The delta/alpha ratio in sleep EEG increases with the severity of depression in patients over 50 years old

Abstract

Background

The delta/alpha ratio (DAR) of sleep electroencephalography (EEG) frequency band parameters is closely related to cognitive control in healthy adults, but its role in patients with depression over the age of 50 remains unclear. Currently, a comprehensive and reliable indicator to distinguish the severity of depression in this population is still lacking.

Methods

This study included 88 participants, all of whom underwent overnight polysomnography. Among them, 30 were in the normal control group (Hamilton Depression Rating Scale (HAMD) ≤ 8), 22 in the mild-to-moderate depression group (HAMD ≤ 35), and 36 in the severe depression group (HAMD > 35). We compared demographic characteristics, polysomnographic features, relative EEG spectral power, and their ratios among the three groups.

Results

The DAR value in the severe depression group was significantly higher than that in the mild-to-moderate depression group. Spearman correlation analysis indicated that the DAR value was positively correlated with depression scores. After adjusting for age, gender, BMI, alcohol, disease duration, and age of onset, logistic regression analysis demonstrated that both the DAR value during Non-Rapid Eye Movement (NREM) sleep (p = 0.017) and Rapid Eye Movement (REM) sleep (p = 0.029) were closely associated with depression. ROC curve analysis revealed that the area under the curve (AUC) for NREM-DAR in depression patients was 0.691, with a sensitivity of 65.7% and a specificity of 81.8%.

Conclusions

DAR values strongly correlate with depression severity, suggesting their potential use as a diagnostic marker. NREM-DAR may be a risk factor for severe depression, while REM-DAR may serve as a protective factor.

Supplementary Information

The online version contains supplementary material available at 10.1186/s12888-025-07367-1.

Background

Depression is one of the most common disorders among people over the age of 50 [1]. According to data released by the World Health Organization in 2023, approximately 3.8% of the global population across all age groups suffers from depression, with a significantly higher prevalence among middle-aged and elderly individuals. Additionally, the risk of serious consequences such as cognitive decline, social isolation, and an increased risk of suicide is higher in middle-aged and elderly stages than other stages among patients with depression [2–4]. Furthermore, the overall health and quality of life of middle-aged and elderly stages patients with depression were worse than other stages due to the progression of existing chronic diseases [5]. Therefore, improvement of prognosis and quality of life among middle-aged and elderly individuals suffering from depression is the most concerned issue in society and academia [6, 7].

However, the lack of diagnostic tools limits the development of accurate diagnosis, treatment, and long-term management [8]. According to international standards, the severity of depression can be categorized into mild, moderate, and severe [9]. Assessment typically relies on subjective diagnosis by physicians and clinical scales such as the HAMD [10]. However, these methods lack sufficient objectivity and sensitivity, which may affect diagnostic accuracy. Major depression poses a significant threat to patients’ quality of life and health, particularly in individuals aged 50 and above. During this period of age, major depression often presents primarily with physical discomfort or cognitive impairments, leading to a higher risk of misdiagnosis or neglect, alongside an increased suicide risk [11, 12]. Diagnosing late-onset major depression is even more challenging, as its symptoms frequently coexist with chronic illnesses, exhibiting complexity and overlap that further complicate identification and intervention [13]. Therefore, early and accurate diagnosis and intervention are crucial to improve prognosis and quality of life among the middle-aged and elderly individuals suffer from depression [14]. Current studies on the quantification of depression severity have identified some preliminary biological markers. For instance, serotonin and dopamine levels are reduced in depression [15–17]while cortisol, C-reactive protein (CRP), and cytokines (such as IL-6 and TNF-α) levels are elevated [18–20]. However, the specificity and reliability of these markers are limited, making them insufficient for standalone diagnosis and significantly influenced by individual differences. Consequently, developing new biomarkers to enhance the early detection of depression in individuals over 50 and to assess its severity more accurately has become a focal point of current research efforts [21].

EEG is a non-invasive, real-time, and cost-effective neurophysiological technique that captures rhythmic neuronal activity at high temporal resolution. It has been widely used in the evaluation of neurological and psychiatric disorders [22, 23]. Among various EEG paradigms, sleep EEG provides a more physiologically stable and internally regulated window for observing brain activity. Resting-state recordings are often confounded by fluctuations in attention, emotional state, environmental noise, and pharmacological effects-factors especially prominent in psychiatric populations [24, 25, 28]. In contrast, during sleep-particularly NREM stages-cortically driven slow-wave activity predominates, offering a less contaminated view of endogenous brain function. For example, in frontal lobe epilepsy, typical interictal discharges may be undetectable during wakefulness but readily observed during sleep, underscoring the superior sensitivity of sleep EEG for detecting subtle abnormalities [25, 28]. Therefore, sleep EEG may offer a more suitable platform for characterizing neurophysiological dysfunctions in complex psychiatric conditions such as depression.

