Perceived stress correlates with disturbed sleep: A link connecting stress and cardiovascular disease

Abstract

The association between stress and cardiovascular disease (CVD) risk is becoming established. A mechanistic link clarifying the intermediate steps between the experience of stress and the development of CVD would support this association. We sought to examine the role of perceived stress as a factor associated with disturbed sleep with the goal of providing an explanation for the stress–CVD connection. We performed a cross-sectional analysis of data recorded by subjects at entry to our CVD prevention program. Data collection included questionnaire surveys, anthropometrics, and a CVD-relevant laboratory panel. Of 350 consecutively enrolled subjects (mean age 54.4 ± 12.4 [SD] years, 138 men, 39%), 165 (47%) scored above the mean for stress measures. These high-stress subjects displayed an increased cardiovascular risk profile including elevated body mass index (mean ± SD 31.1 ± 5.9 vs. 29.0 ± 5.9, r_s = 0.175), increased waist circumference (102 ± 17 cm vs. 98 ± 14, r_s = 0.135), and elevated high-sensitivity serum C-reactive protein (0.384 mg/dl vs. 0.356, r_s = 0.109). High-stress subjects also demonstrated greater daytime sleepiness (Epworth Sleepiness Scale: 10.4 ± 5.0 vs. 7.8 ± 4.8, r_s < 0.316), greater fatigue (fatigue scale: 5.4 ± 2.2 vs. 3.4 ± 2.4, r_s = 0.484), poorer sleep quality (Pittsburgh Sleep Quality Index: 8.5 ± 4.4 vs. 5.9 ± 4.0, r_s = 0.416), and shorter sleep duration (20 min less/24 h, r_s = negative 0.177) with a higher risk for sleep apnea (60% at high risk vs. 40%, p = 0.003) than low-stress subjects. High stress was associated with significant disturbances in sleep duration and sleep quality. Stress levels also correlated with daytime consequences of disturbed sleep. The stress–sleep connection may be an important mechanistic mediator of the association between stress and CVD.

Keywords::

• Cardiovascular disease
• perceived stress
• risk factors
• sleep
• sleep quality
• stress

Introduction

An emerging body of evidence substantiates the observation that there is an association between stress and the occurrence of cardiovascular disease (CVD; Belkic et al. 2004; Rosengren et al. 2004). While the stress–CVD connection has been promoted and taught for decades, a number of difficulties have slowed a productive line of investigation in this area.

The impediments include how to define and measure stress (Cohen et al. 1983; Kocalevent et al. 2007), how to assess stress levels in a reproducible fashion over time (Kocalevent et al. 2009), uncertainty regarding causation between stress and CVD (Tindle et al. 2010), and the expense of performing a study to follow large numbers of subjects over a substantial period of time (Kadojić et al. 1999; Hamer et al. 2008). Furthermore, some studies appear to contradict the association between stress and CVD (Greenlund et al. 1995; Riese et al. 2000; Heslop et al. 2002a,b; Belkic et al. 2004). Nevertheless, the preponderance of studies to date supports the conclusion that stress, variously defined in a variety of approaches, does correlate with increased cardiovascular risk, both for heart disease (Melamed et al. 1992; Belkic et al. 2004; Rosengren et al. 2004; Brborović et al. 2009; Holden et al. 2010; Puustinen et al. 2010) and stroke (Everson et al. 2001; Surtees et al. 2008; Tsutsumi et al. 2009). Moreover, studies of depressive behaviors in female primates subjected to social stressors over a 4-year period have demonstrated significant acceleration of coronary artery atherosclerosis, suggesting a causal relationship between stress and CVD (Shively et al. 2008).

To substantiate and explain the clinical observations correlating stress and CVD, it would be useful to clarify the underlying pathophysiology to outline mechanisms that link stress and CVD. It is clear that the hypothalamus–pituitary–adrenal axis plays a major role, by stimulating cortisol secretion, as do increased aldosterone and catecholamine levels, with a resulting detrimental effect on the cardiovascular system (Kubzansky and Adler 2010). It remains less clear what maladaptive conditions initiate the cascade of mediators that trigger these responses.

