The effect of caffeinated beverage consumption on the relationship between sleep quality and major adverse cardiovascular events: Sleep Heart Health Study

Abstract

Study Objectives

Evidence is limited on the role of caffeine intake in the relationship between sleep quality and the incidence of major adverse cardiovascular events (MACE) particularly in patients with sleep breathing disorders. Therefore, this study’s primary aim was to determine the potential confounding effects of total caffeine consumption on the relationship between sleep quality parameters (total sleep time [TST], sleep efficiency [SE], sleep latency [SL], daytime sleepiness, and wakefulness after sleep onset [WASO]) and MACE.

Methods

This study is a secondary analysis of data from the Sleep Heart Health Study (SHHS). Sleep assessments (TST, SE, SL, daytime sleepiness, and WASO) were performed objectively using in-home polysomnography. Caffeine was measured using a survey asking about the average number of cups/cans/glasses of tea, soda, and coffee consumed per regular day and during the last night before polysomnography.

Results

A final sample of 5628 participants was included in SHHS Visit 1 (78 per cent White/Caucasian; 54 per cent female). Cumulative incidence rates measured over 10.9 ± 2.8 years were 15.1 per cent for MACE and 19.7 per cent for all-cause mortality. In univariate models, all sleep measures except SL were associated with MACE; but after adjustment, only TST remained a significant predictor (odds ratio [OR] = 1.122, p = .011). No confounding effect of caffeine was observed in the associations between sleep measures and MACE. Moderate-high intake attenuated MACE risk among individuals with greater daytime sleepiness (OR = 0.91, p = .011).

Conclusions

Caffeine was not a confounding factor in the relationship between sleep measures and MACE. While exploratory analyses suggested potential modification of the association between hypersomnolence and cardiovascular outcomes, these effects were attenuated after statistical adjustment and correction for multiple testing and should be interpreted cautiously.

Statement of Significance.

In this secondary analysis of the Sleep Heart Health Study, we evaluated whether caffeine consumption confounds or moderates the associations between sleep, major adverse cardiovascular events (MACE), and all-cause mortality. Results indicate that caffeine does not significantly confound the role of sleep in predicting these outcomes. While caffeine intake showed a potential to modify the effect of daytime sleepiness on MACE and coronary heart disease risk, these interactions did not maintain statistical significance after adjusting for multiple comparisons. Altogether, these findings emphasize the independent importance of sleep for cardiovascular health and mortality, while suggesting that future sleep research should incorporate prospective assessments of caffeine exposure.

Introduction

Sleep-disordered breathing (SDB) is increasingly becoming recognized as a major, and potentially modifiable, risk factor for major adverse cardiovascular events (MACE) [1, 2]. People with SDB have a twofold higher risk of cardiovascular disease (CVD) [3] and CVD-related mortality [4]. Poor sleep quality in patients with SDB, characterized by reduced sleep duration or total sleep time (TST), low sleep efficiency (SE), prolonged sleep latency (SL), and frequent awakenings or wakefulness after sleep onset (WASO), has also been strongly associated with an increased risk of coronary heart disease (CHD) [5] and MACE [1].

Although the research is limited, patients with SDB consume more caffeine than people without the condition [6], often used as a strategy to combat daytime sleepiness or prevent falling asleep. Research on the effects of daily caffeine consumption (primarily in the form of coffee) on CVD outcomes has been inconsistent [7]. Some evidence suggests that moderate caffeine consumption (typically defined as 300–400 mg per day or 3–5 cups of coffee) may offer protective effects against cardiovascular outcomes such as congestive heart failure (CHF) and all-cause mortality [8, 9]. However, a J-shaped relationship has also been observed, wherein moderate intake is associated with a reduced risk of CHD [10], but excessive coffee consumption (>5 cups/day or >400 mg) is linked to an increased risk of CHD implying that moderate coffee intake could mitigate the risk of CVDs [10]. Furthermore, caffeine’s disruption of sleep architecture may lead to poor sleep quality, which is especially concerning among patients with SDB, as they may already experience heightened poor sleep quality [11]. However, data on the relationship between caffeine and sleep quality in patients with SDB is scarce.

Despite the identification of these complex relationships, a critical research gap remains regarding the relationship between caffeine consumption and CVD outcomes in patients with SDB. Understanding this potential confounding role of caffeine intake is crucial for mapping the mechanism of action and designing future studies to test the cause and effect of caffeine consumption in development of CVD outcomes in patients with SDB.

The Sleep Heart Health Study (SHHS) [12] is a multi-center observational cohort study designed to investigate the relationships between SDB (mainly obstructive sleep apnea) and cardiovascular outcomes. Using data from the SHHS, this study’s aims were to: (1) determine the potential confounding effects of total caffeine consumption on the relationship between sleep quality parameters (TST, SE, SL, daytime sleepiness, and awakenings or WASO) and MACE outcomes (CVD death, CHD, CHF, and all-cause mortality) in patients with SDB from the SHHS, adjusting for age, sex, race, smoking status, diabetes, hypertension, and body mass index (BMI); and (2) explore the potential interaction effects of total caffeine consumption with sleep quality parameters (TST, SE, SL, daytime sleepiness, and awakenings or WASO) on MACE outcomes.

Materials and Methods

This study is a secondary data analysis of the SHHS observational cohort. The SHHS dataset is available through one of the co-authors. A Michigan State University institutional review board (IRB) exemption was obtained (STUDY00011826 [approval date: 14/1/2025]). Details of the study design have been previously reported [12].

Participants (SHHS Visit 1 and SHHS Visit 2)

A total of 6441 individuals were enrolled in the SHHS Visit 1 (SHHS1) [13] between December 1, 1995, and January 31, 1998. Participants underwent a full-montage unattended home polysomnogram during this initial visit. The study population was drawn from several prospective ongoing cohort studies: the Atherosclerosis Risk in Communities Study (N = 1920), Cardiovascular Health Study (N = 1248), Framingham Heart Study (Offspring and Omni cohorts, N = 1000), Strong Heart Study (SHS; N = 602), New York Hypertension Cohorts (N = 760), and Tucson Epidemiologic Study of Airways Obstructive Diseases and the Health and Environment Study (N = 911). The second follow-up examination (SHHS Visit 2 [SHHS2]) consisted of surviving members of the original cohort who consented to additional assessments including questionnaires, blood pressure measurements, and comprehensive health interviews between January 2001 and June 2003.

All participants provided written informed consent, and the study protocols were approved by the IRBs of each participating institution. Cardiovascular outcomes data were collected from baseline through 2011 and confirmed by the parent cohorts. Events, such as myocardial infarction (MI) or stroke, were adjudicated based on surveillance data from the parent cohorts.

