Electroconvulsive Therapy and the Longitudinal Modulation of Sleep Architecture

Abstract

Background:

Sleep plays a critical role in mood regulation, yet the mechanisms linking sleep to mood improvement remain unclear. Electroencephalographic (EEG) slow-wave activity (SWA) during non–rapid eye movement (NREM) sleep is reduced in patients with major depressive disorder. Here, we examined sleep architecture in patients with depression undergoing electroconvulsive therapy (ECT).

Methods:

Sleep EEG data were collected using wireless Dreem devices before and after each ECT session serially throughout the ECT course. Sleep stages were manually scored into rapid eye movement (REM) and NREM stages (N1–N3) according to modified American Academy of Sleep Medicine guidelines. Depression severity was assessed using the 16-Item Quick Inventory of Depressive Symptomatology-Self Report (QIDS). SWA was quantified as the average 0.5-4 Hz frontal EEG power per minute during N2 and N3 sleep.

Results:

Across 214 sleep records from 22 subjects (median 9 per patient), the percentage of time in N3 sleep increased over the course of therapy in the responders’ group (P = 0.030). SWA increased throughout the ECT course across all participants (P = 0.001), and it was negatively correlated with the QIDS score (P < 0.001). Mediation analysis revealed that SWA mediated the effect of treatment on QIDS only in the responders’ group (P = 0.037).

Conclusion:

Our findings demonstrate that ECT increases N3 sleep and SWA in the responders’ group, with SWA changes mediating improvements in depressive symptoms. These findings highlight SWA as a biomarker of antidepressant response and a potential target for sleep-based strategies to augment brain stimulation therapies for depression.

ELECTROCONVULSIVE THERAPY: A CRITICAL BUT POORLY UNDERSTOOD THERAPY

Depression represents a significant global health burden, contributing substantially to mortality through increased suicide risk and medical comorbidities.^1-5 Refractory depression, in particular, poses a formidable clinical challenge, yet electroconvulsive therapy (ECT) remains the most effective intervention.⁶ With response rates reaching 70%–80% and remission rates between 50% and 63%, ECT offers a powerful therapeutic avenue.^7-9 This treatment involves the induction of a generalized seizure through controlled electrical stimulation, typically administered 2 to 3 times weekly. ECT facilitates a rapid symptom reduction often yielding noticeable improvements within the initial sessions.¹⁰

While ECT offers a rapid and highly effective treatment for refractory depression, it involves general anesthesia with an unprotected airway and pharmacologically-induced paralysis, introducing procedural risks such as aspiration pneumonitis.¹⁰ Despite its efficacy, the underlying mechanisms by which ECT-induced seizures exert therapeutic benefits remain poorly understood. Elucidating these pathways would help improve treatment outcomes and minimize adverse effects.

SLEEP DISTURBANCES IN DEPRESSION

Sleep is a complex physiological process affecting the physiology of the entire organism during wakefulness.¹¹ Sleep disturbances and depression are closely linked, and their interconnections are thought to be informative of underlying pathophysiology.¹² Up to 70%–90% of patients with major depressive disorder (MDD) have sleep disturbances, including initial insomnia, shorter sleep duration, early morning awakening, and nonrestorative sleep.¹³ Patients with depression and with suicidal ideation are more likely to have disrupted sleep than those without.¹⁴ Inversely, individuals with sleep disturbances are at a higher risk of developing major depression.¹⁵

Prior work has investigated the relationship between sleep disruption and depression using polysomnography (PSG), which utilizes EEG and other physiologic signals to characterize sleep into periods of rapid eye movement (REM) and various stages of non–rapid eye movement (NREM) sleep.^13,16,17 NREM stage 2 (N2) sleep constitutes the majority of total sleep time (TST), typically accounting for ~45%–55% in healthy adults.^18-20 NREM stage 3 (N3), also known as slow-wave sleep (SWS), comprises about 15%–25% of sleep and offers restorative physiological benefits across multiple organ systems.^18-20

Sleep slow waves are high-amplitude, low-frequency oscillations (< 4 Hz) that predominate during N3 sleep. Sleep slow wave activity (SWA) is quantified as the EEG power of these oscillations. SWA serves as a physiological marker of homeostatic sleep pressure, with prolonged wakefulness leading to increased sleep pressure and a corresponding elevation of SWA during subsequent non-REM sleep.^21-23 Furthermore, SWA has implication in downscaling of synaptic strengths during sleep which is essential for learning and memory consolidation.^24-27 MDD is associated with deficits in neural and synaptic plasticity.^26,28-30 SWA is reduced in individuals with MDD compared with healthy controls especially during the first cycle of NREM sleep.^31,32 Sleep slow waves may also have a lower amplitude in depressed patients compared with healthy controls.³¹ Finally, in patients with depression, the distribution of SWA across sleep is altered by antidepressant medications and differs between pharmacologic responders and nonresponders.³³ Understanding the effects of antidepressants on sleep markers such as SWA may offer insights into the underlying mechanisms of depression and treatment response.