Building upon traditional EEG, quantitative EEG (qEEG) is an advanced extension of traditional EEG, enables more precise characterization of brain dynamics by extracting features such as frequency, amplitude, and phase through spectral analysis [26, 27]. Among qEEG techniques, frequency power ratios such as the DAR and delta/beta ratio (DBR) have attracted growing interest due to their enhanced test–retest reliability and stronger reflection of global brain states compared to absolute power values [30]. These indices have been shown to be clinically informative across multiple disorders: elevated DAR in obstructive sleep apnea (OSA) is associated with impaired sleep quality and cerebral hypoxia [31, 32]increased theta/beta ratio in attention-deficit/hyperactivity disorder (ADHD) reflects reduced attentional control [33, 34]and altered alpha/beta ratios in Alzheimer’s disease are closely linked to cognitive decline [35]. In depression, both DAR and DBR are implicated in disrupted cortical inhibition and arousal regulation, and are associated with impairments in attention, working memory, decision-making, and emotional regulation [36, 37, 49].

While most existing research on DAR and DBR has been conducted using waking-state EEG, several studies have extended their investigation into sleep and clinical EEG settings. Roeschke and Mann [29] analyzed overnight EEG in depressed patients and reported altered delta–beta phase coupling during both NREM and REM sleep, indicating disrupted cortical rhythm coordination during sleep in depression. In contrast, Livint et al. [43] showed that the Delta-Theta to Alpha-Beta Ratio (DTABR), calculated from resting-state EEG, was significantly associated with the severity of post-stroke depression, suggesting its utility as a quantitative biomarker. Furthermore, Leuchter et al. [44] demonstrated that escitalopram, but not placebo, significantly modulated the (delta + theta)/alpha ratio (DTAR) in the frontal cortex during the first week of treatment, which correlated with subsequent symptom improvement, thus supporting its role as an early neurophysiological marker of antidepressant response. Additionally, a recent study by Xu et al. [51] proposed the frontal DAR (fDAR), derived from early postoperative EEG in stroke patients, as a reliable predictor of both malignant cerebral edema and functional outcome at 3 months, further confirming the diagnostic and prognostic utility of DAR-related indices across various neurological conditions.

Taken together, while findings from resting-state EEG have provided valuable insights into the neurophysiological mechanisms of depression, their interpretation is often limited by internal and external confounding factors. In contrast, sleep EEG-recorded in a more intrinsic and regulated physiological state-may offer a more stable and sensitive window into underlying brain dysfunctions. This study focuses on individuals aged 50 years and above, aiming to characterize cortical electrical activity during sleep and to assess the clinical relevance of EEG power spectral density ratios across different sleep stages. By systematically analyzing DAR and DBR during sleep in patients with varying depression severity, the study seeks to bridge existing gaps in the literature and provide objective, quantifiable EEG biomarkers to support the assessment and stratification of depression in middle-aged and older populations.

Methods

Participants

This retrospective study initially included 159 patients with depression admitted to the Fourth People’s Hospital of Hefei between May 2019 and June 2023. According to the inclusion and exclusion criteria shown in Figs. 1 and 58 patients were finally enrolled. Additionally, 30 control subjects were recruited from the community during the same period. In this study, all patients with Major Depressive Disorder (MDD) were diagnosed through the joint evaluation of two attending psychiatrists or higher-level specialists, based on the criteria outlined in the Diagnostic and Statistical Manual of Mental Disorders, Fifth Edition (DSM-5). All patients underwent a comprehensive clinical psychiatric assessment upon admission. The diagnostic criteria included the presence of five or more symptoms from Criterion A of the DSM-5 for at least two weeks, with mandatory inclusion of either depressed mood or anhedonia, and significant impairment in social or occupational functioning. Symptoms related to substance use or physical illness were excluded as the cause of mood disturbances. The healthy control group was recruited from the community, with explicit exclusion of individuals with a history of psychiatric consultations, antidepressant medication use, or a prior diagnosis of depression. To ensure the absence of sleep disorders, preliminary screening was conducted via clinical interviews, followed by objective verification of sleep efficiency through polysomnography. The HAMD-24 was used to assess their depression status [50]. The participants were aged between 50 and 80 years, with no history of head trauma, brain surgery, or seizures, and were not using psychoactive drugs at the time of the study. In this study, participants were grouped according to the severity of their depression. Subjects with a HAMD score ≤ 8 were classified as the healthy control group (n = 30), those with a HAMD score ≤ 35 were classified as the mild-to-moderate depression group (n = 22), and those with a HAMD score > 35 were classified as the severe depression group (n = 36). Written informed consent was obtained from all participants prior to inclusion in the study. This study followed the principles of the Declaration of Helsinki and was approved by the Ethics Committee of the Fourth People’s Hospital of Hefei. All research procedures adhered to relevant guidelines and regulations.