Inclusion and exclusion criteria

To be enrolled at SHHS1, participants were required to be at least 40 years of age. Individuals of both sexes and all races/ethnicities were included. Those with obstructive sleep apnea who received oxygen therapy, continuous positive airway pressure, oral devices for breathing during sleep, or tracheostomy were excluded. Subsequent treatment for OSA was assessed at SHHS2.

Prevalent CVD was defined as having a history of physician-diagnosed angina, heart failure, MI, stroke, or coronary revascularization. This information was obtained through adjudicated surveillance data from the parent cohorts or participant self-report at enrollment.

Exposures

Sleep measurements

The first visit or SHHS1 involved the collection of in-home polysomnography data during the baseline examination (n = 5112). Repeat examination was conducted as part of the second visit or SHHS2 roughly 4 years later (1999 to 2001). Sleep studies were done in an unsupervised environment, typically at the participants’ residences, with the Compumedics PS polysomnography system conducted by trained and qualified personnel [14, 15].

Sleep quality parameters

The five sleep quality parameters (a primary exposures) examined in this study included: TST, SL (interval from lights-off to the first epoch of sleep), SE (ratio of TST to time in bed [expressed as a percentage] based on polysomnography data), WASO (total minutes awake after sleep onset until final awakening), and daytime sleepiness (measured by the Epworth Sleepiness Scale [ESS]) [16]. Detailed information about the sleep measures [12, 14, 15, 17] is provided in Supplementary Table 1.

Caffeine consumption

Caffeine consumption, the second primary exposure of interest in this study, was assessed in both SHHS1 and SHHS2 using participant questionnaires, but only SHHS1 data were used in this analysis. SHHS1 provided complete, baseline-consistent exposure, and outcome data aligned with the study’s aims. Participants were asked about their daily intake of caffeinated beverages, specifically the number of cups of coffee, glasses or cans of soda, and cups of tea consumed. The unit of measurement was defined as cups for coffee and tea, and cans or glasses for soda. For all caffeinated drinks, participants were asked to report their consumption during two reference periods: (1) the 4 h before bed on the previous night (of polysomnography) and (2) on a regular day. Total cups/cans/glasses of caffeinated beverages were calculated as total cups/cans/glasses per day using the following formula:

If either the regular day or last night’s value was missing, then the average was represented by the non-missing value. The analysis further stratified the data (Table 2) to examine the independent effects of “last night” (acute) and “regular day” (chronic) consumption periods separately. For the sake of brevity, the term cups will be used henceforth to refer collectively to cups, cans, and glasses.

Table 2.

Regression models of caffeinated beverages for prediction of outcomes

Outcome and intake period	Characteristic	Univariate OR (95% CI) [P-value]	Model 1 OR (95% CI) [P-value]
MACE
Last night	Total cups	1.022 (0.923 to 1.125) [.663]	1.031 (0.919 to 1.149) [.588]
	Tea	1.010 (0.806 to 1.239) [.926]	1.056 (0.817 to 1.336) [.661]
	Soda	0.849 (0.680 to 1.042) [.132]	0.968 (0.753 to 1.222) [.793]
	Coffee	1.119 (0.974 to 1.272) [.097]	1.058 (0.902 to 1.228) [.472]
Regular day	Total cups	1.016 (0.993 to 1.038) [.172]	1.035 (1.007 to 1.064) [.013]
	Tea	1.041 (0.986 to 1.096) [.131]	1.070 (1.001 to 1.138) [.036]
	Soda	0.997 (0.930 to 1.064) [.930]	1.100 (1.014 to 1.189) [.019]
	Coffee	1.012 (0.984 to 1.038) [.395]	1.016 (0.984 to 1.048) [.320]
Total (overall)	Total cups	1.013 (0.957 to 1.073) [.646]	1.065 (0.994 to 1.141) [.072]
	Tea	1.334 (1.048 to 1.686) [.018]	1.264 (0.951 to 1.670) [.102]
	Soda	1.061 (0.836 to 1.337) [.623]	1.463 (1.102 to 1.932) [.008]
	Coffee	1.009 (0.928 to 1.096) [.834]	1.023 (0.925 to 1.130) [.660]
CHF
Last night	Total cups	0.959 (0.823 to 1.098) [.565]	1.072 (0.906 to 1.242) [.383]
	Tea	0.803 (0.546 to 1.104) [.217]	0.829 (0.521 to 1.219) [.385]
	Soda	0.725 (0.513 to 0.981) [.051]	1.033 (0.705 to 1.440) [.859]
	Coffee	1.131 (0.942 to 1.326) [.153]	1.177 (0.951 to 1.418) [.105]
Regular day	Total cups	0.935 (0.898 to 0.971) [<.001]	0.999 (0.954 to 1.041) [.954]
	Tea	0.966 (0.879 to 1.047) [.439]	1.021 (0.913 to 1.123) [.699]
	Soda	0.851 (0.754 to 0.949) [.006]	0.971 (0.846 to 1.102) [.664]
	Coffee	0.944 (0.900 to 0.987) [.014]	0.997 (0.945 to 1.045) [.919]
Total (overall)	Total cups	0.850 (0.780 to 0.923) [<.001]	0.967 (0.874 to 1.067) [.509]
	Tea	0.755 (0.519 to 1.074) [.130]	0.735 (0.477 to 1.103) [.148]
	Soda	0.609 (0.422 to 0.860) [.006]	0.931 (0.604 to 1.405) [.739]
	Coffee	0.885 (0.785 to 0.993) [.041]	1.014 (0.877 to 1.167) [.849]
Revascularization
Last night	Total cups	1.084 (0.954 to 1.217) [.191]	1.044 (0.905 to 1.186) [.530]
	Tea	1.156 (0.876 to 1.470) [.270]	1.164 (0.871 to 1.499) [.269]
	Soda	1.038 (0.793 to 1.321) [.771]	0.983 (0.731 to 1.279) [.902]
	Coffee	1.092 (0.901 to 1.287) [.330]	1.035 (0.835 to 1.243) [.731]
Regular day	Total cups	1.051 (1.023 to 1.079) [<.001]	1.035 (1.002 to 1.067) [.029]
	Tea	1.019 (0.940 to 1.090) [.625]	1.039 (0.955 to 1.117) [.327]
	Soda	1.079 (0.995 to 1.163) [.055]	1.054 (0.958 to 1.150) [.262]
	Coffee	1.051 (1.019 to 1.083) [.001]	1.029 (0.991 to 1.065) [.118]
Total (overall)	Total cups	1.121 (1.041 to 1.205) [.002]	1.087 (0.999 to 1.181) [.051]
	Tea	1.338 (0.969 to 1.821) [.070]	1.421 (0.991 to 2.006) [.051]
	Soda	1.376 (1.011 to 1.851) [.038]	1.463 (1.035 to 2.044) [.028]
	Coffee	1.120 (1.004 to 1.247) [.040]	1.056 (0.932 to 1.192) [.388]
Any CHD
Last night	Total cups	1.044 (0.933 to 1.158) [.432]	1.020 (0.899 to 1.146) [.748]
	Tea	1.131 (0.895 to 1.396) [.276]	1.165 (0.898 to 1.476) [.225]
	Soda	0.912 (0.716 to 1.137) [.436]	0.961 (0.732 to 1.232) [.765]
	Coffee	1.079 (0.918 to 1.246) [.328]	0.998 (0.829 to 1.177) [.984]
Regular day	Total cups	1.026 (1.001 to 1.051) [.036]	1.029 (0.998 to 1.059) [.057]
	Tea	1.038 (0.975 to 1.098) [.214]	1.052 (0.976 to 1.125) [.152]
	Soda	1.013 (0.939 to 1.087) [.720]	1.053 (0.963 to 1.146) [.241]
	Coffee	1.024 (0.995 to 1.052) [.096]	1.018 (0.983 to 1.052) [.292]
Total (overall)	Total cups	1.036 (0.972 to 1.103) [.277]	1.048 (0.972 to 1.129) [.222]
	Tea	1.326 (1.011 to 1.723) [.038]	1.288 (0.943 to 1.742) [.106]
	Soda	1.056 (0.807 to 1.369) [.687]	1.234 (0.901 to 1.673) [.183]
	Coffee	1.049 (0.955 to 1.149) [.315]	1.036 (0.928 to 1.154) [.529]
Mortality
Last night	Total cups	1.017 (0.925 to 1.111) [.725]	1.141 (1.022 to 1.273) [.018]
	Tea	0.932 (0.751 to 1.137) [.507]	1.064 (0.823 to 1.351) [.621]
	Soda	0.753 (0.608 to 0.919) [.007]	1.107 (0.868 to 1.393) [.396]
	Coffee	1.194 (1.056 to 1.344) [.004]	1.207 (1.039 to 1.404) [.014]
Regular day	Total cups	0.970 (0.947 to 0.993) [.012]	1.032 (1.003 to 1.061) [.027]
	Tea	1.009 (0.956 to 1.061) [.728]	1.053 (0.979 to 1.124) [.145]
	Soda	0.902 (0.840 to 0.965) [.004]	1.085 (0.994 to 1.178) [.061]
	Coffee	0.974 (0.946 to 1.000) [.061]	1.021 (0.989 to 1.053) [.198]
Total (overall)	Total cups	0.931 (0.882 to 0.982) [.009]	1.074 (1.003 to 1.151) [.040]
	Tea	0.949 (0.750 to 1.193) [.658]	0.934 (0.700 to 1.239) [.640]
	Soda	0.688 (0.545 to 0.863) [.001]	1.052 (0.784 to 1.405) [.732]
	Coffee	0.942 (0.871 to 1.017) [.128]	1.063 (0.963 to 1.172) [.225]