IMPACT OF ELECTROCONVULSIVE THERAPY ON SLEEP STRUCTURE

The effect of ECT-induced generalized seizures on sleep still requires better characterization.³⁴ Generalized seizures in patients with epilepsy have known associated sleep changes, including increases in N2 sleep^35,36 and decreases in SWS.³⁷ Some studies have examined changes in sleep macrostructure before and after an entire treatment course with inconsistent results. A completed course of ECT improved sleep macrostructure, evidenced by increases in TST, sleep efficiency, and augmented daytime activity.³⁸ More specifically, a course of right unilateral ECT was associated with increased slow wave sleep in 15 patients with depression, a finding limited to patients who remitted (N = 6).³⁸ Findings regarding ECT’s impact on REM sleep remain inconsistent. While some studies suggest a suppressive effect on REM sleep, others report increased REM sleep.^38-40 These inconsistent findings may be caused by variability in daytime somnolence and artifacts that impair sleep staging. However, a key limitation across these studies is the discrete nature of sleep assessment. Sleep data were typically acquired only at baseline and end of the ECT course, rather than longitudinally throughout the ECT course, preventing a comprehensive understanding of sleep’s temporal changes and its relationship with the trajectory of depression improvement.

In this study, we sought to better understand the effects of ECT on sleep architecture by examining sleep longitudinally throughout a course of treatment longitudinally. We accomplished a longitudinal study of sleep markers by employing wireless wearable EEG devices for at-home use. We hypothesized that successive ECT treatments would increase SWA over the course of ECT, and that this increase would mediate the therapeutic effects, which would be observed only in responders (Fig. 1).

FIGURE 1.

Hypothesized interaction between electroconvulsive therapy (ECT), sleep slow wave activity (SWA), and antidepressant response. ECT treats depression through the repeated induction of generalized seizures. We hypothesize that ECT enhances sleep slow wave activity, which in turn contributes to synaptic renormalization and neuroplastic changes underlying antidepressant effects. According to this model, improvements in sleep SWA following ECT may mediate downstream reductions in depressive symptoms.

MATERIALS AND METHODS

Participants

All participants were recruited as part of the study protocol for Correlating Electroconvulsive Therapy Response to Electroencephalographic Markers (CET-REM) study between 2020 and 2022.⁴¹ This single-center, prospective, observational study (ClinicalTrials.gov: NCT04451135) examined a clinically ascertained and treated cohort. Participants were recruited from the inpatient psychiatry service and the outpatient ECT referral service at Barnes-Jewish Hospital/Washington University in St. Louis. Inclusion criteria included age 18 and above, a diagnosis of MDD or bipolar disorder (BD) with a depressive episode, and planned initiation of ECT. Potential participants were evaluated by consulting ECT psychiatrists to confirm diagnoses according to the Diagnostic and Statistical Manual of Mental Disorders, Fifth Edition (DSM-5) and eligibility for ECT and ascertain interest in study participation.⁴² Given the observational nature of the study, concomitant pharmacotherapy was managed by the treating psychiatrist as part of routine clinical care. Exclusion criteria include schizophrenia and schizoaffective disorder. Participants were not excluded for psychosis or other comorbid psychiatric conditions.

ECT Procedure

Subjects underwent a course of ECT following usual clinical care, with treatments administered 2 to 3 times per week. The number of treatments, stimulation parameters (pulse width, frequency, and charge), electrode placement, and frequency of sessions were determined by the treating psychiatrist. Three placements were utilized: right unilateral (RUL), bilateral (BL), and bifrontal (BF). ECT stimulation was administered using a Thymatron System IV. Typically, subjects initially received brief pulse (0.3 msec) RUL ECT at 6 times the seizure threshold. If sufficient improvement was not seen after ~6 treatments, subsequent escalations to 100% RUL charge or bilateral ECT were done. General anesthesia during ECT sessions was administered with either etomidate (typically 0.2 mg/kg) or methohexital (typically, 1 mg/kg), with the specific regimens determined by the treating anesthesiologist.

