Serum inflammatory markers and total antioxidant capacity in relation to sleep outcome in the elderly

Abstract

Background

Sleep disturbance in the elderly has been linked to systemic inflammation and oxidative imbalance. However, most studies are cross-sectional and rarely evaluate inflammatory markers and a functional antioxidant summary alongside both subjective and objective sleep measures. We examine baseline and 12‑month within‑person co-variation between inflammatory markers, total antioxidant capacity (TAC), and sleep outcomes in the elderly.

Methods

We conducted a 12‑month observational cohort to examine associations between serum inflammatory markers (C-reactive protein [CRP], Interleukin-6 [IL-6], Tumor necrosis factor-α [TNF-α]) and total antioxidant capacity (TAC) with sleep outcomes quality in elderly. Paired tests and multivariable linear regression were used to evaluate within-person change and cross-sectional associations, adjusting for prespecified covariates. Participants with Pittsburgh Sleep Quality Index (PSQI) ≥ 8 were offered Cognitive behavioral therapy for insomnia (CBT‑I) as usual care; treatment effects were not evaluated.

Results

Over the study period, inflammatory markers (CRP, IL-6, TNF-α) showed a modest decrease, while TAC slightly increased. Concurrently, actigraphy revealed a small rise in average sleep duration, and the PSQI improved. Regression analyses demonstrated that reductions in inflammatory markers and increases in antioxidant capacity were independently associated with improved sleep outcomes, while a composite index combining these biomarker changes correlated significantly with greater PSQI improvements after adjust for age, sex, and BMI.

Conclusions

The study confirms that higher levels of inflammatory markers and lower levels of antioxidant are associated with decreased sleep quality. Moreover, improvements in sleep were associated with decreases in CRP, IL‑6, and TNF‑α and increases in TAC over 12 months. These findings support inflammation and antioxidant capacity as complementary correlates of sleep health and motivate future causal studies to test directionality and mechanisms.

Supplementary Information

The online version contains supplementary material available at 10.1186/s12877-026-07133-2.

Introduction

Sleep disorders are highly prevalent among the elderly, significantly impacting their quality of life and overall health. Approximately 58.43% to 75% of elderly experience sleep disturbances, including insomnia, excessive sleep, and poor sleep quality [1, 2]. These disorders are linked to various adverse health outcomes, such as decreased immunity, cognitive decline, and increased susceptibility to chronic diseases like diabetes and cardiovascular conditions [3, 4]. Factors such as sex, health status, and education significantly influence the prevalence of sleep disorders in this population [2].

Systemic inflammation significantly impacts sleep quality. Poor sleep characteristics, including insomnia and sleep inconsistency, correlate with elevated C-reactive protein (CRP) levels [5–7]. Elevated CRP has been associated with poor sleep quality independent of nocturnal symptoms in patients with inflammatory bowel disease [8]. Furthermore, genetic studies suggest that chronic inflammation can influence habitual sleep duration [9]. Sleep fragmentation and short sleep can activate sympathetic and hypothalamic pituitary adrenal pathways and NF-κB signaling, increasing interleukin (IL)-6 and tumor necrosis factor (TNF)-α and hepatic CRP, which can degrade sleep continuity and architecture [10]. Higher daytime sleepiness and lower sleep efficiency are linked to increased IL-6 and CRP levels [5]. Notably, variability in sleep patterns, including night-to-night fluctuations, has been shown to correlate with heightened inflammatory responses [11]. Overall, these findings underscore the critical role of systemic inflammation in shaping sleep quality and, consequently, overall health.

Antioxidants play a crucial role in mitigating oxidative stress particularly in aging individuals. Oxidative stress contributes to cellular damage and age-related diseases [12, 13]. Recent studies indicate a significant association between dietary antioxidants and improved sleep quality [14, 15]. Additionally, antioxidants like vitamin C and E have shown protective effects against sleep disturbances [15, 16]. The mechanisms underlying this relationship may involve the modulation of oxidative stress and inflammation [14]. Thus, a diet rich in antioxidants may not only combat oxidative stress but also enhance sleep quality.