Model 1 adjusted age, sex, race, diabetes, hypertension, T90, smoking status, and BMI.

Cardiovascular disease outcomes

The primary outcome for this study was the incidence of MACE (follow-up period of 10.9 ± 2.8 years from the baseline polysomnography). MACE was defined as a composite of the following cardiovascular outcomes: Acute MI, cardiovascular death, CHF, coronary surgical intervention/revascularization [18]. Incident cardiovascular events including CVD mortality, any CHD, CHF, and revascularization procedures such as coronary artery bypass grafting or percutaneous coronary intervention (secondary outcomes) were assessed within the six parent cohorts using comprehensive, standardized protocols similar to those used in the Cardiovascular Health Study [2, 19–24].

All-cause mortality, which served as the secondary endpoint for this report, was determined through combination of methods including follow-up interviews, annual written questionnaires or phone contacts with participants or their next-of-kin, monitoring of local hospital records and community obituaries, and cross-referencing with the Social Security Administration Death Master File [25]. Other secondary outcomes included CHD (all types), CHF, and invasive revascularization procedures.

Covariates

To control for confounding, the following nine covariates were collected: age (continuous), sex (male, female), race (White/Caucasian, Black/African American, others), smoking status (never, current, former), diabetes (yes, no), hypertension (yes, no), BMI (continuous), Apnea-Hypopnea Index (AHI; continuous), and T90 (continuous; percentage of TST that oxygen saturation [SpO2] remains below 90 per cent). All multivariate regression models were adjusted for all demographic/lifestyle and clinical covariates. This approach allowed us to isolate the unique contributions of sleep parameters and caffeine to the observed outcomes while minimizing confounding.

Statistical analysis

Descriptive and exploratory

Descriptive statistics were calculated for all exposures, covariates and outcomes, with continuous variables presented as mean ± SD and categorical/nominal variables reported as frequencies and percentages. All exposures (sleep quality and caffeine consumption) and covariates were compared to the primary outcome (MACE; 0 = no, 1 = yes [same as secondary outcomes]) using an independent t-test for continuous variables and a chi-square (or Fisher’s exact) test for categorical/nominal variables. Univariate odds ratios (ORs) and 95 per cent corresponding confidence intervals were calculated for all variables in relation to the primary MACE outcome as well as all secondary outcomes.

Additionally, bivariate correlations between sleep parameters (SL, WASO, SE, and TST) and total cups of caffeinated beverages, specifically tea, coffee, and soda/day, were analyzed using Pearson coefficients. For all exposure variables, the first (Q1) and third (Q3) quartiles were calculated, and the interquartile range (IQR) was defined as Q3 − Q1. Observations lying below Q1 − 1.5 × IQR or above Q3 + 1.5 × IQR were considered outliers and were excluded from subsequent analyses.

Modeling strategy

To address the aims, multivariable logistic regression models were performed to examine the association between sleep quality parameters and MACE outcomes, adjusting for covariates. All five sleep measures (TST, SE, SL, daytime sleepiness, and WASO) were retained as separate variables in the analytical models. These indicators reflect distinct yet complementary aspects of sleep quality, and prior research has demonstrated that each captures unique variance associated with health outcomes [26, 27]. Treating them separately avoids the loss of information that may occur when combining these different dimensions of sleep into a single composite index [28]. This analytic approach has been used previously in epidemiologic studies examining sleep and cardiometabolic outcomes [29].

Aim 1: First, each individual sleep measure was modeled as a separate parameter in a single multivariable logistic regression model that also included all nine covariates (Model 1). To determine the confounding effect of caffeine, we next added total cups of caffeinated beverages to the model (Model 2). Total cups of caffeinated beverages were a confounder if inclusion in Model 2 changed the estimated associations (beta coefficients) between each of the five sleep quality parameters and the MACE outcomes by more than 10 per cent (from Model 1).