Sleep EEG Measurements and Staging

Overnight sleep macrostructure EEG measures were taken before ECT, as baseline, and then following each ECT session during the ECT course using the Dreem headband (second version), a consumer-grade, wireless, dry-electrode device that records EEG without the need for a continuous internet or Bluetooth connection.⁴³ It is equipped with 4 frontal EEG electrodes (Fp1, Fp2 (ground), F7, and F8) and 2 occipital sensors (O1 and O2). This device has previously been validated against polysomnography for monitoring sleep-related physiological signals and accurate sleep staging across broad age demographics.⁴⁴ A single night of recording being adequate for baseline sleep assessment in perioperative settings,⁴⁵ the Dreem device was used for longitudinal sleep EEG data acquisition throughout an ECT course. In addition, it incorporates a light source for heart rate monitoring and a set of accelerometers to assess movement, which can aid in sleep staging and arousal detection.

Sleep EEG recordings were imported and processed using custom MATLAB (MathWorks, Natick, MA) scripts and EEGLAB. The signals underwent a 0.3–50 Hz bandpass filter and were down-sampled to 250 Hz and exported in European Data Format (EDF). Certified sleep technologists staged the recordings using Philips Respironics Sleepware G3 Software following modified American Academy of Sleep Medicine (AASM) criteria.⁴⁵ Epochs were labeled as nonscorable (NS) if they did not meet the criteria for scoring wakefulness (W), NREM sleep stages (N1, N2, N3), or REM sleep. Total nocturnal sleep time (TST) was determined from the timing of the first and last sleep epochs.

Clinical Response: Definition and Measurement

Depression symptom severity was assessed using the 16-Item Quick Inventory of Depressive Symptomatology-Self Report (QIDS).⁴⁶ This questionnaire evaluates 9 categories of depressive symptoms, including sleep quality, depressed mood, weight and eating changes, concentration, self-image, suicidal ideation, anhedonia, energy level, and psychomotor agitation. Patients completed the QIDS in the ECT suite immediately before each ECT session, in accordance with standard clinical practice at Washington University in St. Louis. Consistent with prior literature, we defined ECT responsiveness as at least a 50% decrease in QIDS from baseline.⁴⁷

Subgroup Analysis

For subgroup analysis, participants were classified as nonresponders or responders. Nonresponders were defined as those who did not achieve ≥ 50% symptom improvement by the end of study participation and had completed at least 10 ECT sessions, which we considered a full course of treatment. As the goal was to find predictors of potential ECT response, the remaining participants were included in the responder group. Responders, therefore, comprised those who met the ≥ 50% improvement criterion after at least 10 sessions, as well as those who did not complete a full course due to side effects or preference for alternative treatments, as these individuals still had the potential to benefit from ECT.

Slow Wave Activity (SWA) Measures

Sleep EEG data were recorded at a sampling rate of 250 Hz and underwent 0.1–50 Hz band-pass filtering using EEGLAB.⁴⁸ We then applied a band-stop filter between 0.1 and 0.6 Hz to remove the respiratory artifact. Spectral analyses were performed using the Chronux toolbox⁴⁹ based on 5-second nonoverlapping time windows, a time-bandwidth product of 3, and 5 tapers. Five-second windows with a maximum absolute EEG value of larger than 250 µVwere excluded for further analysis. Outlier detection based on the 3 SD rule was used on EEG power in the 0.5–4 Hz during N2/N3 stages across the nights to further remove EEG artifacts. Custom-written scripts in MATLAB (MathWorks Inc., Natick, MA) were used to quantify SWA as the average 0.5–4 Hz frontal EEG power per minute for each 30-second epoch, utilizing EEG derivative F7–F8.

Clinical Variables and Definitions

Clinical comorbidities, anthropometric measures, and periprocedural characteristics were collected from the electronic medical record. Obstructive sleep apnea (OSA) was defined as a prior clinical diagnosis documented in the medical record or a history of treatment with continuous positive airway pressure (CPAP).⁵⁰ Body mass index (BMI) was calculated as weight in kilograms divided by height in meters squared (kg/m²) using measurements obtained during routine clinical care.⁵¹ Preprocedural physical status was classified using the American Society of Anesthesiologists (ASA) Physical Status Classification System, as assigned by the treating anesthesiologist at the time of ECT.⁵²

Statistical Analysis

The D’Agostino-Pearson test was used to assess the normality of continuous variables. Unpaired t tests were applied to compare mean differences of normally distributed variables. The Wilcoxon Mann-Whitney test was used to assess unpaired differences in medians. The χ² test was used to assess categorical differences between groups.

Separate linear mixed-effects models were used to examine associations between sleep macrostructure measures and treatment number, with age and sex included as covariates. Changes in depression severity over the course of therapy were assessed using a similar mixed-effects model, with QIDS scores normalized to the maximum QIDS score observed for each patient.