One mechanism was proposed in the Massa Lombarda Project, an epidemiological study including 7000 northern Italian adults (Bove et al. 2010). In a subset of 106 men and women, selected for older age, psycho-emotional stress and depression disorder were associated with the development of metabolic syndrome, a cluster of multiple cardiovascular risk states. Another study examined whether self-reported job strain was associated with early, potentially modifiable cardiovascular (CVD)-related health behaviors (Hellerstedt and Jeffery 1997). This study of 3843 randomly selected men and women employees of 32 worksites in Minnesota showed that work stress, defined as high demand and low latitude, was positively associated with smoking, smoking intensity, and high fat intake in men, and with body mass index (BMI) and smoking intensity in women. In 2008, the most sophisticated studies to date were published to describe the mechanistic links between stress and CVD (Chandola et al. 2008; Hamer et al. 2008). These studies used statistical models to assess the relative contributions of potential mediators of stress and CVD events. The Whitehall II study followed over 10,000 male and female civil servants for an average of 12 years (Chandola et al. 2008). The study showed that two factors, health behaviors and metabolic syndrome, accounted for around 32% of the effect of work stress on CVD. Another study that used statistical modeling was prospective and included 6576 healthy men and women followed over an average of 7.2 years (Hamer et al. 2008). Psychological distress was measured with the validated General Health Questionnaire, and actual CVD events (hospitalization for nonfatal myocardial infarction, coronary artery bypass, angioplasty, stroke, heart failure, and CVD-related mortality) were used as the main outcome. The investigators reported that behavioral factors explained the largest proportion of variance (approximately 65%), whereas pathophysiological factors accounted for a modest amount (C-reactive protein approximately 5.5%; hypertension approximately 13%).

The mechanisms proposed by these studies, while supported by objective data, fail to fully account for the observed relationship between stress and CVD. An often overlooked contributor to ill health and bad medical outcomes is sleep, with important sleep parameters including sleep duration and sleep quality. Failure to include the role of sleep in the stress–CVD connection is especially surprising in view of the substantial personal experience that all humans have of the ill effects of sleep deprivation and disrupted quality of sleep. Understanding the role of sleep as a possible link between stress and CVD is especially appealing because sleep behaviors can be taught and improved. Furthermore, it has been shown that improving sleep quality through the implementation of behavior modification does lower perceived stress levels (Eliasson et al. 2010). Disrupted sleep is thus a modifiable risk factor for stress levels and may therefore be, in extension, a modifiable risk factor for CVD.

We therefore hypothesized that high levels of perceived stress would correlate with disturbed sleep parameters. Such mechanistic link is indicated by substantial prior research showing that short sleep and disrupted sleep are associated with high risk for CVD (Heslop et al. 2002a,b).

Methods

The investigation was conducted with the approval of our institutional review board. The study design is a retrospective analysis of data collected on consecutive patients participating in a CVD prevention program at the Walter Reed Army Medical Center Integrative Cardiac Health Project (ICHP). ICHP is a cardiovascular prevention research center for the US Department of Defense. All data were retrospectively gathered and no blood samples were taken specifically for this study. The institutional review board, therefore, did not request informed consent from the study subjects.

Patients were self-referred or referred to the program by a health-care provider to improve habits of diet, exercise, sleep, and stress management. ICHP is accessible to military health-care beneficiaries including active duty service members, retirees, and dependents. The program, therefore, enrolls a broad spectrum of subjects including a variety of races and ethnic backgrounds, both genders, and a range of ages from 18 to 90 years. The typical patient entering the program is found to have two to four risk factors for CVD.

Upon entry, subjects are asked to complete a series of questionnaires to gather information on demographics, current symptoms, past and current medical conditions including medications and lifestyle habits. Among the questionnaires are the validated surveys to assess stress levels, sleep behaviors, sleep quality, and daytime symptoms from inadequate sleep. Data from the questionnaires are reviewed during a medical interview with a nurse practitioner who also performs a physical examination to include anthropomorphic measures.