Aim 2: To evaluate whether caffeine intake modified the associations between each sleep parameter and the MACE outcomes, we tested separate interaction models for each of the five sleep measures: TST, SE, SL, daytime sleepiness (ESS score), and WASO. For interpretability and to facilitate interaction assessment, daily caffeine intake (total cups of caffeinated beverages) was categorized from a continuous variable into three groups: 0 (none), 1 (regular, 1–2 cups), and 2 (moderate–high, ≥3 cups). This categorization was based on the distribution of intake in the analytic sample to ensure adequate observations within each category for interaction analyses, and to reflect conceptually meaningful distinctions between non-consumers, regular, moderate-high habitual caffeine intake. Each sleep measure was analyzed as a continuous variable. Interaction terms were created by multiplying the continuous sleep variable by the categorical caffeine variable (sleep × caffeine group). Statistical significance of the interaction was evaluated using the Wald χ² test with three degrees of freedom (reflecting the four-level categorical variable for caffeine intake). A p-value <.05 was considered evidence of significant effect modification. Analyses were conducted in two stages. The minimally adjusted model included the sleep parameter, the categorical caffeine variable, and their interaction term. The fully adjusted model additionally included relevant covariates. To account for multiple comparisons in secondary and exploratory interaction analyses, p-values were adjusted using the Benjamini–Hochberg false discovery rate (FDR) procedure.

Results

A total of 5628 participants remained in the dataset after removing 134 participants who had withdrawn consent and 679 participants with prevalent MACE at baseline (revascularization [n = 169], MI [n = 188], stroke [n = 156], and CHF [n = 166]; see Figure 1). Most of the final sample were White/Caucasian (78.2 per cent) and just over half were female (54.7 per cent). The incidence of MACE was 15.1 per cent. The incidence of the secondary outcomes was as follows: all-cause mortality (19.74 per cent), any CHD (11.92 per cent), CHF (7.80 per cent), and any revascularization procedure (7.90 per cent). Demographic/lifestyle and clinical data of the included participants are provided in Table 1.

Figure 1.

Flowchart of participants remained in the dataset for final analysis.

Table 1.

Baseline demographic, clinical, and lifestyle characteristics by MACE outcome: univariate logistic regression results

Characteristic	N	Overall N = 5628*	No N = 4745*	Yes N = 883*	P-value†
Age	5628	62.26 (10.86)	60.98 (10.58)	69.09 (9.75)	<.001
BMI	5587	28.43 (5.26)	28.36 (5.30)	28.82 (5.00)	.012
Gender	5628				<.001
Female		3081 (54.74%)	2694 (56.78%)	387 (43.83%)
Male		2547 (45.26%)	2051 (43.22%)	496 (56.17%)
Race as reported by parent cohort	5628				.780
White/Caucasian		4406 (78.29%)	3708 (78.15%)	698 (79.05%)
Black/African American		467 (8.30%)	394 (8.30%)	73 (8.27%)
Others		755 (13.42%)	643 (13.55%)	112 (12.68%)
Smoking status	5585				<.001
Never		2617 (46.86%)	2272 (48.28%)	345 (39.25%)
Current		641 (11.48%)	522 (11.09%)	119 (13.54%)
Former		2327 (41.67%)	1912 (40.63%)	415 (47.21%)
Hypertension	5385	1935 (35.93%)	1528 (33.82%)	407 (46.94%)	<.001
Diabetes mellitus	5355	448 (8.37%)	287 (6.39%)	161 (18.68%)	<.001
Assigned parent cohort ID	5628				<.001
CHS		993 (17.64%)	634 (13.36%)	359 (40.66%)
ARIC		1737 (30.86%)	1504 (31.70%)	233 (26.39%)
STRONG-HEART		396 (7.04%)	304 (6.41%)	92 (10.42%)
FRAMINGHAM		953 (16.93%)	820 (17.28%)	133 (15.06%)
NEW YORK		760 (13.50%)	760 (16.02%)	0 (0.00%)
TUCSON		789 (14.02%)	723 (15.24%)	66 (7.47%)
Total cups of caffeinated beverages	5456				.774
None		1177 (21.57%)	1001 (21.74%)	176 (20.66%)
Regular (1–2 cups/day)		3425 (62.77%)	2885 (62.66%)	540 (63.38%)
Moderate-high (≥3 cups/day)		854 (15.65%)	718 (15.60%)	136 (15.96%)
AHI	4951	9.35 (12.44)	8.99 (12.28)	11.27 (13.16)	<.001
Hypopnea (4% desaturation) Index (T90%)	5628	3.33 (9.91)	3.01 (8.99)	5.07 (13.74)	<.001
TST, h	5549	6.05 (1.00)	6.07 (0.99)	5.95 (1.03)	.002
ESS	5361	7.56 (4.18)	7.55 (4.17)	7.63 (4.24)	.636
WASO, min	5394	55.39 (34.31)	54.25 (33.98)	61.63 (35.43)	<.001
SL, min	5290	9.81 (11.76)	9.79 (11.75)	9.93 (11.85)	.750
SE, %	5443	84.04 (8.60)	84.32 (8.52)	82.52 (8.87)	<.001
Total cups of caffeinated beverages (overall)	5456	1.44 (1.27)	1.44 (1.27)	1.46 (1.28)	.647
Total cups of caffeinated beverages (regular day)	5352	2.49 (2.09)	2.49 (2.09)	2.52 (2.08)	.688
Total cups of caffeinated beverages (last night)	5494	0.30 (0.72)	0.29 (0.72)	0.31 (0.73)	.668
Total cups of tea/day (overall)	5236	0.14 (0.30)	0.14 (0.29)	0.16 (0.32)	.025
Total cups of coffee/day (overall)	5357	0.81 (0.88)	0.81 (0.89)	0.81 (0.86)	.831
Total cans of soda/day (overall)	5179	0.18 (0.31)	0.17 (0.31)	0.18 (0.32)	.631

^* n (%); mean (SD).

^†Pearson’s chi-squared test; Welch two sample t-test.

ARIC, Atherosclerosis Risk in Communities Study; CHS, the Cardiovascular Health Study.

No significant differences were observed between the MACE and non-MACE groups in total cups of caffeinated beverages or beverage-specific intake (soda, coffee, tea). Shorter TST (5.95 vs. 6.07 h), lower SE (82.52% vs. 84.32%), and greater WASO (61.63 vs. 54.25 min) were significantly associated with MACE (Table 1). Results of correlation analysis (Figure 2; supplements) showed a significant weak positive correlation between total cups of caffeinated beverages and SE (r = 0.07; p < .001) and daytime sleepiness (r = 0.04; p < .001). Also, a negative weak correlation between total cups of caffeinated beverages and WASO (r = −0.07; p < .001) was noted (Figure 2).

Figure 2.

Associations between sleep parameters and caffeine intake (coffee, tea, soda, and total cups of caffeinated beverages): acute (last night) and chronic (regular day).

