.

The Borderline Symptoms List (BSL-23) [62] is a 23-item instrument that assesses self-rated past-week borderline symptomatology. The BPDSI-IV [39, 40] is a semistructured interview evaluating the frequency and intensity of BPD symptomatology over a specific 4-week period.

Objective sleep assessment: PSG

The PSG encompassed six main leads of EEG activity (EEG: F3-M2, F4-M1, C3-M2, C4-M1, O1-M2, O2-M1, bilateral electro-ocular activity (EOG), electrocardiogram and submental and leg electromyogram (EMG) (right and left anterior tibial muscles)). Respiration (oral and nasal airflow, chest and abdominal movements, oxygen saturation, and snoring) was recorded on both nights and scored in 30-second epochs according to the rules, terminology and specifications of the Sleep Research Manual of Interpretation of Sleep and Associated Events version 2.6 (AASM) [63] by experienced scorers.

To assess RQ1.2, the following PSG outcome measures were considered for analysis, most of which are related to the total sleep time (TST), which is calculated as the total time spent in sleep excluding any wake epochs (a specific 30-second time unit used to analyse and score sleep data according to the ASSM).

• (a) Sleep architecture measures. Proportionate to TST, all architectural variables were calculated as the percentage of rapid-eye-movement sleep (REM), stage 1 nonrapid-eye-movement sleep (N1), Stage 2 nonrapid-eye-movement sleep (N2), Stage 3 nonrapid-eye-movement sleep or slow wave sleep (N3), also known as slow-wave-sleep (SWS). Further measures analysed were spindle density in N2 (a specific train of distinct waves with a frequency of 11–16 Hz, detected using a real-time spindle detector via BRAIN RT^® software), REM sleep latency after sleep onset, and wake latency after sleep onset.

• (b) Sleep continuity measures. The time in bed (TIB) from lights out to lights on, total sleep time (TST), and sleep period time (SPT), which is defined as the total sleep time, including wake time after the first sleep epoch. Further parameters included sleep onset latency (SOL), defined as the time elapsed between the start of bedtime and the start of the first sleep epoch; sleep efficiency (SE), defined as the total time of sleep without wake epochs (TST) divided by the time spent in bed; defined as the sum of N3 and REM in minutes divided by TST in minutes; wake after sleep onset (WASO) latency, including all wake activity (WASO), including time out of bed; the number of awakenings per hour of total sleep time (NWAKE); and arousal and awakening indices. Arousals are defined as abrupt shifts in EEG frequency, including alpha, theta and/or frequencies greater than 16 Hz (but not spindles) [63].

Data processing and statistical analysis

A sample size of 65 patients and 65 controls provides 80.8% power (two-tailed) to detect a medium-sized effect in an independent sample t-test with age-matched participants. A post hoc power calculation for our final primary ANCOVA (with age as a covariate) was performed on the final sample sizes (HC: n = 41, BPD: n = 44). To achieve the same power level as the prospective analysis (80%), the calculation indicated a medium-sized effect (Cohen’s f = 0.31) via G*Power (version 3.1.9.4). Age was included as a covariate because age matching could not be fully achieved due to funding and time limitations. Data are expressed as the mean ± standard deviation or as a percentage unless otherwise indicated. Cases with missing data were excluded on an analysis-by-analysis basis. We did not remove any outliers, but the outcome measures were either log-transformed or reflected and square-rooted if not normally distributed [64]. All the statistical analyses were performed in JMP^®, Version Pro 17. SAS Institute Inc., Cary, NC, 1989–2023.

To test for group differences, independent samples t-tests (for continuous variables), chi-square tests (for categorical variables) or Mann‒Whitney U tests (for nonnormally distributed continuous variables) were performed (Tables 1 and 2). To answer RQ1 and test between-group differences in objective continuity and architecture PSG parameters and subjective sleep parameters, we performed the following: 1) ANCOVA with group as the independent variable, sleep parameters as the dependent variables and age as the covariate (Tables 3 and 4) and an ordinal logistic regression model with age as a covariate and group as the independent variable (Table 3).