Laboratory studies

Subjects gave blood for cardiac-relevant biochemical studies. For all blood samples, subjects were instructed to present to the laboratory between 06:00 and 08:00 h having fasted from 20:00 h the previous evening. The biochemical measurements on blood samples included a standard lipid panel with total cholesterol concentration, low-density lipoprotein (LDL) cholesterol concentration, high-density lipoprotein (HDL) cholesterol concentration, triglyceride concentration, as well as lipoprotein (a) and lipoprotein PLA2 concentrations. Measures of glucose metabolism include fasting plasma glucose concentration, insulin concentration, and hemoglobin A1C %. High-sensitivity C-reactive protein concentration (hsCRP) was also measured.

The laboratory studies were performed in the institution's certified central laboratory. The lipid panel was measured on a Roche Cobas c501 with appropriate reagents. The technique has documented traceability to the National Reference System for Cholesterol by performing a direct comparison with the cholesterol reference method using fresh human specimens, which cover the National Cholesterol Education Program (NCEP) medical decision points. The system has demonstrated the ability to meet the NCEP's performance criteria for accuracy and precision.

Perceived Stress Scale (PSS)

The PSS is one of the most widely accepted measures of stress (Cohen et al. 1983). This validated 14-item questionnaire asks the subject how often certain experiences of stress occurred in the last month and is designed to measure the degree to which situations in one's life are appraised as stressful. With item responses from 0 to 4, the range of possible scores is 0–56 with higher scores correlating with higher stress. The PSS is designed for use in community samples with at least a junior high school education. The items are easy to understand and the response alternatives are simple to grasp. Moreover, the questions are quite general in nature and hence relatively free of content specific to any subpopulation group. Score in the low 20s reveal moderate stress levels, while scores approaching 30 are substantial and concerning.

Pittsburgh Sleep Quality Index

The Pittsburgh Sleep Quality Index (PSQI) is a self-rated questionnaire which assesses sleep quality and disturbances over a 1-month time interval (Buysse et al. 1989). Nineteen individual items generate seven component scores whose sum yields one global score with a range of 0–21. The psychometric and clinical properties of the PSQI suggest its utility both in clinical practice and research activities. A global score of greater than 5 indicates a poor sleeper. Sleep perturbations can be categorized by scores: 0–5 is a good sleep score; 6–10 shows mild sleep difficulty; 11–15 moderate sleep difficulty; and 16–21 severe sleep difficulty.

Epworth Sleepiness Scale

The Epworth Sleepiness Scale (ESS) is the most widely used tool to estimate the subjective symptom of daytime sleepiness (Johns 1992). Subjects were asked to use a scale of 0–3 to estimate their likelihood of dozing in eight different situations in recent weeks. The individual scores were summed and possible scores range from 0 to 24. Sleepy subjects score 10 or higher and sleepiness can be categorized by scores: 10–14 as mild sleepiness, 15–19 as moderate sleepiness, and 20–24 as severe sleepiness.

Fatigue Scale

The Fatigue Visual Numeric Scale is borrowed from the Stanford Patient Education Research Center (see http://patienteducation.stanford.edu/research/vnsfatigue.html, accessed 1 July 2010). This fatigue scale asks subjects to express their experience of fatigue from 0 to 10 for the previous 2-week period. Subjects who circle 5–6 express mild fatigue, 7–8 moderate fatigue, and 9–10 severe fatigue.

Berlin Questionnaire

Of questionnaires available to screen patients for sleep apnea, the Berlin Questionnaire is one of the most commonly utilized and best validated (Netzer et al. 1999). As measured by the questionnaire, patients with persistent and frequent symptoms are considered to be at high risk for sleep apnea. Questions about symptoms demonstrated internal consistency (Cronbach correlations, 0.86–0.92). With a positive Berlin Questionnaire, sleep apnea was predicted with a sensitivity of 0.86, a specificity of 0.77, a positive predictive value of 0.89, and a likelihood ratio of 3.79.