Inflammation and oxidative stress are tightly coupled biological processes, and TAC provides a functional summary of systemic antioxidant defenses rather than a single antioxidant concentration [13, 17, 18]. Sleep disturbance can amplify inflammatory signaling and oxidative stress, while inflammatory cytokines and oxidative imbalance may, in turn, impair sleep continuity and quality, suggesting a bidirectional loop [10, 19–21]. Therefore, we considered inflammatory markers (CRP, IL-6, TNF-α) and TAC as complementary indicators of an integrated inflammatory–oxidative milieu [10, 19–21], aiming to evaluate its relationship with sleep health in older adults, using both PSQI and actigraphy, and to test whether within‑person changes in these biomarkers track with within‑person changes in sleep outcomes over 12 months, by examining (1) baseline associations between biomarkers and sleep outcomes and (2) within‑person co‑variation over 12 months between changes in biomarkers and changes in sleep outcomes (PSQI and actigraphy) in the elderly.

Methods and materials

This was a 12‑month, single-center observational cohort conducted at Second Medical Center, PLA General Hospital (Beijing, China), with repeated assessments at baseline (month 0), month 6, and month 12. As part of routine clinical care (outside the study protocol), participants with PSQI ≥ 8 were offered CBT‑I. Allocation was non‑random, and the study was not designed to estimate a treatment effect. Primary analyses were cohort‑level and observational. The study was conducted at Second Medical Center, PLA General Hospital. All eligible participants gave written informed consent after being informed of the study objectives, procedures, potential risks, and benefits. The protocol was approved by the Institutional Review Board of Second Medical Center, PLA General Hospital (KY-2022-13). All procedures complied with the Declaration of Helsinki and followed the STROBE guideline for observational studies (Fig. 1).

Fig. 1.

Participant flow diagram. Flow of participants through the 12‑month observational cohort, including eligibility screening, exclusions, baseline enrollment (Month 0), and follow‑up assessments at Months 6 and 12

Inclusion criteria

1), Age ≥ 65 years; 2), able to provide informed consent; 3), completed a baseline fasting blood draw and 7-day actigraphy; 4), willing to undergo repeated blood tests and sleep assessments at follow-up.

Exclusion criteria

1), known, untreated severe sleep disorders requiring urgent therapy; 2), acute infection or active inflammatory disease at baseline; 3), systemic immunosuppressive therapy; 4), end-stage renal/hepatic disease; 5), shift work; 6), conditions precluding safe participation or valid actigraphy.

Data collection and assessments

Demographic and clinical data

At baseline, participants provided information including age, sex, ethnicity, living status, medical history (hypertension, diabetes, and any major comorbidities), Lifestyle factors (smoking status, alcohol use, physical activity), and anthropometric measures (height, weight, and calculation of body mass index (BMI)). Alcohol use was assessed at baseline as participants’ usual drinking habits over the prior 12 months using a structured questionnaire on typical frequency and quantity. Participants were then categorized as none, moderate, or heavy drinkers according to prespecified thresholds. Medical history was obtained by structured questionnaire and/or medical record review and included hypertension (blood pressure ≥ 140/90 mmHg), diabetes, dyslipidemia, coronary heart disease, and stroke history where available. Current long‑term medications were recorded where available, including aspirin/antiplatelet therapy, statins/lipid‑lowering agents, and hypoglycemic medications.

Biomarker measurements

Participants had blood drawn at baseline (Month 0), Month 6, and Month 12. Venipuncture was performed in the morning (07:00–09:00) after an overnight fast (8–10 h). Inflammatory markers CRP was measured by automated particle-enhanced immunoturbidimetric assay (CRP4, Roche Diagnostics; Limit of detection: 0.3 mg/L, Intra-assay CV: 1.2–4.8%; Inter-assay CV: 2.3–5.5%). IL-6 (HS600C, R&D Systems. Analytical sensitivity: 0.031 pg/mL; Intra-assay CV: 3.6–4.7%; Inter-assay CV: 3.9–10.8%) and TNF-α (HSTA00E, R&D Systems. Analytical sensitivity: 0.022 pg/mL; Intra-assay CV: 1.9–2.2%; Inter-assay CV: 6.2–6.7%) were quantified via enzyme-linked immunosorbent assay (ELISA). TAC refers to the cumulative reducing capacity of serum/plasma against a standardized oxidant and is expressed as Trolox equivalent. It is not a concentration of a single antioxidant. TAC was measured in fasting serum using a colorimetric ABTS/TEAC method (Antioxidant Assay Kit, #709001, Cayman Chemical, USA), run in duplicate with intra‑/inter‑assay CVs of 5.5%/8.0%. The limit of detection was 0.02 mM. Results are reported as mM Trolox equivalents.