Fig. 1.

Flowchart of participant selection. A total of 159 patients diagnosed with major depressive disorder were initially screened. After applying the inclusion and exclusion criteria - including missing HAMD-24 assessments (n = 8), lack of PSG-EEG recordings (n = 24), age not between 50–80 years (n = 28), history of epilepsy, head trauma, or brain surgery (n = 32), and low-quality EEG recordings (n = 9) − 58 patients were finally included. Among them, 36 were classified as having severe depression (SDD, HAMD > 35) and 22 as having mild-to-moderate depression (MMD, HAMD ≤ 35). Additionally, 30 healthy controls (HC, HAMD ≤ 8) were recruited from the community during the same period

Polysomnography

All participants spent the first night adapting to the sleep laboratory environment without polysomnography (PSG) recording; formal PSG monitoring was conducted on the second night. The experiment was conducted in a comfortable and soundproof sleep room. Participants arrived at the sleep room in their usual pajamas, completed the setup and adjustment of standard equipment as required, and began sleeping at 8:00 PM. All PSG data were collected using a poly graphic sleep monitoring system produced by Bio-logic Systems, a subsidiary of NATUS Group, with the model number 580G2cGss. The primary monitoring parameters included electroencephalographic (EEG) activity, electromyographic (EMG) activity, bilateral electrooculographic (EOG) activity, electrocardiographic (ECG) activity, airflow, chest and abdominal movements, blood oxygen saturation, and snoring. These signals were amplified and recorded using the poly graphic physiological recorder for amplification and recording. EEG signals were collected using the traditional monopolar montage method with a sampling rate of 512 Hz and two reference points. Pre-attached Ag/AgCl adhesive electrodes (20 in total) were used for EEG signal collection. Electrodes from the frontal, central, and parietal regions (F3-F4, C3-C4, O1-O2) were connected to the earlobes [38]with reference electrodes A1 and A2 placed on the mastoids. The ground electrode N was placed on the forehead above the eyes through a 10 KΩ resistor. The following day, recordings were initially analyzed using automated scoring software, followed by manual verification of all monitoring parameters by experienced sleep technicians. EEG recordings were visually analyzed period by period according to the 2017 version of the American Academy of Sleep Medicine (AASM) Sleep Scoring Manual. Abnormal periods with movement, respiratory muscle artifacts, electrical artifacts, baseline shifts, or electrode issues were excluded [39].

Quantitative EEG power spectral density analysis

The entire night’s EEG signals were exported in European Data Format (EDF). Visual evaluation of the data from electrodes F3, F4, O1, O2, C3, and C4 was performed using the EEGLAB toolbox in MATLAB R2019a. Electrical and movement artifacts were visually inspected, and sleep periods containing artifacts were excluded. After automatic artifact removal, EEG recordings were re-referenced to the average of the mastoid channels (A1 and A2). Signal preprocessing included applying a bandpass filter (0.5 Hz to 80 Hz) and a notch filter (48 Hz to 52 Hz). Subsequently, spectral analysis of artifact-free 30-second sleep epochs was conducted using the fast Fourier transform (FFT) via the MATLAB Welch function. The power content (in µV²) was calculated for each 30-second epoch across sleep stages N1, N2, N3, and REM. The frequency bands for spectral distribution were divided as follows: delta (0.5–4.0 Hz), theta (4.1–8.0 Hz), alpha (8.1–13.0 Hz), and beta (13.1–30 Hz). The absolute power spectral density for each frequency band across the entire brain was determined for N1, N2, N3, REM.

Statistical analysis

First, the normality and homogeneity of variances of the data were tested, presented as mean ± standard deviation. A P-value < 0.05 was considered statistically significant. All data were analyzed using the Statistical Product and Service Solutions (SPSS, version 25.0). Demographic characteristics, PSG features, relative spectral power, frequency band ratios (DAR and DBR) were compared between the two groups using two-tailed independent t-tests, chi-square tests, Wilcoxon rank-sum test, Kruskal–Wallis H test and Analysis of Variance. Pearson and Spearman correlation analysis was then performed to evaluate the correlation between HAMD scores and the relative spectral power of each sleep cycle. Based on the correlation analysis results, a binary logistic regression model and ROC analysis were used to describe the relationship between changes in DAR values and the severity of depression. The odds ratio (OR) and 95% confidence interval (CI) were calculated to evaluate the factors influencing the severity of depression. ROC curve analysis was used to identify the optimal cutoff value for distinguishing depression severity, and the area under the curve (AUC), sensitivity, specificity, and corresponding 95% confidence intervals (95% CI) were calculated, along with the optimal cutoff value.