Statistical analysis

Data are presented as mean ± SD. Two sample t-tests were used to compare continuous variables between groups, and categorical data were compared between groups using Fisher's exact test. Body habitus, sleep variables, and hsCRP did not satisfy assumptions of normality (as tested by the Shapiro–Wilk statistic) therefore, Spearman's correlation coefficient; (r_s) was used to examine the association of these variables with the PSS. All tests were two-tailed and p values < 0.05 were presumed to represent statistical significance. Data were analyzed using SPSS for Windows (v. 17.0, SPSS, Inc., (IBM), Chicago, IL, USA).

Results

We studied data from 350 participants entering ICHP's CVD prevention program. The mean age ( ± SD) of our participants was 54.4 ± 12.4 years, consistent with a spectrum of lifestyles from actively working to semi-retired and fully retired adults. Heavily represented racial categories were Caucasian and African American, but a substantial number of subjects identified themselves as mixed race or declined to pick a category. There was a majority of women (61%) in our study sample (see Table I).

Table I.  Characteristics of the subjects according to perceived stress levels.

	All subjects (n = 350)	Low stress (n = 185)	High stress (n = 165)	t statistic	Degrees of freedom	p value
Age, years (mean ± SD)	54.4 ± 12.4	57.4 ± 11.5	51.1 ± 12.6	4.9	348	< 0.001^*
Race (%)
Caucasian	134 (38%)	73 (39%)	60 (36%)			0.673^†
African American	105 (30%)	53 (29%)	52 (31%)
Hispanic	14 (4%)	5 (3%)	9 (5%)
Asian	2 (1%)	1 (1%)	1 (1%)
Others	96 (27%)	53 (29%)	43 (26%)
Gender, male (% male)	138 (39%)	78 (42%)	60 (36%)			0.28^†
PSS score (56 points max)	22.4 ± 8.1	16.3 ± 4.7	29.3 ± 5.0	− 25.0	348	< 0.001^*

Notes: Values are mean ± SD or actual number of subjects in a category (with proportion). Statistical comparisons are between low-stress and high-stress subjects using the two sample t-test (or Fisher exact test as noted) with p values less than 0.05 representing statistical significance. Low-stress subjects were defined by a Perceived Stress Score less than the mean of 23 points, while high-stress subjects were defined by a score equal to or greater than the mean of 23 points

^* denotes two sample t-test between low-stress and high-stress subjects

^† denotes Fisher's exact test between low-stress and high-stress subjects.

As the mean PSS score was 22.4 points, we elected to define subjects with PSS scores of 23 or more points as belonging to the “high-stress” group and subjects with PSS less than 23 as the “low-stress” group. This allowed for analysis of data for nearly equal sized cohorts of high- and low-stress groups. While there are no defined ranges of “normality” or published degrees of severity based upon the PSS scores, the cut point of 23 does conform to a threshold value above which stress becomes a concerning issue from a clinical point of view in our experience within our program.

As summarized in Table I, there were no significant differences with regard to race or gender for high-stress and low-stress groups, though high-stress subjects were somewhat younger (p < 0.001).

As summarized in Table II, the cohort of subjects with high stress had a higher BMI (obese indices vs. merely overweight, p = 0.001) and larger measured waist circumferences. The biochemical measurement of hsCRP showed a positive correlation with perceived stress. The high-stress group also showed shorter total sleep times (20 min less per 24 h), poorer sleep quality, higher likelihood of sleep apnea diagnosis, greater sleepiness, and greater fatigue. The correlation between perceived stress and sleep quality is illustrated in Figure 1.

Table II.  Correlations between perceived stress levels vs. anthropometrics, behavior scores, symptom scores, and laboratory values.