Sleep assessments

Objective sleep was measured by wrist actigraphy. Participants wore an Amazfit Arc device (Huami Technology, China) on the non‑dominant wrist for 7 consecutive days at baseline, 6 months, and 12 months. Raw acceleration was sampled at 30 Hz and aggregated to 30‑s epochs. Sleep–wake was scored with the validated Cole–Kripke adult algorithm. Rest intervals were defined from participant sleep logs and event markers with visual review by trained staff. Valid‑night criteria required a scorable rest interval with continuous wear. Nights with extended non‑wear, operationalized as ≥ 60 min of consecutive zero activity with a 2‑min spike tolerance, were excluded. At each time point, participants were required to have ≥ 5 valid nights, and outcomes were averaged across valid nights. Total sleep time (TST) and time in bed (TIB) were derived nightly. Sleep duration refers to TST (hours/night), and sleep efficiency was calculated as (TST ÷ TIB) × 100%. Although Amazfit Arc is a consumer wearable, its raw accelerometry can be used for actigraphy‑style sleep/wake scoring. Prior validation studies of the Huami Arc have shown sleep/wake estimates comparable to a research‑grade actigraph in free‑living conditions and, in PSG validation, approximately 90% epoch‑by‑epoch accuracy with high sleep sensitivity and more limited wake specificity—performance comparable to a clinical‑grade Actiwatch, consistent with known actigraphy limitations in detecting quiet wakefulness [22, 23]. Subjective sleep quality was assessed with the Pittsburgh Sleep Quality Index (PSQI) at baseline, month 6, and month 12 (global score 0–21, higher scores = worse sleep). A PSQI cutoff ≥ 8 was used solely to triage participants to usual‑care CBT‑I and was not used to define study eligibility or primary analyses.

Usual‑care CBT‑I availability

Participants with PSQI ≥ 8 were offered cognitive behavioral therapy for insomnia (CBT‑I) as part of routine clinical care (outside the study protocol). Allocation to CBT‑I was non‑random, and the study was not designed to estimate treatment effects. no causal inference or group comparisons by CBT‑I exposure were performed.

Sample size considerations

No formal a priori power calculation was performed. The target sample (n ≈ 300) reflected feasibility for a 12‑month cohort with repeated measures, allowing estimation of prespecified associations with adjustment for a limited covariate set. We emphasize effect sizes with 95% CIs and consistent patterns across subjective and objective sleep measures.

Statistical analysis

Continuous variables were presented as mean ± SD (or median [IQR]), while categorial variables were presented as frequency and percentage. Paired tests were used to assess differences in biomarkers and sleep from baseline to month 6 and month 12. For within‑person comparisons, we used paired t‑tests for approximately normally distributed outcomes and Wilcoxon signed‑rank tests for skewed outcomes. Analyses used available paired observations for each comparison (no imputation). For regression analyses, we used complete‑case data for the outcome, predictor, and prespecified covariates. The analytic sample size for each model is reported in the corresponding table. The primary sleep outcome for regression models was the PSQI global score. Actigraphy outcomes were sleep duration and sleep efficiency =((total sleep time ÷ time in bed) × 100%). We fit linear regression models with PSQI as the dependent variable and each biomarker (CRP, IL‑6, TNF‑α, TAC) as the predictor (univariate), then a multivariable model including prespecified covariates (age, sex, BMI, hypertension, elevated fasting glucose, smoking status). We examined whether within‑person changes in biomarkers were associated with changes in sleep by computing Δ values as 12‑month minus baseline. We then regressed ΔPSQI on each Δ biomarker (univariate) and on all Δ biomarkers together (multivariable) with the same covariates as above. For interpretability, β coefficients are reported on the original units. To summarize overall inflammatory/antioxidant change, we created a composite by z‑scoring ΔCRP, ΔIL‑6, and ΔTNF‑α, reverse‑scoring ΔTAC, and averaging the z‑scores. Linearity and homoscedasticity were assessed via residual vs. fitted plots, normality of residuals via Q‑Q plots; leverage/influence via Cook’s distance, and multicollinearity via VIF (threshold < 5). Because CRP, IL‑6, and TNF‑α can be right‑skewed, we repeated key models with log‑transformed biomarkers as a sensitivity check. In sensitivity analyses, we additionally adjusted multivariable models for baseline statin and aspirin use. Analyses conducted in R (version 4.4) with p < 0.05 considered statistically significant.