Across caffeine-related models, univariate analyses occasionally suggested protective associations for outcomes such as CHF; however, these associations were generally attenuated or reversed after multivariable adjustment. For caffeine consumed the night before assessment, higher total cups of caffeinated beverages (OR = 1.141, p = .018) and coffee intake (OR = 1.207, p = .014) were significantly associated with increased odds of mortality in adjusted models. For regular-day caffeine intake, greater consumption of total caffeinated beverages (OR = 1.035, p = .013), tea (OR = 1.070, p = .036), and soda (OR = 1.100, p = .019) was associated with higher risk of MACE. Notably, soda intake emerged as a particularly strong risk factor, showing consistent associations with both MACE (OR = 1.463, p = .008) and revascularization (OR = 1.463, p = .028), regardless of whether intake was assessed for the prior night or a regular day (Table 2; Supplementary Table 3).

Primary outcome confounding analysis results

In univariate analyses, greater WASO (OR = 1.006, 95% confidence interval [CI] = 1.004 to 1.008, p < .001), lower SE (OR = 0.977, 95% CI = 0.969 to 0.985, p < .001), and shorter TST (OR = 0.891, 95% CI = 0.830 to 0.957, p  = .002) were significantly associated with increased odds of MACE. However, after adjustment for demographic/lifestyle and clinical covariates (Model 1) and further adjustment for caffeinated beverage intake (Model 2), only TST remained significant, with longer sleep duration associated with higher odds of MACE (OR = 1.122, 95% CI = 1.027 to 1.227, p = .011). The results were robust to changes in last night’s total cups of caffeinated beverages, compared with a regular day. No significant associations were observed for SL or daytime sleepiness in either univariate or adjusted models (Table 3). No confounding effect of total cups of caffeinated beverages on MACE prediction was noted.

Table 3.

Regression models (Binary logit [Fisher’s scoring]) for prediction of MACE

	Univariate			Model 1			Model 2
Characteristic	N	OR [95% CI]	P *	N	OR [95% CI]	P *	N	OR [95% CI]	P *
Adjusted for overall total cups of caffeinated beverages
WASO (min)	5396	1.006 [1.004 to 1.008]	<.001	4421	0.999 [0.996 to 1.001]	.375	4421	0.999 [0.996 to 1.002]	.452
SL (min)	5292	1.001 [0.995 to 1.007]	.741	4349	1.000 [0.993 to 1.008]	.958	4349	1.000 [0.992 to 1.007]	.990
SE (%)	5445	0.977 [0.969 to 0.985]	<.001	4468	1.005 [0.995 to 1.015]	.361	4468	1.004 [0.994 to 1.015]	.407
ESS (0–24)	5362	1.004 [0.987 to 1.022]	.642	4427	1.005 [0.984 to 1.026]	.631	4427	1.005 [0.984 to 1.025]	.665
TST (h)	5551	0.891 [0.830 to 0.957]	.002	4563	1.121 [1.026 to 1.226]	.011	4563	1.122 [1.027 to 1.227]	.011
Adjusted for regular day total cups of caffeinated beverages
WASO (min)	5396	1.006 [1.004 to 1.008]	<.001	4421	0.999 [0.996 to 1.001]	.375	4421	0.999 [0.996 to 1.002]	.458
SL (min)	5292	1.001 [0.995 to 1.007]	.741	4349	1.000 [0.993 to 1.008]	.958	4349	1.000 [0.992 to 1.007]	.996
SE (%)	5445	0.977 [0.969 to 0.985]	<.001	4468	1.005 [0.995 to 1.015]	.361	4468	1.004 [0.994 to 1.015]	.415
ESS (0–24)	5362	1.004 [0.987 to 1.022]	.642	4427	1.005 [0.984 to 1.026]	.631	4427	1.005 [0.984 to 1.026]	.637
TST (h)	5551	0.891 [0.830 to 0.957]	.002	4563	1.121 [1.026 to 1.226]	.011	4563	1.122 [1.027 to 1.226]	.011
Adjusted for last night’s total cups of caffeinated beverages
WASO (min)	5396	1.006 [1.004 to 1.008]	<.001	4421	0.999 [0.996 to 1.001]	.375	4323	0.999 [0.996 to 1.002]	.430
SL (min)	5292	1.001 [0.995 to 1.007]	.741	4349	1.000 [0.993 to 1.008]	.958	4254	1.001 [0.993 to 1.008]	.811
SE (%)	5445	0.977 [0.969 to 0.985]	<.001	4468	1.005 [0.995 to 1.015]	.361	4370	1.004 [0.993 to 1.014]	.481
ESS (0–24)	5362	1.004 [0.987 to 1.022]	.642	4427	1.005 [0.984 to 1.026]	.631	4331	1.010 [0.989 to 1.031]	.359
TST (h)	5551	0.891 [0.830 to 0.957]	.002	4563	1.121 [1.026 to 1.226]	.011	4463	1.118 [1.022 to 1.223]	.015

^*Maximum likelihood estimates. Model 1 adjusted age, sex, race, diabetes, hypertension, T90, smoking status, and BMI; Model 2 adjusted age, sex, race, DM, HTN, T90, BMI, smoking status, and total cups of caffeinated beverage.

Secondary outcomes confounding analysis results

For secondary outcomes (Table 4), in univariate models, greater WASO (OR = 1.001, 95% CI = 1.009 to 1.013, p < .001), lower SE (OR = 0.955, 95% CI = 0.948 to 0.963, p < .001), shorter TST (OR = 0.750, 95% CI = 0.702 to 0.801, p < .001), and lower daytime sleepiness (OR = 0.975, 95% CI = 0.959 to 0.991, p = .003) were all strongly associated with higher odds of all-cause mortality. For CHF, each additional minute of WASO was associated with higher odds (OR = 1.009, 95% CI = 1.006 to 1.012), whereas higher SE (OR = 0.968, 95% CI = 0.958 to 0.979) and longer TST (OR = 0.838, 95% CI = 0.761 to 0.922) were associated with lower odds. Similarly, for any CHD, greater WASO (OR = 1.005, 95% CI = 1.003 to 1.007) and shorter TST (OR = 0.861, 95% CI = 0.795 to 0.932) were associated with increased odds, while higher SE was protective (OR = 0.979, 95% CI = 0.970 to 0.988). In contrast, revascularization showed a distinct pattern, with higher daytime sleepiness associated with increased odds (OR = 1.038, 95% CI = 1.014 to 1.062), whereas higher SE (OR = 0.988, 95% CI = 0.977 to 0.999) and longer TST (OR = 0.904, 95% CI = 0.821 to 0.994) were associated with reduced odds. However, after multivariable adjustment (Model 2 and Model 3), several of these associations were attenuated and no longer significant.

Table 4.