Results

Demographic and PSG characteristics of the subjects

This study included a total of 88 participants, with their demographic and clinical characteristics shown in Table 1. The participants were divided into three groups: normal controls (n = 30), patients with mild to moderate depression (n = 22), and patients with severe depression (n = 36). There were no statistically significant differences among the groups in terms of age, BMI, gender, or the prevalence of hypertension, diabetes, or cerebral infarction. The polysomnographic characteristics of the participants are presented in Table 2. While significant differences were found in total sleep time (F = 3.785, p = 0.027) and sleep latency (H = 7.428, p = 0.024) among the three groups, no statistically significant differences in sleep-related indices were observed between the mild to moderate depression group and the severe depression group.

Table 1.

Demographic and clinical characteristics of all participants

	HC(n = 30)	DD(n = 58)	T/χ² value (HC vs. DD)	MMD(n = 22)	SDD(n = 36)	F/χ² value (HC vs. MMD vs. SDD)	T/χ² value (HC vs. MMD)	T/χ² value (HC vs. SDD)	T/χ² value (MMD vs. SDD)
Age	63.5 ± 8.51	64.97 ± 8.68	−0.756	62.82 ± 9.54	66.28 ± 7.97	1.404	0.271	−1.367	−1.488
BMI	23.48 ± 4.01	28.14 ± 39.08	−0.083	24.38 ± 3.30	23.11 ± 2.72	1.108	−0.858	0.539	1.585
Male	9,30%	23,39.7%	0.797	12,54.5%	11,28.2%	4.193	3.176	0.002	3.284
Hypertension	12,40%	27,46.6	0.344	12,54.5%	15,41.7%	1.262	1.081	0.019	0.910
Diabetes	5,16.7%	9,15.5%	0.020	2,9.1%	7,19.4%	1.114	0.625	0.085	1.117
Cerebral infarction	5,16.7%	15,25.9%	0.952	7,31.8%	8,22.2%	1.668	1.641	0.319	0.656

AbbreviationsHChealthy control,DDdepressive disorder,MMDmild-to-moderate depressive,SDDsevere depressive disorder,BMIBody Mass Index

Table 2.

Polysomnographic characteristics of all participants

	HC(n = 30)	DD(n = 58)	T/Z value (HC vs. DD)	MMD(n = 22)	SDD(n = 36)	F/H value (HC vs. MMD vs. SDD)	T/Z value (HC vs. MMD)	T/Z value (HC vs. SDD)	T/Z value (MMD vs. SDD)
Time in Bed	600.97 ± 76.49	585.95 ± 84.29	0.815	585.61 ± 76.09	586.15 ± 89.78	0.328	0.707	0.714	−0.023
Total sleep time	495.95 ± 69.97	433.48 ± 115.52	3.15**	442.64 ± 86.33	427.89 ± 131.07	3.785*	2.458*	2.69**	0.468
Sleep efficiency	83.65(75.73–87.6)	77.45(67.73–85.25)	−2.61**	74.4(67.73–81.43)	80.35(62.32–86.15)	7.428*	−2.806**	−1.867	−0.561
Sleep latency	16.1(5.85–31.9)	18.4(11.05–39.05)	−1.576	26.65(11.7–44.93)	15.7(9.77–34.9)	3.712	−1.899	−0.985	−1.122
REM Latency	150.75(57–315.75)	234(115–348.5)	−1.564	221.25(118.13–360.38)	235.5(115–348.5)	2.466	−1.456	−1.27	−0.018
arousal index	2.5(2.08–4)	3.4(2.1–4.65)	−1.255	3.3(2–4.65)	3.4(2.15–5.03)	2.081	−0.64	−1.392	−0.762
AHI	1.45(1.1–2.5)	1.5(0.98–2.1)	−0.586	1.2(0.68–1.6)	1.65(1.13–2.3)	5.799	−1.753	−0.361	−2.423*
REM	47.5(16.88–74.63)	29.25(15.88–51.5)	−2.026*	31.5(17.25–54)	25.5(15–44.5)	4.302	−1.417	−1.968*	−0.516
N1	71.25(48.13–156)	88.75(48.75–164.88)	−0.37	81(48.75–128.75)	126(45.13–179.63)	1.095	−0.213	−0.689	−1.042
N2	286.75(212.88–350.75)	246.75(171.63–319.38)	−1.378	282.75(187.75–319.38)	238.25(169.13–319.63)	2.37	−0.704	−1.526	−0.697
N3	40.5(17.38–83.63)	31.5(12.38–68.75)	−0.889	35.75(11.25–92.88)	28(17.75–67.13)	0.822	−0.741	−0.786	−0.24
REM%	9.57 ± 6.16	7.42 ± 5.02	1.625	7.79 ± 5.25	7.19 ± 4.94	1.567	1.075	1.714	0.424
N1%	13.55(9.58–29.85)	22.5(12.1–38.63)	−1.739	17(11.33–32.65)	24.5(12.35–44.3)	4.294	−0.852	−1.951	−1.242
N2%	60.2(42.7–71.78)	57.8(46.58–66.4)	−0.198	61.7(46.8–71.73)	54.75(46.33–63.83)	1.629	−0.454	−0.605	−1.378
N3%	8.2(3.9–14.9)	7.55(3.5–15.75)	−0.59	8.1(2.48–18.25)	6.7(3.8–15)	0.436	−0.611	−0.438	−0.329
NREM%	90.5(85.3–96.8)	91.5(84.03–96.02)	0	90.5(80.13–94.6)	91.85(85.98–96.43)	0.746	−0.593	−0.412	−0.794