To answer RQ2, we calculated the Spearman correlation coefficient (r), which was used to evaluate the (multivariate) correlation between macrostructural and subjective sleep outcomes in both groups separately (secondary analysis). To adjust the P values according to the false discovery rate (FDR), the Benjamini‒Hochberg (BH) procedure was used. Correlation coefficients above |0.3| are reported [65], with values of weak <|0.4|, moderate <|0.7| and strong > |0.7 [66] (Supplementary Table 1.1 and 1.2, page 4).

RQ3 pertained exclusively to our BPD cohort. To answer this research question, we conducted three standard least squares regressions to explore (a) whether the ISI and PSQI scores are associated (dependent variable) with PSG outcomes (independent variables) (RQ3.1); (b) whether BPD severity (dependent variable) is associated with objective (PSG) outcomes (independent variables) (RQ3.2 Table 5.1); and (c) whether clinical measures (independent variables) are associated with insomnia scores on the ISI (dependent variable) (RQ3.3 Table 5.2). At least ten observations were used per variable to prevent overfitting our models.

To answer RQ4, exploratory analyses via ANOVAs were conducted to explore whether PSG sleep parameters (independent variables) differed between subgroups (dependent variables) of HC and participants with BPD with and without clinical comorbid PTSD, MDD, past month reports of current NSSI or suicidal behaviour (impulses, thoughts, intentions and/or actions), and medication use (antidepressants and antipsychotics). To compare the three subgroups, we used Tukey/Steel–Dwass tests. We adjusted the P values by applying the Benjamini‒Hochberg false discovery rate (FDR) procedure to correct for multiple testing (Supplementary Table 2.1 and 2.2, pages 5 and 6).

Results

RQ1.1 Demographics/descriptive statistics

A total of 130 women (65 BPD patients, 65 HC) were enrolled between March 2021 and September 2023. The dropout rate was 29% in the control group (n = 19) versus 23% in the BPD group (n = 15). The final sample consisted of 44 participants with BPD (mean age 31.81 ± 11.23 years) and 41 with HC (mean age 26.21 ± 6.95 years). The full sociodemographic data, comorbidities and use of medications are displayed in Table 1. See Supplementary Fig. 1 for a detailed study recruitment flow diagram, page 3.

RQ1.2 subjective sleep parameters in BPD patients vs. HC

Our results were consistent with our hypotheses, where on the basis of the subjective sleep reports, we found that in our BPD sample, 53% of the participants scored above the threshold for clinically relevant sleepiness (ESS) [44], 22% had clinically significant insomnia (cut-off ≥ 15), 85% reported mild (subthreshold) insomnia (cut-off > 10) and 94% had disturbed sleep quality (PSQI) (Table 2). On the basis of the CSD morning self-reports after the PSG assessment, patients with BPD reported significantly worse sleep quality, waking up earlier than expected, not feeling rested or refreshed, worrying when going to bed, dreaming more, thinking more at night, not sleeping through the night or until the moment planned, and personal circumstances (such as pain, transition symptoms, stressful events) that affected sleep compared with HC. Age had a significant main effect on self-reported dreams, with older participants reporting more dreams during the night than younger participants did (Table 3).

RQ1.2 objective sleep PSG assessments in BPD patients vs. HC

Compared with those of HC and as hypothesised, we found worse sleep continuity measures in participants with BPD, who spent significantly more TIB, had a longer SPT and took a longer time to have a wake epoch after sleep started (WASO) latency. However, there were no significant differences between the groups in SE, TST, SOL, or WASO (Table 4). With respect to sleep architecture, our findings did not support our hypothesis. Compared with HC, participants with BPD presented significantly greater proportions of N1 and reduced N2 sleep and longer REM sleep latency, but there were no significant differences between the groups in the N3 or REM sleep proportion, as hypothesised (Table 4). A significant main effect of age was present for SE, WASO, N1%, and N3%, with older participants presenting lower SE and N3% values and longer WASO and N1% values than younger individuals did (list of abbreviations in Supplementary file 1 pages 1 and 2).