Multivariate regression model results to predict secondary outcomes

Outcome and sleep measure	Univariate		Model 1*		Model 2†
	OR [95% CI]	P ‡	OR [95% CI]	P ‡	OR [95% CI]	P ‡
All-cause mortality (n = 1111)
WASO (min)	1.011 [1.009 to 1.013]	<.001	1.003 [1.000 to 1.005]	.045	1.003 [1.000 to 1.005]	.019
SL (min)	1.005 [0.999 to 1.010]	.113	1.005 [0.997 to 1.012]	.203	1.004 [0.996 to 1.011]	.336
SE (%)	0.955 [0.948 to 0.963]	<.001	0.985 [0.975 to 0.994]	.002	0.984 [0.974 to 0.994]	.002
Daytime sleepiness (ESS [0–24])	0.975 [0.959 to 0.991]	.003	0.985 [0.964 to 1.005]	.143	0.985 [0.965 to 1.006]	.167
TST (h)	0.750 [0.702 to 0.801]	<.001	0.898 [0.825 to 0.978]	.014	0.897 [0.822 to 0.978]	.014
CHF (n = 444)
WASO (min)	1.009 [1.006 to 1.012]	<.001	0.999 [0.996 to 1.002]	.607	1.000 [0.996 to 1.003]	.796
SL (min)	0.998 [0.989 to 1.006]	.595	0.992 [0.982 to 1.003]	.148	0.991 [0.981 to 1.001]	.099
SE (%)	0.968 [0.958 to 0.979]	<.001	1.007 [0.993 to 1.021]	.351	1.005 [0.991 to 1.019]	.487
Daytime sleepiness (ESS [0–24])	1.007 [0.984 to 1.032]	.540	1.018 [0.989 to 1.047]	.221	1.017 [0.988 to 1.047]	.241
TST (h)	0.838 [0.761 to 0.922]	<.001	1.009 [0.897 to 1.136]	.877	1.008 [0.895 to 1.137]	.897
Any CHD (n = 671)
WASO (min)	1.005 [1.003 to 1.007]	<.001	0.998 [0.996 to 1.001]	.239	0.999 [0.996 to 1.001]	.321
SL (min)	1.001 [0.994 to 1.008]	.834	0.999 [0.990 to 1.007]	.769	0.997 [0.988 to 1.005]	.435
SE (%)	0.979 [0.970 to 0.988]	<.001	1.006 [0.995 to 1.018]	.269	1.006 [0.994 to 1.017]	.326
Daytime sleepiness (ESS [0–24])	1.010 [0.990 to 1.030]	.320	1.006 [0.984 to 1.029]	.600	1.004 [0.981 to 1.027]	.741
TST (h)	0.861 [0.795 to 0.932]	<.001	1.077 [0.978 to 1.187]	.135	1.071 [0.971 to 1.183]	.171
Revascularization (n = 448)
WASO (min)	1.002 [0.999 to 1.005]	.150	0.998 [0.994 to 1.001]	.191	0.998 [0.994 to 1.001]	.192
SL (min)	1.000 [0.992 to 1.008]	.980	0.999 [0.989 to 1.008]	.767	0.996 [0.986 to 1.006]	.398
SE (%)	0.988 [0.977 to 0.999]	.027	1.004 [0.991 to 1.018]	.516	1.005 [0.991 to 1.018]	.501
Daytime sleepiness (ESS [0–24])	1.038 [1.014 to 1.062]	.002	1.021 [0.995 to 1.047]	.117	1.017 [0.991 to 1.044]	.203
TST (h)	0.904 [0.821 to 0.994]	.037	1.063 [0.949 to 1.193]	.294	1.057 [0.941 to 1.188]	.350

^*Adjusted age, sex, race, DM, HTN, T90, smoking status, and BMI.

^†Adjusted age, sex, race, diabetes, hypertension, T90, BMI, smoking status, and total cups of caffeinated beverage.

^‡Maximum likelihood estimates.

For all-cause mortality, longer WASO remained a significant predictor even after full adjustment (OR = 1.003, 95% CI = 1.000 to 1.005, p = .019), as did lower SE (OR = 0.984, 95% CI = 0.974 to 0.994, p = .002) and shorter TST (OR = 0.897, 95% CI = 0.822 to 0.978, p = .014). For CHF, CHD, and revascularization, all univariate associations lost significance after adjustment, indicating that the crude relationships were largely explained by demographic/lifestyle and clinical covariates (Table 4). No confounding effect of total cups of caffeinated beverages on secondary outcomes was noted.

After applying FDR correction to account for multiple comparisons, only the associations related to all-cause mortality remained statistically significant. Specifically, SE (P-FDR = .011), WASO (P-FDR = .038), and TST (P-FDR = .038) continued to demonstrate significant relationships with the risk of death. Conversely, the association between TST and MACE, which was significant in the final model (p = .011; Table 3), was attenuated and failed to maintain significance following FDR adjustment (P-FDR = .054). All other sleep-related measures across cardiovascular outcomes, including CHD and revascularization, did not reach statistical significance after correcting for multiple testing (Supplementary Table 5).

Interaction analysis of outcomes

For MACE, a significant interaction was maintained in the fully adjusted Model 2 between moderate-high caffeine intake and daytime sleepiness (≥3 cup/day × ESS: OR = 0.91, 95% CI = 0.85 to 0.98, p = .011), suggesting that caffeine may mitigate some risks associated with hypersomnolence (Figure 3). Also, the association between daytime sleepiness and CHD appeared to be moderated by moderate-to-high caffeine intake (≥3 cups/day × ESS: OR = 0.93, 95% CI = 0.86 to 1.00, p = .047), suggesting a potential attenuation of CHD risk among individuals with higher caffeine consumption (Figure 4; Supplementary Table 4).Additionally, CHD models demonstrated a significant interaction between regular caffeine intake and TST in Model 2 (1–2 cup/day × TST: OR = 0.77, 95% CI = 0.60 to 0.98, p = .036), even as the main effect of 1–2 cups/day became more pronounced (OR = 5.55, 95% CI = 1.23 to 25.8, p = .027; Figure 5). In contrast, the protective main effect of caffeine against CHF observed in the WASO group remained robust across both models (Model 2 1–2 cup/day: OR = 0.51, 95% CI = 0.29 to 0.92, p = .023), whereas many univariate sleep associations with mortality were attenuated upon the inclusion of covariates in Model 2. After applying FDR correction to the interaction models, all previously observed associations between sleep measures and caffeine categories across the health outcomes became statistically non-significant (Supplementary Table 6).

Figure 3.

Odds of MACE across levels of daytime sleepiness (ESS score), stratified by caffeine intake group (0, 1–2, 3≥ cups-cans/day).

Figure 4.

Odds of CHD across levels of daytime sleepiness, stratified by caffeine intake group (0, 1–2, 3≥ cups-cans/day).

Figure 5.

Odds of CHD across levels of TST, stratified by caffeine intake group (0, 1–2, 3≥ cups-cans/day).