AbbreviationsREMRapid Eye Movement,NREMNon-Rapid Eye Movement,N1non-REM (NREM) stage 1,N2 NREM stage 2,N3NREM stage 3

Relative spectral power and band ratio of all participants

The differences in relative spectral power and frequency band ratios across various sleep stages for all participants are shown in Fig. 2. Significant differences in delta power were observed among the three groups during N1(H = 9.845, p = 0.007), N2(H = 15.125, p = 0.001), N3(H = 14.28, p = 0.001), NREM (H = 24.567, p < 0.001), and total sleep periods (H = 25.503, p < 0.001). Additionally, the DAR values differed significantly among the three groups during N1(H = 8.17, p = 0.017), N3(H = 11.276, p = 0.004), NREM (H = 13.807, p = 0.001), and total sleep periods (H = 13.241, p = 0.001). Similarly, significant differences in DBR values were found among the three groups during N1(H = 10.13, p = 0.006), N2(H = 13.805, p = 0.001), N3(H = 24.539, p < 0.001), NREM (H = 22.272, p < 0.001), and total sleep periods (H = 19.909, p < 0.001).

Fig. 2.

Differences in EEG Power and Power Ratio Across Sleep Stages. The healthy control group (HC) generally exhibited higher delta power, DAR, and DBR across all sleep stages, especially during the NREM stage and total sleep stage, with significant differences compared to the severe depressive disorder group (SDD), mild-to-moderate depressive disorder group (MMD), and mixed depressive disorder group (DD) (p < 0.05). Additionally, DAR and DBR were overall lower in the MMD group compared to the SDD group. Theta power in the total sleep stage was higher in HC than in DD and SDD. Alpha power in the REM stage was significantly lower in the MMD group than in the SDD group. No significant differences were observed in Beta power across groups

The delta power in the severe depression group was significantly higher than that in the mild-to-moderate depression group during the total sleep period (Z=−2.292, p = 0.022). Although no significant differences were observed between the two groups during other periods, the mean value of delta power in the severe depression group was still higher than that in the mild-to-moderate group. Similarly, the DAR in the severe depression group was significantly higher than that in the mild-to-moderate group during N1(Z=−2.305, p = 0.033), N2(Z=−2.396, p = 0.017) and NREM (Z=−2.179, p = 0.029). While no significant differences in DAR were found between the two groups during other periods, the mean value of the severe depression group remained higher than that of the mild-to-moderate group, except during the REM period.

Correlation between HAMD scores, relative spectral power, and band ratios for all participants

The correlation between HAMD scores and relative spectral power among all participants is presented in Figure3. During N1 (r = 0.324, p = 0.015), N2 (r = 0.337, p = 0.01), NREM (r = 0.361, p = 0.005), and total sleep duration (r = 0.373, p = 0.004), delta power in all participants showed a positive correlation with HAMD depression scores. Furthermore, during N1 (r = 0.299 p = 0.027), N3 (r = 0.298, p = 0.024) and NREM (r = 0.274, p = 0.039), DAR values were positively correlated with the HAMD depression scores. Similarly, DBR values exhibited a positive correlation with HAMD depression scores across N2(r = 0.319, p = 0.015).

Fig. 3.

Correlation Heatmap Between HAMD Depression Scores and EEG Power/Power Ratio. A There is a significant positive correlation between delta power and HAMD depression scores in all participants during N1 (r = 0.324, p = 0.015), N2 (r = 0.337, p = 0.01), NREM (r = 0.361, p = 0.005), and total sleep duration (r = 0.373, p = 0.004). B. Theta power is positively correlated with HAMD depression scores during REM sleep (r = 0.312, p = 0.027). C. Alpha power shows a significant positive correlation with HAMD depression scores during REM sleep (r = 0.378, p = 0.006). D. There is no significant correlation between beta power and HAMD depression scores in any sleep stages. E. DAR values are positively correlated with HAMD depression scores during N1 (r = 0.299, p = 0.027), N3 (r = 0.298, p = 0.024), and NREM sleep (r = 0.274, p = 0.039). F. DBR values show a significant positive correlation with HAMD depression scores during N2 sleep (r = 0.319, p = 0.015) 

Logistic regression analysis of depression level for all participants

Figure 4 and Table 3 present the results of the binary logistic regression analysis. Across all models (both unadjusted and adjusted), NREM-DAR was positively correlated with the occurrence of severe depression. From Model 1 to Model 3, for each unit increase in NREM-DAR, the probability of severe depression increased 10%, 13%, and 41%. Conversely, REM-DAR was negatively correlated with the occurrence of severe depression. From Model 1 to Model 3, for each unit increase in REM-DAR, the probability of severe depression decreased by 7%, 8%, and 29%. In Model 3, both NREM-DAR and REM-DAR showed significant effects. The OR (95% CI) for NREM-DAR was 1.411(1.064,1.871), and for REM-DAR, it was 0.711(0.524,0.965). 