Significant interaction effects between group and age were found for the TST arousal index and the NREM arousal and awakenings index, which increased with age in individuals with BPD but decreased with age in HC (Table 4).

RQ2. Correlations between objective and subjective sleep parameters

In both cohorts, the objective (PSG) and subjective (CSD) TSTs were only weakly correlated. In the control group, a weak negative correlation between objective WASO and self-reported TST (CSD) was found. In the BPD group, objective and subjective SOLs were moderately correlated. Furthermore, the correlations between subjective WASO and objective NWAKE (PSG) in the BPD group were weak (Supplementary Tables 1.1 and 1.2, page 4).

Fisher’s z-value analysis indicated no significant differences (p > 0.05) in the correlation coefficients between the two samples in the following parameters: objective (PSG) and subjective (CSD) TST (p = 0.072), objective WASO and subjective TST (CSD) (p = 0.077), and subjective WASO and objective NWAKE (PSG) (p = 0.051), indicating that the relationships between the variables are similar across the two groups being studied. The significant Fisher’s z results (p < 0.05) provide evidence that the correlation between these variables differed significantly between two groups: objective and subjective SOL (p = 0.042), and subjective WASO and objective SOL (PSG) (p = 0.037), indicating that the mechanisms linking subjective and objective sleep measures differ across the groups (Table 5.1 and 5.2).

RQ3. Exploratory analysis of the BPD cohort: clinical variables and sleep (subjective and objective)

Sleep quality measured by the PSQI was not associated with any objective PSG outcomes. Nevertheless, higher degrees of insomnia on the ISI were associated with reductions in TST [F(1, 42) = 6.171, p = 0.017] and increased awakenings and arousals [F(1. 42) = 5.22, p = 0.027] (RQ3.1).

Higher BPD symptom severity scores were associated with lower TST, SE, %N3, and WASO latency and higher %REM and %N2 spindle density (Table 5.1). BPD symptomatology scores were not associated with any subjective CSD outcomes (SOL, TIB, TST, NWAKE, or WASO) (RQ3.2).

The past-month severity of BPD (BPDSI-IV), emotional reactivity (ERS), presleep arousal (PSAS), emotional dysregulation (DERS): GOALS subscale, alexithymia (TAS), and comorbid depression (BDI) and PTSD symptomatology (ZIL) were positively associated with subjective insomnia complaints. Childhood trauma (CTQ), Borderline Symptoms List (BSL-23), and DERS total score and subscales: NONACCEPTANCE, IMPULSE, AWARENESS, STRATEGIES, and CLARITY were not significantly associated with subjective insomnia complaints in individuals with BPD during the past two weeks (RQ3.3) (Table 5.2).

RQ4. Depression, PTSD, suicidality and medication use in the HC and BPD subgroups

Compared with HC and BPD patients not on antipsychotic medication, BPD patients receiving antipsychotics reported worse sleep quality (Z = 2.838, p = 0.012) [χ2(2, N = 83) = 10.973, p = 0.004] (Supplementary Table 2.1 page 5).

Similarly, participants with BPD taking antidepressants had longer REM sleep latency (Z= -2.624, p = 0.023) than the other two groups did [χ2(2, N = 84) = 14.1654, p = 0.008]. No significant differences were found between subgroups of BPD patients with or without recent suicidality or those with comorbid depression or PTSD (Supplementary Table 2.2 page 6).

Several PSG parameters, including time in bed (TIB), %N1, %N2, %REM, REM latency, WASO latency, and spindle density in %N2, differed significantly (p < 0,05) between controls and BPD patients, regardless of comorbidities or medication use (Supplementary Tables 2.1 and 2.2, pages 5 and 6).

Discussion

This study represents the largest examination of insomnia and sleep in individuals with BPD to date, using comprehensive subjective and objective measures to assess sleep continuity, architecture, and the relationship between clinical measures and insomnia. Our findings not only confirm high rates of subjective sleep complaints but also establish the prevalence rate of the ISI in BPD, with 94% of individuals with BPD reporting disturbances in sleep quality [12] and 22% self-reporting clinically significant insomnia (cut-off ≥ 15) [51]. Notably, 85% of our sample reported mild insomnia (cut-off > 10) [30, 38, 48]. This variability in insomnia cut-off scores highlights the need for future research to establish standardised thresholds for BPD patients.