Discussion

The aim of this study was to determine the potential confounding effects of total caffeine consumption on the relationship between sleep quality parameters (TST, SE, SL, daytime sleepiness, and WASO) and MACE outcomes. Regarding the primary aim, we found that total cups of caffeinated beverages did not show any significant confounding effect on primary and secondary outcomes. Basically, differences in caffeine consumption did not influence the strength or direction of association between sleep quality parameters and the cardiovascular outcomes. In fact, several indicators of poor sleep quality and insomnia, including shorter TST (predictive of MACE and all-cause mortality), greater WASO, and lower SE (both predictive of all-cause mortality and CHD), remained significant predictors even after full multivariable adjustment. This finding is aligned with another study [30] that found poor SE and prolonged WASO were independently associated with higher risks of CVD, composite cardiovascular outcomes, and cardiovascular mortality, even after multivariable adjustment. Additionally, we found caffeine intake may modestly modify associations of daytime sleepiness (ESS) with MACE and CHD, whereas caffeine intake modified associations of TST with CHD. To our knowledge, this study is the first of its kind to assess simultaneously the role of caffeine intake and sleep quality measures as well as the interaction between them on cardiovascular outcomes using a nationwide dataset.

The role of sleep in predicting MACE has been frequently reported [31–33] especially in older adults (65 years and older). Poor sleep quality increased the risk of MACE up to 96 per cent [33]; and compared with older frail people with poor sleep, maintaining good sleep habits was associated with an almost 44 per cent lower risk of MACE in healthy non-frail older adults [31]. Compared to people with poor sleep and poor cardiovascular health, those with healthy sleep had about a 69 per cent lower risk of MACE, and this protective effect was consistent for heart failure, stroke, heart attack, and overall mortality [32]. People with clinical insomnia had roughly a 40%–60% higher risk of recurrent major cardiovascular events, and insomnia explained about one-sixth of all such events, ranking just behind smoking and inactivity as major modifiable risk factors [34]. Our results align with all these findings, emphasizing that poor sleep quality drastically increases the risk of MACE. Knowing that sleep is a modifiable risk factor, patients can reduce the risk of MACE by controlling underlying sleep disorders. The importance of sleep health in developing MACE is even more important in older adults as aging is associated with sleep changes such as declining SE and prolonged SL [35]. This situation is concerning because older adults are less likely to report sleep disorders and related symptoms [36]. Furthermore, in all adjusted models we noticed that age was a significant covariate for MACE outcomes, highlighting the crucial role of sleep in older adults.

Although, our results are rooted in a different generation of people in the US, caffeine intake has changed over the last two decades in the United States [37, 38]. From 2015 to 2025, the US caffeine landscape shifted from widespread moderate consumption to more concentrated and coffee-centered patterns [38]. While fewer people now consume caffeine daily, those who do, especially adults, consume more, largely through coffee. This occurrence reflects the rise of premium coffee culture, energy beverages, and evolving lifestyle patterns in caffeine consumption [38]. In bivariate analysis of caffeine and sleep variables without adjustment for any demographics/clinical data, our results unexpectedly showed that more total cups of caffeinated beverages may be associated with improved quality (higher SE and lower WASO) and less hypersomnolence (ESS). However, these results should not be conclusive as they do not consider the vital roles of age, sex, and comorbidities which impact drinking behaviors and sleep pattern. We found that the regular day caffeine exposure (habitual) is driving the CVD outcome and acute caffeine intake (last day/night) may not be a good predictor of CVD. Although, in models without sleep measures (Table 2) we found that higher total cups of caffeinated beverages (on a regular day) were associated with an increased likelihood of all-cause mortality and higher soda intake (overall) was dramatically related to an increased risk of MACE and revascularization, all caffeine-related variables became non-significant in sleep models (Tables 3 and 4) after adjusting for covariates, implying that the observed associations may be confounded by other lifestyle (smoking) or clinical factors (diabetes, hypertension, AHI) rather than representing independent effects of caffeine itself. A prior study [9] also found not only an inverse relationship between coffee consumption and CHF, but also that the relationship may not be purely linear or independent of other factors. Similarly, large observational meta-analyses of long-term coffee consumption reported that moderate intake (roughly 3–5 cups/day) is inversely associated with CVD risk (relative risks ~0.85) compared to no consumption [39]. Yet, mendelian randomization and genetic-instrument studies challenge a causal interpretation, showing essentially no association when genetic proxies for higher caffeine/coffee intake are used [40]. This result that mirrors our finding that caffeine-related associations disappeared after adjustment for confounders. Thus, while observational evidence suggests potential cardioprotective effects of coffee/caffeine consumption, both our results and genetic evidence highlight the likelihood that such associations are driven by residual confounding rather than a direct protective effect of caffeine itself. Further, our finding of increased revascularization risk (prior to adjustment) may hint at complexities, perhaps reflecting differences in beverage types or unmeasured comorbidities, that align with literature showing heterogeneity of effects by source or context of consumption (e.g. filtered vs unfiltered coffee; heavy vs moderate intake) [41]. Our null findings post-adjustment reinforce the importance of careful control for covariates in caffeine/CVD research and suggest that caffeine consumption alone should not be interpreted as an independent protective factor in cardiovascular outcomes.

The stimulant effect of caffeine is mediated through adenosine receptor antagonism, which reduces “sleep pressure” and delays the onset of sleep [28]. Based on our findings, moderate-to-high caffeine intake appeared to attenuate the association between daytime sleepiness and CHD. Although, after applying FDR all results of interaction became non-significant, mechanistically, for individuals experiencing excessive daytime sleepiness (sleep fragmentation or hypoxemia) habitual caffeine use may improve daytime functioning and physical activity levels [6]. This improvement could indirectly influence cardiometabolic pathways, potentially mitigating some adverse cardiovascular impacts associated with poor sleep quality. However, it must be noted that the association between ESS and CHD did not reach statistical significance when we adjusted total cups of caffeinated beverages. Furthermore, while these physiological mechanisms provide a plausible framework for our observations, they were not directly measured in the current study and warrant further investigation using objective measures of activity and metabolic health. Caffeine consumed within ~3 h of bedtime is most disruptive; consumption earlier in the day has less consistent effects [11, 28]. The primary caffeine exposure was assessed by calculating the average of two self-reported values: intake from the previous night (acute) and intake on a regular basis (chronic). To enhance the accuracy and granularity of the risk estimations, the analysis further stratified the data to examine the independent effects of “last night” (acute) and “regular day” (chronic) consumption periods separately. Habitual caffeine consumers may develop some tolerance, which may moderate the effects on sleep parameters in real-world observational studies [42, 43]. Older adults may be more sensitive; studies suggest greater disruption in SL and SE with caffeine in older compared to younger adults [44, 45].