The association between SWA and QIDS was first evaluated with a mixed-effects model that did not include treatment number as a covariate. SWA values were log-transformed to achieve a normal distribution across records (Supplementary Figure 1, Supplemental Digital Content 1, http://links.lww.com/JECT/A361). The mediating role of SWA in the relationship between treatment number and QIDS score was assessed using 2 linear mixed-effects models, with age and sex included as covariates (equation S1 in the Supplementary Materials, Supplemental Digital Content 2, http://links.lww.com/JECT/A362). The MaxP test,⁵³ also known as the joint significance test, was used to evaluate the statistical significance of the mediation effect (see details in the Supplementary Materials, Supplemental Digital Content 2, http://links.lww.com/JECT/A362).

Statistical analyses were conducted in GraphPad Prism (Version 9.5.1), MATLAB (Version R2021b), and R statistical software (version 4.1.2; R Core Team, 2021).

RESULTS

Study Population

Participant screening, eligibility assessment, and enrollment are detailed in the CONSORT flow diagram presented in Supplementary Figure 2, Supplemental Digital Content 3, http://links.lww.com/JECT/A363. A total of 243 individuals were initially screened for the study. Of these, 148 were deemed ineligible due to the following reasons: diagnosis of schizophrenia or schizoaffective disorder (n = 49), ineligibility due to court-ordered ECT (n = 50), pursuing non-ECT treatments (n = 47), or not meeting the age criteria (n = 2). Among the remaining 95 eligible individuals, 53 were approached for participation. Twenty-two declined to participate citing lack of interest and transportation issues, resulting in a total of 31 participants recruited into the study. During the course of the study, 9 participants withdrew (detailed in results below), leaving a final sample of 22 participants who completed the study.

Participant demographic characteristics are summarized in Table 1. The majority of subjects (n = 18, 82%) were treated for a diagnosis of major depression, while the remaining 4 subjects (18%) had a diagnosis of bipolar depression. Participants exhibited treatment resistance, having failed a median (IQR) of 4 (2) unique antidepressant trials. The average (IQR) number of ECT sessions recorded per subject was 11 (4). Depression outcomes were assessed longitudinally throughout the ECT course, with rare missing QIDS assessments totaling 10 across the study, corresponding to a median (IQR) of 0 (1) missing assessments per patient, primarily due to missed study visits. By the end of the study period, 68% of subjects received only RUL ECT, 23% had received some BL treatments, and 9% were switched to bifrontal BF treatment.

TABLE 1.

Demographic and Clinical Characteristics of Study Participants

	All Subjects (n = 22)	Responders (n = 14)	Nonresponders (n = 8)
Age (y)	48.7 ± 17.3	45.8 ± 14.8	53.8 ± 21.0
Female sex (%)	15 [68.2]	12 [85.7]	5 [62.5]
Diagnosis (%)
MDD	18 [81.8]	12 [85.7]	5 [62.5]
BD	4 [18.2]	2 [14.3]	3 [37.5]
BMI (kg/m2)	29.2 ± 8.0	27.9 ± 7.8	31.5 ± 8.3
OSA diagnosis (%)	6 [27.3]	2 [14.3]	2 [25.0]
ASA class	3.0 (1.0)	2.0 (1.0)	3.0 (0.0)
No. prior trialed antidepressants	4.0 (2.0)	4.5 (3.8)	2.5 (1.5)
No. prior trialed augmenting agents	2.5 (2.0)	3.0 (2.0)	2.0 (3.0)
No. ECT sessions	11.5 ± 3.6	11.2 ± 3.8	12.1 ± 3.2
ECT laterality (%)
Right unilateral	15 [68.2]	9 [64.3]	6 [75.0]
Bilateral	5 [22.7]	3 [21.4]	2 [25.0]
Bifrontal	2 [9.1]	2 [14.3]
Anesthetic agent(s) (%)
Etomidate	13 [59.1]	8 [57.1]	5 [62.5]
Methohexital	2 [9.1]	2 [14.3]	0
Both	7 [31.8]	4 [28.6]	3 [37.5]
Initial QIDS score	19 (8.5)	19.5 (5.5)	15.0 (5.0)
QIDS improvement	0.51 ± 0.26	0.61 ± 0.23	0.28 ± 0.18

Participants (n = 22) were divided into 2 groups: nonresponders to a full course (10 sessions) of ECT (n = 8), and responders (n = 14).

ASA indicates American Society of Anesthesiologists physical status classification; BD, bipolar disorder; BMI, body mass index; ECT, electroconvulsive therapy; MDD, major depressive disorder; OSA, obstructive sleep apnea; QIDS, Quick Inventory of Depressive Symptomatology.