Exploratory moderation and mediation analyses

Because inflammation and oxidative balance are biologically coupled, we evaluated whether TAC modified (moderated) the association between inflammatory burden and sleep outcomes. We created an inflammation burden index as the mean of z‑scored ln‑transformed CRP, IL‑6, and TNF‑α and tested an interaction term (inflammation index × TAC) in multivariable linear regression models predicting PSQI, adjusting for age, sex, BMI, hypertension, elevated fasting glucose, and smoking. We repeated the same approach for 12‑month change scores (Δ = 12‑month minus baseline). In addition, we conducted an exploratory statistical mediation analysis testing whether ΔTAC partially explained the association between Δinflammation burden and ΔPSQI using nonparametric bootstrap confidence intervals.

Result

Baseline characteristics and sleep health in elderly participants

A total of 300 participants were examined in this study. The average age was 72.4 years, with a slight majority being female (56.3%) (Table 1). Lifestyle assessments indicated that 48.3% of the participants lived alone, and a vast majority were non-smokers (90.7%) and did not consume alcohol (56.3%) (Table 1). Physical activity levels varied, with 41.3% reporting low activity (Table 1). In terms of health metrics, the mean BMI was 27.1, and 23.6% had hypertension (Table 1). Inflammatory markers like CRP, IL-6, and TNF-α presented median levels within expected ranges for an elderly population (Table 1). TAC was 1.2 ± 0.3 mM Trolox equivalents, suggesting moderate antioxidant defenses (Table 1). Sleep analysis revealed an average duration of 6.8 h with a sleep efficiency of 82.4%, and the average PSQI score was 7.1 (Table 1).

Table 1.

Baseline assessment of study participants

Variable	Value
Age (years), mean ± SD	72.4 ± 5.6
Sex, n (%)
Female	169 (56.3%)
Male	131 (43.7%)
Living status, n (%)
Alone	145 (48.3%)
With Family	133 (44.3%)
Assisted Living	22 (7.3%)
Ethnicity, n (%)
Han	286 (95.3%)
Other ethnic minority	14 (4.7%)
Smoking, n (%)
Yes	28 (9.3%)
No	272 (90.7%)
Education
Primary or less	164 (54.7%)
Secondary	108 (36.0%)
College or above	28 (9.3%)
Residence, n (%)
Urban	182 (60.7%)
Rural	118 (39.3%)
Alcohol use, n (%)
None	169 (56.3%)
Moderate	119 (39.7%)
Heavy	12 (4.0%)
Physical activity, n (%)
Low	124 (41.3%)
Moderate	113 (37.7%)
High	63 (21.0%)
Body Mass Index (kg/m2), mean ± SD	27.1 ± 3.8
Comorbidity, n (%)
Hypertension (BP ≥ 140/90)	88 (29.3%)
Dyslipidemia	128 (42.7%)
Coronary heart disease	61 (20.3%)
Stroke	39 (13.0%)
Diabetes	72 (24.0%)
Fasting glucose
High (> 100 mg/dL)	83 (27.7%)
Normal	217 (72.3%)
CRP (mg/L), median [IQR]	2.3 [1.2–4.1]
IL-6 (pg/mL), median [IQR]	3.8 [2.1–5.6]
TNF-α (pg/mL), median [IQR]	5.2 [3.0–7.0]
TAC mM Trolox equivalents, mean ± SD	1.2 ± 0.3
Average Sleep Duration (hrs), mean ± SD	6.8 ± 1.2
Sleep Efficiency (%), mean ± SD	82.4 ± 7.1
PSQI Global Score (0–21), mean ± SD	7.1 ± 3.2
PSQI > 5, n (%)	186 (62.0%)
PSQI ≥ 8, n (%)	109 (36.3%)
Sleep efficiency < 80%, n	97 (32.3%)
Sleep duration < 6 h, n (%)	79 (26.3%)
Medication, n (%)
Aspirin/antiplatelet therapy	84 (28.0%)
Statin/lipid-lowering agents	112 (37.3%)
Hypoglycemic medication	61 (20.3%)
Anti-inflammatory drugs	38 (12.7%)