Subjective sleep complaints vs. objective sleep continuity and architecture in BPD

Despite the high prevalence of subjective sleep disturbances, objective measures (PSG parameters) did not reveal significant differences in SOL, SE, WASO, or the number of awakenings between the BPD and control groups [15]. However, individuals with BPD presented longer durations in bed, longer sleep periods, and WASO latency, as well as increased arousal indices in the TST during NREM stages; these findings partially support our hypothesis. We found no significant disruption in sleep continuity parameters between the BPD subgroups [67]. Notably, arousals increased with age in BPD patients but decreased with age in HC, suggesting age-related differences in sleep fragmentation in NREM sleep.

Our findings suggest that BPD patients experience more disrupted sleep architecture, characterised by increased N1 sleep, reduced N2 sleep, and prolonged REM latency. Our results indicate that BPD patients experience more light sleep (N1) coupled with frequent disruptions (arousals) in NREM, highlighting that the quality and amount of NREM sleep might be a more significant parameter in BPD patients than the hypothesised sleep continuity parameters are [19]. The absence of group differences in SOL and SE in combination with these architectural findings may partially reflect the high levels of insomnia reported subjectively. Accurately diagnosing and treating sleep disorders in individuals with BPD could improve therapeutic outcomes [19]. These disturbances in objective sleep underscore the importance of accurately diagnosing and treating sleep disorders in BPD patients to improve therapeutic outcomes.

Impact of medications on REM sleep

Our analysis revealed that psychotropic medications, particularly antipsychotic medications, significantly affect objective sleep quality in BPD patients. Furthermore, our finding of prolonged REM latency in BPD patients contrasts with meta-analytic data showing shortened REM latency [15] and contributes to mixed results [23]. This discrepancy may be attributed to the fact that most of our patients were medicated, particularly with serotonin reuptake inhibitors (SSRIs), serotonin-norepinephrine reuptake inhibitors (SNRIs), and tricyclic antidepressants (TCAs), which are known to prolong REM latency, reduce REM sleep, and disrupt sleep continuity [41]. Our subgroup analysis confirmed that BPD patients on these medications had longer REM latency than those not taking them [41, 42], emphasising the need to consider the effects of antidepressants on sleep when treating BPD patients with comorbid depression [25].

Discrepancies between subjective and objective sleep measures

A notable finding is the weak correlation between subjective and objective sleep measures in both BPD patients and HC, supporting the concept of “sleep-state misperception“ [68]. This discrepancy, commonly observed in individuals with insomnia and referred to as “paradoxical insomnia” [69], has also been proposed for BPD patients [26, 38]. The weak correlations between subjective sleep reports and objective AASM PSG outcomes in our findings corroborate the notion that physiological sleep and its perception are two distinct dimensions of psychiatric disorders, including BPD [70]. This mismatch underlines the importance of independently assessing both subjective sleep experiences and objective measures.

Role of BPD symptom severity in sleep disturbances

BPD symptomatology significantly impacts perceived sleep quality [19]. Poor sleep quality has been shown to impair recovery in patients with BPD, with nonrecovered BPD patients reporting longer SOLs [14]. While this has primarily been shown through subjective measures, our study additionally confirmed that greater BPD symptom severity is correlated with several AASM-PSG parameters, underscoring the importance of objective sleep assessments at the beginning of treatment. Variables such as TST, SE, spindle density, %N3, and %REM were associated with past-month BPD symptom severity, aligning with meta-regression findings that identified sample type (inpatient vs. outpatient) as a source of heterogeneity. These findings suggest that the more severe the BPD symptoms are, the more disrupted the sleep architecture. Despite the impact of symptom severity in the BPD group, we found no differences between HC and BPD groups regarding potential controlled contextual factors (i.e., noise, light-darkness, heat-cold, etc.), reinforcing the importance of individual clinical profiles in shaping sleep disturbances in BPD patients. Further research using ambulatory EEG wearables in both settings (inpatient and outpatient) is needed to determine whether this heterogeneity between inpatient and outpatient groups could be related to the impact of the degree of BPD symptom severity in inpatient versus outpatient populations or to specific contextual influences.