ECT Clinical Response

Among the 22 participants who completed the study, ECT significantly reduced depression severity, as reflected by a decrease in median (IQR) QIDS scores from 19 (8.5) at baseline to 8.5 (7) by the end of the study period (Wilcoxon signed-rank test, P < 0.001; Fig. 2A). Of these, 18 (82%) completed at least ten ECT sessions. Ten of those 18 (56%) participants showed a clinical response during the study period, with 9 (41%) responding within 10 sessions. The remaining 8 (44%) completed ten or more treatments but did not meet the criteria for response within the study period. Four participants discontinued ECT before completing 10 sessions, with a median (IQR) of 6 (3.5) treatments. As indicated above, they were included in the responders group.

FIGURE 2.

Effects of ECT treatment on depressive symptoms. (A) Depression severity was assessed using the 16-Item Quick Inventory of Depressive Symptomatology-Self Report (QIDS).⁴⁶ This panel shows paired line plots for each participant illustrating changes in QIDS pre-ECT and post-ECT course. ECT significantly reduced depression severity, as reflected by a decrease in median QIDS scores from 19 (IQR: 8.5) at baseline to 8.5 (IQR: 7) at the end of the treatment course among the 22 participants who completed the study (Wilcoxon signed-rank test, P < 0.001). (B) The plot displays normalized QIDS as a function of treatment number, highlighting the consistent downward trend in depressive symptoms over the course of ECT. A linear mixed-effects model revealed a significant reduction in normalized QIDS scores across successive ECT sessions (F(1, 240.16) = 124.70, P < 0.001), with a 95% CI for the effect size ranging from –0.038 to –0.027. (C) Box plot showing slope of changes in QIDS across treatment number for responders (green) versus nonresponders (orange). Responders exhibited a steeper negative mean QIDS slope (−1.10, 95% CI: −1.50 to −0.71) than nonresponders (−0.24, 95% CI: −0.47 to 0.00), a difference that was statistically significant (unpaired t test, t(20) = 3.38, P = 0.003).

To further characterize the trajectory of symptoms over time, we applied a linear mixed-effects model. Across all participants, there was a significant reduction across successive ECT sessions (F(1, 240.16) = 124.70, P < 0.001). The 95% CI for the effect size ranged from −0.038 to −0.027, indicating a strong and consistent decrease in depression severity with each treatment (Fig. 2B). Longitudinal plots illustrating changes in QIDS scores for each of the 22 participants throughout the course of ECT is presented in Supplementary Figure 3, Supplemental Digital Content 4, http://links.lww.com/JECT/A364. Neither age (F(1, 20.39) = 1.27, P = 0.273; 95% CI: −0.005 to 0.002) nor sex (male vs. female; F(1, 20.76) = 1.76, P = 0.200; 95% CI: −0.043 to 0.201) showed significant associations with QIDS scores. Overall, these results indicate that while ECT treatment significantly reduces depressive symptoms, age and sex do not have a measurable impact on treatment response.

Responders demonstrated a steeper negative mean slope (−1.10, 95% CI: −1.50 to −0.71) compared with nonresponders (−0.24, 95% CI: −0.47 to 0.00). This difference was statistically significant (unpaired 2-tailed t test, t (20) = 3.38, P = 0.003), indicating that response status was associated with greater symptom reduction over the course of treatment. We next examined the subset of data excluding the 8 nonresponders resulting in 150 records from 14 individuals as responders. These 8 nonresponders did not differ significantly from responders in age, racial distribution, sex, educational attainment, psychiatric diagnosis, BMI, OSA diagnosis, ASA class, depression severity, previously trialed antidepressant or augmentation agents, number of ECT treatments during the study, ECT laterality, or anesthetic agents used (Table 1). However, as expected, the nonresponders showed significantly less improvement in QIDS compared with responders (28 ± 18 vs. 61 ± 23; t(20) = 3.74, P = 0.001, Table 1). Furthermore, responders demonstrated a steeper negative mean QIDS slope (−1.10, 95% CI: −1.50 to −0.71) compared with nonresponders (−0.24, 95% CI: −0.47 to 0.00). This difference was statistically significant (unpaired 2-tailed t test, t(20) = 3.38, P = 0.003, Fig. 2C), indicating that response status was associated with greater symptom reduction over the course of treatment.