Data are shown as mean ± SD, median [IQR], or count (percentage). Sleep duration reflects actigraphy‑derived total sleep time. Sleep efficiency = (total sleep time/time in bed) × 100%. PSQI global score ranges 0–21 (higher = worse sleep)

Regression analysis of inflammatory markers, TAC, and other predictors of sleep quality in the elderly

Univariate and multivariate regression analysis were conducted to evaluate factors associated with sleep quality in the elderly (Table 2). CRP, IL-6, and TNF-α were positively associated with poor sleep quality in the univariate model (p < 0.05) (Table 2). These relationships persisted in the multivariate model, albeit with slightly reduced β coefficients for CRP (0.60, 95% CI: 0.20–1.10, p = 0.008), IL-6 (0.30, 95% CI: 0.00–0.60, p = 0.040), and TNF-α (0.70, 95% CI: 0.20–1.20, p = 0.007) (Table 2). Total antioxidant capacity showed a negative association with poor sleep quality in the univariate and multivariate model (Table 2). Age, and sex were modest predictors in univariate analysis but did not maintain significance in the multivariate model (Table 2). Hypertension remained statistically significant after adjustment, while BMI and elevated fasting glucose showed borderline associations in the multivariable model (Table 2). Smoking status was a weaker predictor and did not reach statistical significance in the multivariate model (Table 2).

Table 2.

Univariate and multivariate regression analyses of predictors for sleep quality

	Univariate analysis		Multivariate analysis
	β (95% CI)	p-value	β (95% CI)	p-value
CRP (mg/L)	0.80 (0.30–1.30)	0.002	0.60 (0.20–1.10)	0.008
IL-6 (pg/L)	0.50 (0.10–0.90)	0.010	0.30 (0.00–0.60)	0.046
TNF-α (pg/L)	0.90 (0.40–1.40)	0.001	0.70 (0.20–1.20)	0.007
TAC (mM Trolox equivalents)	-0.60 (-1.10 to -0.10)	0.020	-0.40 (-0.90 to 0.00)	0.047
Age (per year)	0.05 (0.01–0.09)	0.012	0.02 (-0.01 to 0.05)	0.200
Sex (female vs. male)	0.30 (0.10–0.50)	0.005	0.20 (-0.05 to 0.45)	0.120
BMI (kg/m2)	0.04 (0.00–0.08)	0.045	0.03 (-0.01 to 0.07)	0.070
Hypertension (yes vs. no)	0.60 (0.20–1.00)	0.004	0.40 (0.10–0.70)	0.010
Elevated Fasting Glucose (> 100 mg/dL) (yes vs. no)	0.50 (0.20–0.80)	0.002	0.30 (0.00–0.60)	0.051
Current Smoking (yes vs. no)	0.40 (0.05–0.75)	0.026	0.20 (-0.10 to 0.50)	0.170

Outcome is PSQI global score (0–21; higher = worse sleep)

β represents PSQI points per unit increase in the predictor

Multivariable models adjust for age, sex, BMI, hypertension, elevated fasting glucose, and smoking status

Model fit summary for regression analyses on sleep quality predictors

Table 3 summarizes the model fit for three different regression models analyzing predictors of sleep quality. Model 1, which considers only inflammation markers as a predictor. The model achieves statistical significance with an F-statistic of 8.65 (df1 = 1, df2 = 371) and a p-value of 0.003, and has a root mean square error (RMSE) of 2.98 (Table 3). Model 2, which uses only Total Antioxidant Capacity as the predictor, exhibits lower explanatory power, with an F-statistic of 3.40 (df1 = 1, df2 = 371) and a non-significant p-value of 0.066, yielding an RMSE of 3.05 (Table 3). The full multivariate Model 3, incorporating multiple predictors, shows a substantial improvement in model fit, indicating a better prediction of sleep quality (Table 3).

Table 3.