Clinical (predisposing/precipitating) factors associated with insomnia in BPD

Clinical features such as emotional reactivity, impulse control difficulties, difficulties engaging in goal-directed behaviour, alexithymia, and presleep emotional arousal were significantly associated with insomnia scores in BPD patients in our study [27, 35]. These results indicate that clinical variables may not only function as predisposing and precipitating factors in the development of insomnia but also exacerbate emotional regulation difficulties, which can worsen BPD symptoms [19]. Specifically, difficulties concentrating or engaging in goal-directed behaviour may disrupt sleep through mechanisms such as cognitive hyperarousal, intrusive thoughts that may interfere with the ability to initiate or maintain sleep, and/or unhelpful nighttime habits that disrupt the ability to relax and initiate sleep [71, 72]. These results are significant for diagnostic and treatment studies, given that untreated insomnia can exacerbate BPD symptoms by disrupting emotional regulation [17]. Our study indicates that addressing these factors in interventions aimed at improving cognitive control and impulse regulation during emotional distress might be particularly effective in addressing both insomnia and overall clinical outcomes in BPD patients.

Presleep hyperarousal: a potential predisposing factor for insomnia in BPD patients

Contrary to our hypothesis, exposure to childhood trauma was not associated with insomnia in this study [27]. However, our findings suggest that addressing clinical factors related to hyperarousal, such as trait-level emotional reactivity, could influence both subjective and objective insomnia markers. Hyperarousal is a central factor in the pathophysiology of insomnia [73]. Our self-report findings are consistent with existing literature, highlighting the relationship between presleep arousal and insomnia. Moreover, heightened emotional reactivity and challenges in maintaining goal-directed behaviour may exacerbate presleep cognitive and emotional disruptions, further amplifying hyperarousal [71, 72]. Objectively, AASM-PSG parameters, such as the TST and the frequency of awakenings/arousals, were correlated with insomnia scores in BPD patients, suggesting that individuals with BPD and higher insomnia scores may experience more fragmented sleep, which is consistent with the literature linking sleep fragmentation to insomnia and depression [26, 74]. Notably, using subjective and objective markers of hyperarousal as markers of treatment response in BPD patients could be highly useful for determining whether current treatments effectively reduce hyperarousal at both levels.

Maladaptive sleep cognitions as perpetuating factors of insomnia

Maladaptive sleep cognitions, such as worry about insomnia and the perceived consequences of poor sleep, have been identified as clinical predictors of BPD and may contribute to increased arousal and emotional distress [15, 75]. Our study demonstrated that subjective reports on the CSD sleep diary may reflect not only dysfunctional sleep symptomatology but also a pattern of perceptions and maladaptive cognitions about sleep. These dysfunctional sleep perceptions could hinder recovery, functioning as perpetuating factors in insomnia, which is consistent with the Spielman and Glovinsky insomnia model [14, 15, 27]. Additionally, these insights emphasise the need for targeted interventions for BPD patients, such as CBT-I, to address maladaptive sleep cognitions.

Limitations

The primary limitation of this study is its exclusive recruitment of women, limiting its generalizability to men. Time constraints and the higher prevalence of BPD in women (70%) than in men (30%) drove this decision [3]. Psychotropic medications affect sleep continuity and architecture [41, 76]; discontinuation was not recommended because of the long wash-out period. To control for medication effects, we limited our sample to one class of antidepressants or antipsychotics, allowing for subgroup analyses. Reduced REM latency and increased REM sleep are speculated if participants are unmedicated [41]. Despite subgroup analyses addressing heterogeneity, small sample sizes restrict interpretation [15]. Finally, the PSG data were analysed via conventional AASM sleep scoring.