Sleep continuity measures, particularly WASO and SE, and TST emerged as robust predictors of all-cause mortality, even after accounting for traditional risk factors and caffeine intake, demonstrating the clinical relevance of fragmented sleep as an independent risk pathway. These findings aligned with prior evidence show that greater nocturnal wakefulness and lower SE are associated with heightened inflammation, autonomic dysregulation, and increased risk of mortality and cardiovascular events [46]. Previous studies [47, 48] showed both shorter and longer sleep durations are significant risk factors (making them a U-shape relationship) of all-cause mortality. Although we did not find a U-shape relationship between TST and CVD outcomes (Supplementary Figures 1 and 2), we found longer TST may have some protective effect on all-cause mortality in this population. A previous study [47] from SHHS showed that long sleep duration and shifting from long to short sleep duration independently increased the risk of all-cause mortality, but the researchers examined TST as categorical variable whereas we tested it as continuous variable. In the present analysis, TST retained an independent association with MACE after covariate adjustment, whereas the interaction with caffeine intake was not significant. Also, the effect of TST on the prediction of MACE did not change based on regular day versus last night caffeine intake. This result suggests that shorter sleep duration itself, rather than its modification by caffeine, may represent a direct physiological risk factor for MACE, consistent with prior studies linking reduced sleep time to impaired endothelial function, hypertension, and atherosclerotic progression [29]. For mortality and revascularization outcomes, both SE and daytime sleepiness (ESS score) demonstrated independent associations even after adjusting for confounders, indicating that poor nocturnal consolidation and elevated daytime somnolence may capture distinct yet complementary pathways of cardiovascular vulnerability [49]. These results are consistent with previous research showing that lower SE and excessive daytime sleepiness independently predict incident coronary events and all-cause mortality [1]. Therefore, our findings reinforce the concept that objectively measured sleep quality, particularly efficiency and continuity, constitutes a meaningful, caffeine-independent predictor of long-term cardiovascular risk.

Limitations

This study has some limitations. First, the SHHS dataset is relatively old (1997–2000), and caffeine consumption and sleep quality measures in the United States have changed in the last two decades. Therefore, our results may not be generalizable to the current US population. Also, we did not assess the interaction between caffeine and sleep architecture variables such as per cent time in rapid eye movement sleep and others. Also, we just used SHHS1 results, and we did not assess SHHS2 in terms of caffeine and sleep measure changes in predicting MACE. Caffeine consumption and some other covariates were subjective/self-reported which can introduce social desirability, recall, and misjudgment bias. Additionally, questions related to caffeine intake in SHHS were very broad and may not reflect actual caffeine intake of participant or could be misinterpreted by some individuals. Although effect sizes of interaction models were estimated with 95% CIs, several higher caffeine intake categories exhibited relatively wide intervals, indicating limited precision; therefore, modest effect modification between sleep measures and caffeine intake cannot be definitively excluded. Finally, this study was not a randomized controlled trial; therefore, there could be additional residual confounding.

Implication for practice

Given the observational design, these findings do not support recommending increased caffeine consumption for patients with SDB. However, moderate habitual caffeine intake does not appear to exacerbate cardiovascular risk in this population based on our findings. Clinical guidance should remain individualized and tailored, particularly considering timing of intake and coexisting hypertension or arrhythmia. Finally, it is important to consider that caffeine sources may influence observed associations. Caffeinated beverages differ substantially in their nutritional composition; for example, sugar-sweetened soda contains added sugars that are independently associated with adverse cardiometabolic outcomes [50]. In contrast, coffee [51] and tea [52] contain polyphenols and other bioactive compounds with potential antioxidants and anti-inflammatory properties. Therefore, the pure health impact of caffeine may vary depending on the beverage type in which it is consumed.

Implication for future research

Caffeine consumption in SHHS was assessed subjectively and retrospectively. Prospective caffeine consumption assessment, including ideally blood, saliva, or hair, may be a better approach. Additional analyses examining the longitudinal change of caffeine and sleep measures over time using SHHS1 and SHHS2 can contribute to current literature. Investigating sleep architecture and caffeine intake is critical for understanding mechanisms of other CVD outcomes such as arrhythmia. Future studies with larger sample sizes and prespecified interaction hypotheses are needed to more definitively evaluate potential effect modification between sleep measures and caffeine consumption.

Conclusions

Caffeine intake was not a confounding factor in the relationship between TST, SE, WASO, daytime sleepiness, and cardiovascular outcomes. Although total cups of caffeinated beverages appeared to modify the association between daytime sleepiness and MACE and CHD, this effect attenuated after statistical adjustment and correction for multiple testing, implying that the observed interaction may reflect residual confounding or limited precision rather than a robust independent modifying effect. Further research to confirm this relationship is needed.

Author contributions

Hesam Varpaei (Conceptualization [lead], Formal analysis [equal], Methodology [lead], Project administration [equal], Software [lead], Visualization [lead], Writing—original draft [lead], Writing—review & editing [lead]), Mathew Reeves (Investigation [equal], Methodology [equal], Supervision [equal], Validation [equal], Writing—review & editing [equal]), Lorraine B. Robbins (Investigation [equal], Methodology [equal], Supervision [lead], Validation [equal], Writing—review & editing [equal]), Fabrice Mowbray (Investigation [equal], Methodology [equal], Supervision [equal], Validation [equal], Writing—review & editing [equal]), Pallav Deka (Investigation [equal], Methodology [equal], Supervision [equal], Validation [equal], Writing—review & editing [equal]), and Stuart F. Quan (Data curation [lead], Investigation [equal], Methodology [equal], Supervision [equal], Validation [equal], Writing—review & editing [equal])

Funding

The first author was awarded funding from Michigan State University via a dissertation completion fellowship. The SHHS was supported by National Heart, Lung, and Blood Institute cooperative agreements U01HL53940 (University of Washington), U01HL53941 (Boston University), U01HL53938 (University of Arizona), U01HL53916 (University of California, Davis), U01HL53934 (University of Minnesota), U01HL53931 (New York University), U01HL53937 and U01HL64360 (Johns Hopkins University), U01HL63463 (Case Western Reserve University), and U01HL63429 (Missouri Breaks Research).

Disclosure statement

Financial disclosure: SFQ has served as a consultant for Teledoc, Bryte Foundation, Jazz Pharmaceuticals, Summus, Apnimed, SleepRes, and Whispersom. The other authors have no financial conflict of interest to declare.

Non-financial disclosure: None declared.

Data availability

SHHS data is available through National Sleep Research Resource (https://sleepdata.org/datasets/shhs). All code and syntax used to conduct the analysis for this paper are available upon request from the corresponding author.

Ethics approval and informed consent

This study obtained institutional review board (IRB) exemption from Michigan State University (STUDY00011826 [approval date: 14/1/2025]).

AI use statement

AI was not used to generate the ideas, concepts, or scientific content of this study. OpenAI (Chap GPT 5.1) tools were used solely for English language-related support, including grammar correction, punctuation, and improving clarity and fluency.