Model fit summary for predictors of sleep quality

Model	R 2	Adjusted R2	p-value	RMSE
Model 1: Univariate (inflammation markers only)	0.18	0.14	0.003	2.98
Model 2: Univariate (TAConly)	0.12	0.09	0.066	3.05
Model 3: Full Multivariate	0.46	0.44	< 0.001	2.65

Changes in inflammatory markers, TAC, and sleep metrics over 12 months

Changes in biomarker levels and sleep metrics were evaluated over 12 months (Table 4). For inflammatory markers, median levels of CRP decreased from 2.4 mg/L at baseline to 2.0 mg/L at 12 months, showing significant reductions at both 6 and 12 months (Table 4). Similarly, IL-6 levels decreased from 3.3 pg/mL to 2.9 pg/mL, with significant changes noted across the same intervals (Table 4). TNF-α also showed a significant reduction (Table 4).

Table 4.

Changes over 12 months

Measure	Baseline	Month 6	Month 12	p-value (Baseline vs. 6 Mo)	p-value (Baseline vs. 12 Mo)
CRP (mg/L)	2.4 ± 1.6	2.1 ± 1.4	2.0 ± 1.3	0.030	0.015
IL-6 (pg/mL)	3.3 ± 2.2	3.1 ± 2.0	2.9 ± 1.9	0.040	0.022
TNF-α (pg/mL)	5.0 ± 2.6	4.8 ± 2.4	4.6 ± 2.3	0.035	0.018
TAC (mM TE)	1.1 ± 0.3	1.2 ± 0.4	1.3 ± 0.3	0.020	0.020
Actigraphy: Sleep Duration (h/night)	6.6 ± 1.3	6.7 ± 1.2	6.8 ± 1.1	0.045	0.018
Actigraphy: Sleep Efficiency (%)	81.5 ± 7.2	82.0 ± 7.1	82.4 ± 6.9	0.038	0.012

Values are mean ± SD unless otherwise noted. p values reflect paired comparisons vs. baseline using paired t‑tests or Wilcoxon signed‑rank tests depending on distribution. n indicates participants with valid paired data for each comparison

Total Antioxidant Capacity remained relatively stable, showing a slight increase, with statistically significant improvement observed (Table 4).

Sleep metrics, assessed via actigraphy, indicated slight improvements over the year. Sleep duration increased from an average of 6.6 h to 6.8 h, with significant differences noted at both 6 and 12 months (p-values: 0.045 and 0.018). Sleep efficiency also improved, rising from 81.5% ± 7.2 to 82.4% ± 6.9, with significant changes observed (Table 4).

Regression analysis: associations of biomarker changes with sleep quality

Table 5 details the results from regression analyses exploring the impact of changes in inflammatory markers and TAC on sleep quality. In the univariate analysis, increases in C-reactive protein (ΔCRP) were strongly associated with poorer sleep quality (p = 0.002, Table 5). Similarly, IL-6 (ΔIL-6) and TNF-α (ΔTNF-α) also negatively impacted sleep (Table 5). On the other hand, increases in total antioxidant capacity (ΔTotal Antioxidant Capacity) positively influenced sleep quality. The multivariate analysis confirmed these findings but with slightly reduced impact: ΔCRP at a β of 0.88 (p = 0.007), ΔIL-6 at 0.72 (p = 0.034), ΔTNF-α at 0.54 (p = 0.005), and ΔTotal Antioxidant Capacity at -0.61 (p = 0.012). These results highlight the significant role that both inflammatory and TAC play in affecting sleep quality over time (Table 5). The combination of the inflammatory markers and total antioxidant capacity demonstrated a better prediction power than the any biomarker alone (Table 5).

Table 5.

Univariate and multivariate regression of changes in inflammatory and antioxidant markers predicting sleep quality

	Univariate analysis		Multivariate analysis
	β (95% CI)	p-value	β (95% CI)	p-value
ΔCRP (mg/L)	0.95 (0.80–1.40)	0.002	0.88 (0.65–1.21)	0.007
ΔIL-6 (pg/L)	0.80 (0.62–1.38)	0.030	0.72 (0.52–0.96)	0.034
ΔTNF-α(pg/L)	0.62 (0.45–1.49)	0.001	0.54 (0.37–0.87)	0.005
ΔTAC (mM Trolox equivalents)	-0.67 (-0.45 to -0.89)	0.005	-0.61 (-0.35 to -1.25)	0.012
Δ(CRP, IL-6, TNF-α, TAC)	1.35 (1.09–2.14)	< 0.001	1.24 (1.01–2.03)	< 0.001