Fig. 4.

Regression Forest Plot. In Model 1, the odds ratio (OR) for NREM-DAR is 1.1 (95% CI 1.011–1.198), indicating that for every one-unit increase in NREM-DAR, the odds of the event occurring increase by approximately 10%. Meanwhile, the OR for REM-DAR is 0.928 (95% CI 0.865–0.995), meaning that for every one-unit increase in REM-DAR, the odds of the event occurring decrease by about 7.2%, and its 95% confidence interval does not cross 1, suggesting statistical significance. In Model 2, the OR for NREM-DAR is 1.139 (95% CI 1.023–1.268), further indicating a positive correlation between NREM-DAR and the occurrence of the event, with the odds of the event increasing by about 13.9% for each unit increase in NREM-DAR. The OR for REM-DAR is 0.918 (95% CI 0.847–0.996), showing that an increase in REM-DAR reduces the odds of the event by 8.2%, and its confidence interval does not include 1, suggesting statistical significance. In Model 3, the OR for NREM-DAR is 1.411 (95% CI 1.064–1.871), further reinforcing the positive correlation between NREM-DAR and the occurrence of the event, with the odds of the event increasing by approximately 41.1% for each unit increase in NREM-DAR. However, the OR for REM-DAR is 0.711 (95% CI 0.524–0.965), indicating that for every one-unit increase in REM-DAR, the odds of the event occurring decrease by approximately 28.9%, and its confidence interval does not cross 1, suggesting statistical significance

Table 3.

Logistic regression analysis of depression level for all participants

	NREM-DAR			REM-DAR
	OR	P	CI(95%)	OR	P	CI(95%)
Model 1	1.1	0.027	1.011–1.198	0.928	0.037	0.865–0.995
Model 2	1.139	0.017	1.023–1.268	0.918	0.039	0.847–0.996
Model 3	1.411	0.017	1.064–1.871	0.711	0.029	0.524–0.965

Model 1: Prototype

Model 2: Adjustment for age, gender, alcohol and BMI

Model 3: Model 2 + adjusted for disease duration and age at first presentation

ROC curve analysis of depression level for all participants

The ROC curve was used to assess the predictive performance of DAR and DBR values across different sleep stages, N1, N2, N3, NREM, REM, and total sleep periods. The DAR values demonstrated significant predictive capability for N1, N3, NREM, and total sleep periods, while the DBR values showed notable performance specifically for N2 sleep periods. As shown in Fig. 5, the optimal cutoff value for N1-DAR is 13.06, with an AUC of 0.690(95% CI 0.55–0.83), a p-value of 0.02, sensitivity of 42.9%, and specificity of 95%.The optimal cutoff value for N3-DAR is 15.04, with an AUC of 0.689(95% CI 0.54–0.83), a p-value of 0.018, sensitivity of 86.1%, and specificity of 52.4%.The optimal cutoff value for NREM-DAR is 14.65, with an AUC of 0.691(95% CI 0.55–0.84), a p-value of 0.016, sensitivity of 65.7%, and specificity of 81.8%.The optimal cutoff value for total sleep period DAR is 13.04, with an AUC of 0.671(95% CI 0.53–0.82), a p-value of 0.03, sensitivity of 68.6%, and specificity of 77.3%. The optimal cutoff value for N2-DBR is 38.76, with an AUC of 0.675(95% CI 0.53–0.82), a p-value of 0.026, sensitivity of 75%, and specificity of 59.1%.

Fig. 5.

ROC Analysis for DAR Value.A.N1-DAR: Optimal cutoff is 13.06, with AUC = 0.690 (95% CI 0.55–0.83), p = 0.02, sensitivity = 42.9%, specificity = 95%.B.N3-DAR: Optimal cutoff is 15.04, with AUC = 0.689 (95% CI 0.54–0.83), p = 0.018, sensitivity = 86.1%, specificity = 52.4%.C.NREM-DAR: Optimal cutoff is 14.65, with AUC = 0.691 (95% CI 0.55–0.84), p = 0.016, sensitivity = 65.7%, specificity = 81.8%.D.N2-DBR: Optimal cutoff is 38.76, with AUC = 0.675 (95% CI 0.53–0.82), p = 0.026, sensitivity = 75%, specificity = 59.1%.E.Total Sleep Period DAR: Optimal cutoff is 13.04, with AUC = 0.671 (95% CI 0.53–0.82), p = 0.03, sensitivity = 68.6%, specificity = 77.3%

Discussion

To our best knowledge, this is the first study to use the frequency band ratio DAR value to assess the changes in sleep EEG in patients with depression aged 50 and above. Our study revealed significant differences in DAR values between the normal control group and patients with depression during the N1, N3, NREM and total sleep periods. Additionally, a significant correlation was observed between HAMD scores and DAR values during NREM, REM, and total sleep periods, suggesting the potential of DAR values as biomarkers related to depression.