Directions for future research

Conventional AASM sleep scoring may miss crucial disparities in sleep patterns, which may be important for understanding subjective sleep experiences [77, 78]. Innovative methodologies, such as microstructural scoring with epochs shorter than 30 s, are necessary for detecting fine-grained sleep disruptions in BPD patients. Future research should explore microstructural measures of sleep fragmentation and local sleep across different brain areas to explain the objective-subjective sleep measurement mismatch [68, 77,78,79].

The complex link between BPD and sleep disturbances, influenced by emotional dysregulation and psychiatric comorbidities, necessitates extensive transdiagnostic and longitudinal research [70, 80]. Future studies should routinely include subjective and objective sleep assessments, such as home-tested headbands for ambulatory sleep evaluation [81].

Clarifying the bidirectional relationships between critical psychopathological variables in BPD (e.g., emotional reactivity, pre-sleep arousal, alexithymia, and BPD-specific symptomatology) and clinical insomnia cut-off scores may provide valuable insights into phenotypical subgroups within BPD, aiding in the identification of specific sleep profiles and their clinical implications. These predisposing and precipitating clinical variables are likely to interact dynamically over time, influencing the onset, maintenance, and resolution of insomnia. Targeting these shared vulnerabilities through interventions could simultaneously address underlying mechanisms of emotional dysregulation and insomnia, potentially breaking the cycle of perpetuation [82]. For example, improving emotional reactivity and impulse control may reduce nighttime hyperarousal, a hallmark feature of insomnia, while enhancing goal-directed behaviour and addressing alexithymia could promote adherence to behavioural treatments such as cognitive-behavioural therapy for insomnia (CBT-I). This approach could ultimately refine diagnostic accuracy and inform tailored interventions, paving the way for future research into the dynamic interactions between sleep disturbances and BPD psychopathology.

Clinical recommendations

Current treatment options for BPD often fail to adequately address sleep disturbances despite their high prevalence and significant impact on recovery [83]. For example, Weinhold et al. (2017) reported that sleep structure at baseline plays a crucial role in treatment success during trauma-focused interventions for BPD patients. Specifically, individuals with longer total sleep time and increased REM sleep duration before treatment experienced better outcomes in narrative exposure therapy (NET), particularly in terms of reducing PTSD symptoms. These findings underscore the importance of addressing sleep disturbances early in the treatment process [84].

Given the critical role of sleep in recovery, we recommend routine screening and management of insomnia and sleep quality in all treatment settings for patients with BPD. While sleep hygiene is included in some evidence-based BPD treatments [17, 85], its effectiveness is significantly less than that of CBT-I [86]. Interestingly, CBT-I has been shown to resolve objective‒subjective mismatch in insomnia patients and improve clinical outcomes in individuals with major depressive disorder [87]. Although the effectiveness of sleep interventions specifically for BPD patients with or without chronic insomnia remains unproven, ongoing research is currently investigating the application of CBT-I in this population [88].

Winsper and Tang (2014) proposed that emotional dysregulation may serve as a mechanism through which insomnia exacerbates suicidality in individuals with BPD. This suggests that future interventions focusing on enhancing cognitive control and emotion regulation could be promising avenues for improving outcomes in this population [32]. Currently, tailored interventions for sleep problems in BPD patients have not yet been developed; nevertheless, understanding which clinical factors are linked to BPD could provide a key avenue to use BPD-specialised skills to address emotional dysregulation, such as those specified in dialectical behaviour therapy (DBT), which could indirectly assist in reducing insomnia complaints [85].

Conclusion

Our study confirms that high rates of subjective sleep complaints are present in individuals with BPD, with 85% self-reporting clinically significant insomnia and 94% reporting sleep quality disturbances. Standard PSG parameters fail to capture sleep disturbances specific to BPD, whereas subjective reports do, with few weak correlations between them. In individuals with BPD, PSG parameters are predicted by clinical severity. Higher clinical severity, emotional dysregulation, depression, PTSD, alexithymia and presleep arousal were associated with higher insomnia scores in BPD patients. These findings emphasise the need for innovative methodologies to detect subtle microstructural changes for diagnostic and therapeutic purposes and increase awareness of insomnia as a key clinical characteristic of BPD.