Δ indicates 12‑month minus baseline. Outcome is ΔPSQI (negative values indicate improvement). Multivariable models adjust for age, sex, BMI, hypertension, elevated fasting glucose, and smoking. The composite inflammatory–antioxidant change index was computed as the average of z‑scored ΔCRP, ΔIL‑6, and ΔTNF‑α and reverse‑scored z‑ΔTAC (higher values indicate more unfavorable inflammatory/antioxidant change)

In models including an inflammation burden index, TAC significantly moderated the association between inflammation and PSQI such that the inflammation–PSQI association was stronger at lower TAC (Supplementary Table S1). Similar moderation was observed for within‑person changes: increases in TAC attenuated the coupling between worsening inflammation and worsening PSQI over 12 months (Supplementary Table S1).

In an exploratory mediation analysis using 12‑month change scores, more unfavorable change in the inflammatory burden index was associated with a smaller increase in TAC (path a: β = −0.075 mM TE per 1 SD higher inflammatory change, p < 0.001) (Supplementary Table S2). Greater increases in TAC were, in turn, associated with greater improvement in sleep quality (path b: β = −1.46 PSQI points per 1 mM increase in TAC, p < 0.001), after accounting for inflammation change (Supplementary Table S2). The bootstrapped indirect effect through TAC was statistically significant (a×b = 0.110; 95% CI 0.050–0.176), corresponding to ~7.5% of the total association between inflammatory change and PSQI change (Supplementary Table S2). However, these findings are hypothesis‑generating given contemporaneous biomarker and PSQI measurement at each time point.

Discussion

Our study systematically examined the longitudinal relationships between inflammatory markers, TAC, and sleep quality in the elderly over a 12-month period. We observed significant correlations between elevated inflammatory markers, antioxidant capacities and sleep quality. The inflammatory markers showed a progressive reduction in levels correlating with an improvement in sleep quality, whereas increased antioxidant capacities were associated with improvements in sleep quality. Beyond baseline associations, our longitudinal analyses extended prior work by showing that within‑person improvements in sleep quality tracked with concurrent favorable shifts in inflammatory markers and TAC over 12 months. The combination of inflammatory markers and TAC supports an integrated inflammatory–oxidative perspective on sleep health in the elderly.

The significance of these findings lies in their implications for understanding and managing sleep health in the elderly. The positive and negative correlations observed between these biomarkers and sleep metrics highlight the potential for targeted interventions that could modify inflammatory responses and enhance antioxidant defenses to improve sleep outcomes in this population.

In our 12-month observational cohort of the elderly, within person improvements in sleep tracked with declines in CRP, IL 6, and TNF α and increases in TAC, which is supported with experimental sleep loss/fragmentation activates inflammatory pathways in humans, and meta-analyses link sleep disturbance with higher CRP/IL 6 at the population level [19–21]. Clinically, insomnia disorder shows a daytime shift of IL 6/TNF α secretion, while OSA displays elevated systemic inflammation with partial reductions after CPAP [24–26]. Differences in age/comorbidity, sleep phenotype, measurement modality, and assay platform plausibly account for heterogeneity without changing the overall direction. Professional guidance supports actigraphy as a valid clinical tool, and standard TAC assays highlight method dependent scaling [17, 18, 27]. Observational work in OSA also reports lower antioxidant capacity at baseline with signals of improvement after CPAP, consistent with the TAC increases accompanying better sleep in our cohort [28, 29].

Sleep fragmentation/short sleep can elevate sympathetic/HPA tone and activate NF κB, increasing IL 6 and TNF α, which drive hepatic CRP. Reciprocally, cytokines can perturb sleep architecture, indicating a bidirectional loop [10, 20, 24]. Oxidative stress interlocks with inflammation, prominent in OSA, where intermittent hypoxia amplifies ROS and inflammatory signaling, so a functional redox readout like TAC complements inflammation when characterizing the milieu accompanying poor sleep [17, 18, 30]. Immunosenescence/inflammaging provides a higher basal inflammatory–oxidative tone that can magnify these links, helping explain why cohort level associations were detectable even after covariate adjustment and when modeled as within person change across 0/6/12 months [31].