Longitudinal Impact of ECT on Sleep Macrostructure

As an initial step, we characterized how ECT impacts sleep macrostructure at the group level, independent of treatment response status. A total of 214 overnight sleep records from 22 participants were analyzed, with a median (IQR) of 9 (5) recordings per participant. Missing sleep recordings occurred with a median (IQR) of 2 (3) per patient, commonly due to occasional device nonwear, technical recording failures, or scheduling constraints around ECT sessions. On average, participants spent 11.58 minutes (SD = 12.24) in N1 sleep, 167.14 minutes (SD = 92.52) in N2, and 10.25 minutes (SD = 17.84) in N3. REM sleep averaged 38.58 minutes (SD = 42.11), while nonscored (NS) sleep averaged 87.61 minutes (SD = 100.56). The percentage distribution of time spent in each sleep stage is shown in Table 2.

TABLE 2.

Comparison of Sleep Macrostructure and QIDS Scores Across All Treatment Sessions, As Well As At the Initial and Final Recording Sessions

	Across All Treatments	Initiala	Final	P b
Sleep macrostructure
TST (min)	227.53 ± 126.75	237.38 ± 155.84	226.93 ± 121.80	0.79
N1 (min)	11.57 ± 12.23	9.70 ± 8.15	8.11 ± 11.98	0.23
N2 (min)	167.13 ± 92.5	169.50 ± 102.62	170.27 ± 94.57	0.95
SWS (min)	10.24 ± 17.83	5.86 ± 9.95	16.81 ± 22.07	0.02
REM (min)	38.57 ± 42.11	53.31 ± 63.84	31.72 ± 34.73	0.17
N1 (%)	4.96 ± 5.03	4.18 ± 3.54	3.87 ± 7.34	0.13
N2 (%)	75.75 ± 14.74	76.48 ± 19.14	77.59 ± 15.64	0.81
N3 (%)	5.37 ± 8.60	2.80 ± 6.02	7.95 ± 11.36	0.02
REM (%)	13.91 ± 13.09	16.53 ± 18.13	10.56 ± 10.90	0.16
QIDS	13 (9)	19 (8.5)	8.5 (7)	0.001*

Values are presented as Mean ± SD or median (IQR) as appropriate.

^aFor participants with missing data at baseline, the nearest available recording was used for the initial data point.

^bPaired t tests were used for comparing continuous variables with normal distributions, while Wilcoxon signed-rank tests were used for non-normally distributed variables.

^*indicates a statistically significant difference between groups (P < 0.05).

N1 indicates non–rapid eye movement (NREM) sleep stage N1; N2, NREM sleep stage N2; N3, NREM sleep stage N3; REM, rapid eye movement sleep; SWS, slow-wave sleep; TST, total sleep time.

We next examined whether the percentage of time spent in each sleep stage was influenced by treatment number while accounting for individual variability. While treatment number was not significantly associated with the percentage of time spent in N2 (slope = 0.20; 95% CI: −0.24 to 0.64; F(1, 209.16) = 0.79; P = 0.376; Fig. 3A) or SWS (slope = 0.21; 95% CI: −0.05 to 0.46; F(1, 210.41) = 2.54; P = 0.112; Fig. 3B), the percentage of REM sleep significantly decreased with increasing treatment number (slope = −0.47; 95% CI: −0.88 to −0.06; F(1, 212.85) = 5.26; P = 0.023; Fig. 3C).

FIGURE 3.

ECT disrupts sleep macrostructure over the index course of therapy. (A) Sleep stages were manually scored into rapid eye movement (REM) and non–rapid eye movement (NREM) stages (N1–N3) according to modified American Academy of Sleep Medicine guidelines.⁴⁵ Across all participants, regardless of ECT response, the percentage of total sleep time (TST) spent in N2 sleep did not significantly change with treatment number (linear mixed-effects model; slope = 0.20, 95% CI: −0.24 to 0.64; F(1, 209.16) = 0.79, P = 0.376). (B) The percentage of TST spent in slow-wave sleep (SWS) also showed nonsignificant increases across treatment sessions (slope = 0.21, 95% CI: −0.05 to 0.46; F(1, 210.41) = 2.54, P = 0.112). However, in the responders, the percentage of time spent in SWS increased significantly by 0.35 percentage points (95% CI: 0.03–0.66; F(1, 144.11) = 4.80, P = 0.030). (C) The percentage of TST spent in REM sleep significantly decreased with treatment number (slope = −0.47, 95% CI: −0.88 to −0.06; F(1, 212.85) = 5.26, P = 0.023). For all panels in this figure, the solid black line represents the linear global effect, and the shaded blue area indicates the 95% CI.