Further ROC analysis demonstrated the diagnostic value of DAR values in distinguishing the severity of depression. Specifically, the NREM-DAR value had an area under the curve (AUC) of 0.691, with a sensitivity of 65.7% and a specificity of 81.8%. These findings support the potential application of DAR values as a tool for assessing the severity of depression. Moreover, through binary logistic regression models adjusted for age, gender, BMI, disease duration, and age of onset, multiple analyses consistently showed that NREM-DAR values were risk factors for severe depression, whereas REM-DAR values served as protective factors. These findings further reinforce the hypothesis that DAR values may represent novel biomarkers for differentiating the severity of depression.

Sleep EEG can be divided into NREM and REM stages. Among these, slow-wave activity (SWA), particularly delta waves, is a hallmark of deep NREM sleep and plays a critical role in synaptic plasticity, cortical homeostasis, and neural recovery [41, 42]. Delta waves (0.5–4 Hz) represent the lowest frequency band in the EEG spectrum and are essential for restorative sleep. Disruptions in delta activity may reflect underlying impairments in the brain’s ability to recover and reorganize [40, 45]. Previous studies have consistently reported that depressed patients tend to exhibit reduced delta power during NREM sleep, indicative of impaired slow-wave generation and diminished sleep quality [56–59, 61]. For example, Goldschmied et al. found decreased delta activity in MDD patients, which was associated with poor emotional regulation and impaired neurophysiological function [57].

In alignment with this literature, our results confirmed that overall delta power was significantly lower in depressed individuals compared to healthy controls. However, stratified analysis further revealed that the MMD group exhibited the lowest delta activity, while the SDD group demonstrated a slight rebound, albeit still markedly lower than healthy participants. This intragroup variability suggests that delta power alone may not be sufficient to characterize EEG alterations across depression severity, and compensatory mechanisms may be involved. This phenomenon may be attributed to the heterogeneity of sleep disturbances in older adults with depression. Clinically, SDD patients are more likely to experience early morning awakenings, preserving relatively intact slow-wave sleep (SWS) in the first half of the night. In contrast, MMD patients often show sleep-onset insomnia or frequent nocturnal awakenings, leading to fragmented SWS and further reductions in delta activity [52]. From a neurobiological perspective, delta waves are generated by synchronized thalamocortical oscillations and are influenced by GABAergic inhibition, synaptic integrity, and cortical plasticity [53, 54]. Depression-related impairments in these systems may suppress delta generation or result in dysfunctional enhancement, where increased delta power lacks restorative efficacy.

Given that delta activity also influences broader cortical dynamics, this study further incorporated DAR and DBR to evaluate the dynamic balance between slow- and fast-wave activity. These spectral power ratios offer advantages over single-band power measures by minimizing interindividual variability and better reflecting cortical arousal and sleep–wake regulatory integrity [46]. Previous studies have shown that depressed patients often exhibit elevated alpha and beta power during sleep [48]which reflects cortical hyperarousal and may disrupt sleep continuity. While such high-frequency activity could lower DAR and DBR by increasing the denominator of these ratios, our findings indicated that both DAR and DBR were significantly higher in the SDD group than in the MMD group. This result suggests that frequency ratios may reflect complex interactions of multiple mechanisms beyond isolated band changes. Jaimchariyatam et al. reported that a significant subset of MDD patients exhibit alpha–delta sleep intrusion, where alpha activity overlaps with delta waves during NREM, disrupting the functional role of slow-wave sleep and reducing spectral stability [60]. This pathological interplay may lead to elevated DAR or DBR values that do not reflect restorative processes, but rather signify cortical dysregulation and impaired sleep microstructure.