Although our cohort showed relatively favorable objective sleep parameters, subjective sleep quality was less favorable, and objective indices changed only modestly over follow-up. Specifically, among participants with paired measurements, sleep duration increased from 6.6 ± 1.3 h at baseline to 6.8 ± 1.1 h at 12 months and sleep efficiency from 81.5 ± 7.2% to 82.4 ± 6.9%, indicating statistically significant but clinically small mean shifts. These relatively high actigraphy means likely reflect both cohort selection and known measurement characteristics of wrist actigraphy. Current clinical guidance supports actigraphy for longitudinal assessment in insomnia and circadian rhythm sleep–wake disorders, but also highlights limited specificity for wake detection—meaning actigraphy can overestimate total sleep time and sleep efficiency when quiet wakefulness is misclassified as sleep [27]. Validation studies comparing actigraphy with polysomnography similarly report high sensitivity for sleep but poorer detection of wake after sleep onset, a limitation that may be accentuated in the elderly with low nocturnal movement [32]. The discordance between objective estimates and PSQI is therefore expected, because PSQI captures perceived sleep quality and daytime consequences that may not be fully reflected by movement-based metrics [33]. Finally, it is important to interpret these apparently “good” mean actigraphy values in context: even modest sleep disturbance has been linked to higher systemic inflammatory markers in meta-analyses and experimental studies, supporting the plausibility of inflammation–sleep associations observed in our cohort despite relatively favorable objective averages [19].

This study quantified within person trajectories at 0/6/12 months for CRP/IL 6/TNF α/TAC in the elderly, triangulated PSQI with actigraphy, and summarized biomarker dynamics via a composite change index in a real-world cohort where CBT I was available as usual care. Practically, pairing low-cost serum markers with TAC and serial actigraphy is feasible for monitoring trajectories and could inform risk stratification or response tracking in future causal studies, which is consistent with AASM actigraphy guidance and standard PSQI use [27, 33]. Interventional signals that CPAP can reduce inflammatory markers in OSA and CBT I can lower inflammatory risk in late life insomnia, together with our longitudinal results, motivate randomized or mechanistic studies to determine whether modifying inflammatory/oxidative pathways changes sleep outcomes [26, 34, 35].

However, these findings should be interpreted within several boundaries. The design is observational, so results are associational. CBT I was a nonrandom, usual care offering and not analyzed, and we made no group contrasts by CBT I exposure. Insomnia was not required for inclusion, we screened with PSQI and excluded known, untreated severe sleep disorders, yet some residual misclassification is possible. Residual confounding and bidirectionality cannot be ruled out despite prespecified adjustment and within person (Δ) analyses. Actigraphy and assay/platform variability may influence measurement, and TAC is a functional metric with assay dependent ranges. Because biomarkers and sleep outcomes were measured contemporaneously at each time point in an observational cohort, we cannot establish temporal ordering required for causal mediation, and residual confounding may remain despite prespecified adjustment. Nearly half of participants reported living alone, which may reflect functional independence and selection of ambulatory elderly able to complete repeated assessments. Because formal functional status measures were not collected, we could not evaluate whether functional status explains this distribution. Finally, we pursued a pre specified, limited hypothesis set and report unadjusted p values, emphasizing effect sizes and convergent PSQI–actigraphy patterns, and generalizability may be constrained by a single center, clinic-based cohort of elderly. Looking ahead, causal and mechanistic work is needed to test whether modifying these pathways changes sleep outcomes. Studies with denser longitudinal sampling should evaluate mediation and directionality.

In sum, inflammation and antioxidant capacity co varied with sleep over 12 months in the elderly. These associational findings are hypothesis generating and motivate causal and translational studies to determine whether modifying inflammatory/oxidative pathways can improve sleep outcomes.

Authors’ contributions

The authors confirm contribution to the paper as follows: study conception and design: H.L.; data collection: J.L., Y.Y.; analysis and interpretation of results: J.L., Y.Y., J.H., L.B., H.H.; draft manuscript preparation: J.L., Y.Y., J.H., L.B., H.H., H.L. All authors reviewed the results and approved the final version of the manuscript.

Funding

No funding.

Data availability

Data sets generated during the current study are available from the corresponding author on reasonable request.

Declarations

Ethics approval and consent to participate

The protocol was approved by the Institutional Review Board of Second Medical Center, PLA General Hospital (KY-2022-13). All study procedures complied fully with the ethical standards outlined in the Declaration of Helsinki. Written informed consent was obtained from all study participants prior to enrollment.

Consent for publication

Not applicable.

Competing interests

The authors declare no competing interests.