Enhanced SWS in Responders’ Group

We then investigated how SWS changes when excluding nonresponsive subjects. A linear mixed-effects model was utilized to assess the influence of treatment number, age, and sex on the percentage of time spent in N3, while accounting for individual variability. Age (slope = −0.10 with 95% CI: −0.30 to 0.10, F(1, 11.98) = 1.05, P = 0.326) and sex (slope = −5.37 with 95% CI: −13.35 to 2.53, F(1, 10.59) = 2.09, P = 0.177) did not significantly predict percentage of time spent in N3. Though the mixed-effects model revealed a significant effect of treatment on SWS (slope = 0.35 with 95% CI: 0.03 to 0.66; F(1, 144.11) = 4.80, P = 0.030). This indicates a significant positive effect of treatment on N3 sleep, even after accounting for age and sex.

Modulation of Sleep Slow Wave Activity Throughout ECT Treatment Course

Having examined the impact of ECT on sleep macrostructure, we next investigated its effects on sleep microstructure. Specifically, we assessed the effects of ECT treatments on SWA (log-transformed) across all patients. A linear mixed-effect model revealed a statistically significant effect of treatment on the SWA (Fig. 4A; slope = 0.03 with 95% CI: 0.01–0.05, F(1, 192.04) = 9.61, P = 0.002). We then assessed whether SWA and QIDS (normalized) were correlated, without including treatment number as a covariate. We found a significant association between these measures (Fig. 4B; slope = −0.09, 95% CI: −0.15 to −0.04, F (1, 198.35) = 10.70, P = 0.001). To assess whether SWA mediates the effects of treatment on QIDS, we constructed a second linear mixed-effects model (equation S1 in the Supplementary Materials, Supplemental Digital Content 2, http://links.lww.com/JECT/A362) to test the significance of the indirect pathway depicted in Figure 4C (bottom panel). However, this mixed model did not identify a statistically significant effect of SWA on QIDS (slope = −0.039 with 95% CI: −0.082 to 0.004, F(1, 203.72) = 3.11, P = 0.079), though the result approached significance.

FIGURE 4.

Relationship between sleep SWA and depression severity across the ECT index course. (A) Log-transformed SWA during NREM sleep increased over the course of ECT (linear mixed-effects model; slope = 0.03 with 95% CI: 0.01–0.05, F(1, 192.04) = 9.61, P = 0.002). The figure shows the fixed effect trendline and 95% CI, with colored lines representing individual participants. (B) Greater SWA was associated with lower depression severity as measured by QIDS scores (slope = −0.09, 95% CI: −0.15 to −0.04, F(1, 198.35) = 10.70, P = 0.001). Lines and shaded regions reflect the fixed-effect estimate and associated CI. (C) Schematic of mediation analysis testing whether changes in SWA mediate the relationship between ECT treatment and QIDS scores. The top panel illustrates the total effect of ECT treatment on QIDS scores. The bottom panel decomposes this into direct and indirect effects, with the indirect path modeled through changes in SWA. Separate linear mixed-effects models were used to estimate each path of the mediation. The mixed model assessing the indirect pathway did not identify a statistically significant effect of SWA on QIDS (slope = −0.039, 95% CI: −0.082 to 0.004, F(1, 203.72) = 3.11, P = 0.079), although the result approached significance, suggesting a potential trend that warrants further investigation. However, a significant effect was observed in the responders (slope = −0.05, 95% CI: −0.101 to −0.003, F(1, 142.00) = 4.33, P = 0.039). This suggests that the effect of treatment on QIDS scores is partially mediated by SWA in the responders.

Sleep SWA Mediating the Therapeutic Effects of ECT in Responders

To investigate whether SWA mediates the effects of treatment on QIDS after excluding nonresponsive subjects, we constructed 2 linear mixed-effects models (equation S1 in the Supplementary Materials, Supplemental Digital Content 2, http://links.lww.com/JECT/A362), to assess the significance of the indirect pathway depicted in Figure 4C (bottom panel). These models were applied to a subset of data including only responders. This analysis revealed a significant effect of treatment on the SWA (slope = 0.03 with 95% CI: 0.014–0.06, F(1, 133.87) = 10.37, P = 0.002) and significant mediation effect of SWA on QIDS (slope = −0.05 with 95% CI: −0.101 to −0.003, F(1, 142.00) = 4.33, P = 0.039), indicating that the effect of treatment on QIDS scores is partially mediated by SWA in the responders’ group.

DISCUSSION

In this study, we leveraged longitudinal sleep recordings to characterize how a full course of ECT alters both sleep macrostructure and microstructure in patients with refractory depression. We found that slow wave sleep increased over the course of ECT and was negatively correlated with QIDS score. In addition, over a course of ECT, percent of time spent in REM decreased, and percent of time spent in N2 and N3 combined increased. Overall, these findings indicate that ECT is associated with systematic modulation of sleep architecture that parallels clinical improvement.