NREM sleep plays a crucial restorative role and is key to memory consolidation and emotional regulation, with the N3 phase being the core of deep sleep in NREM, essential for restorative sleep and brain function regulation [45, 47]. In contrast, REM sleep is often associated with dreaming, with brain activity approaching that of wakefulness, primarily involved in emotional regulation and memory processing. It helps alleviate negative emotions and enhance emotional adaptability. In patients with depression, an increase in NREM-DAR may indicate impaired ability to achieve restorative deep sleep, reflected by abnormal cortical synchronization. This inefficiency in sleep-related neural regulation may hinder the clearance of emotional and cognitive load accumulated during wakefulness, thereby weakening emotional regulation and increasing the risk of severe depression. Conversely, a relatively lower DAR during REM sleep may reflect more preserved emotional balance and reparative processing, potentially serving a protective role in alleviating depressive symptoms. These results can be explained within the framework of existing neurophysiological models of depression, including the hyperarousal theory and the dysfunction of the default mode network (DMN). The hyperarousal theory posits that patients with depression exhibit sustained cortical hyperarousal during sleep, especially during NREM stages [55]. The increase in NREM-DAR observed in this study may reflect such pathological hyperarousal, suggesting a reduced ability to achieve deep restorative sleep and maladaptive cortical synchronization. On the other hand, the DMN dysfunction model suggests that abnormal DMN activity in patients with depression leads to imbalances in emotional and cognitive regulation [50]which may also partially explain the relationship between REM-DAR changes and emotional reparative processing during REM sleep. This study found that NREM-DAR is a risk factor for severe depression in individuals over the age of 50 and has potential value in predicting the severity of depression. In contrast, REM-DAR acts as a protective factor, helping to reduce the risk of severe depression in this population. Therefore, DAR values could serve as a stable biomarker in the EEG of depression patients over 50 years old. With its high sensitivity and specificity, DAR can accurately differentiate between mild to moderate depression and severe depression, providing a basis for the development of personalized treatment plans.

This study still has several limitations. First, due to its retrospective design, the available data were limited, and cognitive function assessments were not included, preventing exploration of the causal relationship between DAR values and depression-related cognitive decline. Second, all participants in this study were elderly inpatients with depression, with relatively consistent medication use; therefore, sensitivity analyses regarding medication factors were not conducted. Future studies should expand sample sources and increase sample sizes, systematically document different types and dosages of medications, and perform stratified analyses on their effects on EEG power ratios. Additionally, sensitivity analyses and methods such as propensity score matching should be applied to minimize potential selection bias and enhance the reliability and generalizability of the findings. Further research is also needed to validate the utility of DAR in larger samples and multicenter cohorts, and to explore its potential value in predicting treatment response in depression, thereby providing stronger evidence for its clinical application. 

Conclusions

DAR values strongly correlate with depression severity, suggesting their potential use as a diagnostic marker, with NREM-DAR potentially serving as a risk factor for severe depression and REM-DAR possibly acting as a protective factor. Given the current lack of reliable and comprehensive biomarkers for accurately assessing depression severity in adults aged 50 and above, our findings provide valuable insights for the fields of clinical psychiatry and sleep medicine. Utilizing DAR as a potential biomarker could greatly enhance diagnostic and therapeutic approaches for this population.

Abbreviations

DAR

Delta/alpha ratio

HC

Healthy control

DD

Depressive disorder

MMD

Mild-to-moderate depressive

SDD

Severe depressive disorder

BMI

Body Mass Index

REM

Rapid Eye Movement

NREM

Non-Rapid Eye Movement

N1

Non-REM (NREM) stage 1

N2

NREM stage 2

N3

NREM stage 3

Authors’ contributions

Xiaoming Kong take responsibility for all aspect of this research. Letian Yang designed the study and wrote the article. Siwen Lv, Xinyu Gao, Jiaojiao Li, Yangliu Pei, Nannan Zhu and Yu Guo worked for literature search. Letian Yang, Hao Geng, Yan Sun and Pengyu Xie designed figures. Letian Yang, Hao Geng, Siwen Lv, Jiaojiao Li and Pengyu Xie collected data. Letian Yang and Pengyu Xie analyzed data. Xiaoming Kong conducted data interpretation.

Funding

This work was supported by National Clinical Key Specialty Construction Project of China, Anhui Province Clinical Key Specialty Construction Project, Anhui Provincial Health Research Program (AHWJ2023A20208), Hefei Seventh-cycle Key Medical Specialty, and Applied Medical Research Program of Hefei Municipal Health and Wellness Commission (Hwk2022zd016).

Data availability

The underlying dataset used in this study contains sensitive personal identifying information and is therefore not publicly available. The data may be made available from the corresponding author, Xiao-ming Kong, upon reasonable request and subject to the approval of the relevant ethical committee and institutional policies.

Declarations

Ethics approval and consent to participate

All experiments were conducted in accordance with the Declaration of Helsinki and institutional guidelines, and were approved by the Clinical Research Ethics Review Committee of the Fourth People’s Hospital of Hefei (approval date: October 27, 2023; ethical approval reference number: HFSY-IRB-YJ-KYXMKXM(2023–070 − 001)). Written informed consent was obtained from all individual participants included in the study. The authors affirm that patient confidentiality has been fully maintained.

Consent for publication

Not applicable.

Competing interests

The authors declare no competing interests.