ECT-Induced Alterations in Sleep Structure

The analysis of sleep macrostructure revealed a significant decrease in the percentage of TST spent in REM sleep with increasing treatment number (P = 0.023). This finding is consistent with previous studies that have reported ECT-induced suppression of REM sleep.³⁹ However, because antidepressant medications are also known to suppress REM sleep, we cannot exclude the possibility that concomitant medication use contributed to this effect.⁵⁴ When nonresponders were excluded from the analysis, we found a significant positive effect of treatment on the percentage of time spent in SWS sleep (P = 0.030), suggesting that ECT may enhance deep sleep in individuals who respond to treatment. These results corroborate earlier studies that observed an elevation of SWS in individuals who respond to ECT studying sleep at baseline and after an ECT index course.³⁸

These findings extend earlier work that compared sleep architecture before and after an index ECT course³⁸ by demonstrating that sleep changes evolve dynamically across treatment sessions and covary with symptom improvement. The temporal coupling between treatment number, increased SWS, and reduced depressive severity supports the interpretation that sleep macrostructure changes are not simply downstream consequences of mood improvement, but may participate in the therapeutic process itself.

Role of Slow Wave Oscillations in ECT’s Antidepressant Effect

Given the known role of sleep alteration in the pathophysiology of depression, we investigated whether increases in SWA mediated the antidepressant effect of ECT. This study provides more convincing evidence for the important role of slow wave oscillations in depression and its treatment. Converging antidepressant interventions such as clomipramine³³ and ketamine⁵⁵ have been shown to increase SWA. Although a completed course of ECT increased time in slow wave sleep in one study,³⁸ our finding of a longitudinal increase in SWA and a relationship with treatment efficacy is stronger evidence.

Sleep architecture changes may underlie multiple antidepressant mechanisms. Many antidepressant medications enhance SWS including amitriptyline, ketamine, mirtazapine, and trazadone.^54,56 Clomipramine altered SWS distribution differentially between depression responders and nonresponders.³³ SWS is known to have roles in homeostasis and neuroplasticity,^24,57,58 and our findings further support the emerging view that enhancement of SWS is a shared feature of highly effective antidepressant treatments.

Clinical and Mechanistic Implications

These findings have important implications for optimizing the delivery of ECT. If enhancement of slow-wave oscillations contributes to antidepressant efficacy, then treatment-related factors that influence sleep physiology, including anesthetic selection, stimulation parameters, treatment spacing, and concomitant medications, may represent modifiable targets to improve clinical outcomes. Overall, our results support a model in which ECT-induced modulation of deep sleep represents a meaningful physiological correlate of antidepressant response and a promising avenue for treatment optimization.

Limitations

Our study’s primary limitation as an observational study is its inability to draw causal conclusions. However, the fact that our findings were enhanced in those who did respond to ECT suggests that these changes in sleep could be important components in understanding the response to ECT. Responder findings should be interpreted as specific to that subgroup and may vary with how response is defined, as well as with treatment duration. Confirmation in prospective studies with standardized treatment courses will help determine whether these patterns hold broadly and clarify any cause-and-effect relationships.

There are also challenges inherent in longitudinal studies with human subjects. Missing data, commonly arising from incomplete longitudinal sleep EEG recordings, were addressed using mixed-effects models to accommodate the unbalanced longitudinal data structure and varying numbers of observations per participant, rather than to impute missing values. The number of ECT treatments received varied across individuals, which could introduce variability in the data. In addition, because this was an observational study, differences in concomitant medications between groups may have influenced outcomes, and causal inferences cannot be drawn. Future research involving interventions that modify sleep structure could help elucidate the causal relationship between ECT and changes in sleep and thereby improve our understanding of underlying mechanisms.

CONCLUSION

In conclusion, this longitudinal study provides evidence that ECT induces significant changes in sleep architecture in patients with MDD. Notably, we observed a progressive increase in SWA across the ECT course, which was strongly correlated with a reduction in depressive symptom severity, as measured by the QIDS. Furthermore, in responders’ group, we found that ECT increased the percentage of time spent in SWS, and that the observed improvements in depressive symptoms were mediated by changes in SWA. These results underscore the potential role of sleep, particularly SWS and SWA, in the therapeutic mechanisms of ECT. The observed decrease in REM sleep percentage with treatment number warrants further investigation to understand its clinical implications. Future studies should aim to further elucidate the neuro- biological pathways linking ECT, sleep architecture, and mood regulation, potentially leading to the development of more targeted and effective interventions for treatmentresistant depression.

Supplementary Material

Supplemental Digital Content is available for this article. Direct URL citations are provided in the HTML and PDF versions of this article on the journal’s website, www.ectjournal.com.