	Low stress (n = 185)	High stress (n = 165)	Correlation coefficient^*	p value two-tailed
BMI (kg/m^2)	29.0 ± 5.9	31.1 ± 5.9	0.175	0.0011
Waist circumference (cm)	98 ± 14	102 ± 17	0.135	0.012
Total sleep time (hours/24 h)	6.4 ± 1.2	6.1 ± 1.5	− 0.177	0.0011
Pittsburgh Sleep Quality Index (21-point scale)	5.9 ± 4.0	8.5 ± 4.4	0.416	< 0.001
Berlin Questionnaire (% high risk for sleep apnea)	63/152 (41)%	72/120 (60%)		0.003^†
Epworth Sleepiness Scale (24-point scale)	7.8 ± 4.8	10.4 ± 5.0	0.316	< 0.001
Fatigue scale (10-point scale)	3.4 ± 2.4	5.4 ± 2.2	0.484	< 0.001
hsCRP (mg/dl)	0.356	0.384	0.109	0.045

Notes: Values are mean+SD or proportion. Statistical comparisons are between low-stress and high-stress subjects using the two sample t-test (or Fisher exact test as noted) with p values less than 0.05 representing statistical significance. Correlation coefficients are derived from the Spearman's rank correlation coefficient (also called Spearman's rho) using a two-tailed test with p < 0.05 as the predetermined threshold of statistical significance. BMI, body mass index; hsCRP, high-sensitivity C-reactive protein

^* denotes Spearman's rho (r_s)

^† denotes Fisher's exact test.

Figure 1.  Using the Spearman correlation, there is a significant positive relationship between perceived stress scores (PSS) and scores on the PSQI, r_s = 0.43, n = 274, p < 0.0005.

Several measurements (n = 350), not presented in the tables, showed no correlation with levels of perceived stress by Spearman's rank correlation. The lipid panel including total serum concentrations of cholesterol (r_s = 0.04), LDL cholesterol (r_s = 0.004), HDL cholesterol (r_s = 0.015), triglyceride (r_s = − 0.045), and Lp (a) (r_s = − 0.013) did not correlate with PSS. Likewise, parameters of glucose metabolism did not correlate with PSS, including fasting plasma glucose concentration (r_s = 0.057), HgA1C percentage (r_s = 0.013), and the homeostatic assessment model or HOMA (r_s = 0.093).

When sorted by gender, important differences were revealed. By t-tests (n = 350), women were slightly younger (54 ± 11 years vs. 58 ± 12 years, F = 1.63, df = 348, p = 0.04), had higher perceived stress scores (PSS = 24 ± 8 vs. 20 ± 8, F = 0.44, df = 348, p = 0.04), had higher total serum cholesterol concentration (200 ± 37 vs. 176 ± 72 mg/dl, F = 0.58, df = 343, p = 0.009), and higher serum HDL cholesterol concentration (64 ± 22 vs. 48 ± 12 mg/dl, F = 15.21, df = 343, p < 0.001).

Discussion

The salient findings of this study are that increased levels of perceived stress were correlated with shortened total sleep time, worse scores for sleep quality, higher likelihood of sleep apnea, and worse daytime symptoms of sleepiness and fatigue. It is important to note that there were no concomitant correlations between perceived stress and lipid abnormalities or measures of glucose metabolism, two common risk factors for heart disease. It is known that normal values for lipids and glucose metabolism do not preclude an increased CVD risk. The finding that glucose and lipids did not correlate with stress in our study places greater weight on the role of sleep disruption in the development of CVD. In combination with numerous prior studies that connect short sleep and disturbed sleep with CVD (Heslop et al. 2002a,b), our correlations provide a mechanistic link to support the observed association between stress and CVD.

It is important to define stress and what is actually being measured with the PSS as it pertains to the current investigation. Because the PSS questions are general and free of content specificity, the instrument assesses subjectively experienced stress independent of an objective external stimulus or situation (Cohen et al. 1983). Personality aspects and resources of the subjects contribute to the total perceived stress score. The PSS correlates closely with trait neuroticism rather than the state of stress imposed. It therefore follows that trait neuroticism may be a pre-morbid characteristic of some good sleepers, who nonetheless manifest hyperarousal in response to stress and thus develop stress-induced insomnia (Basta et al. 2007; Fernandez-Mendoza et al. 2010).

The tools used to measure sleep in this study evaluate both sleep quality and sleep quantity. The high-stress group got an average of 20 min less sleep per night compared to the low-stress group. This may initially appear to be an inconsequential difference in sleep quantity. However, after only a few days or weeks, a substantial sleep debt can accrue, sufficient to affect mood, performance, and sense of well-being (Dinges et al. 1997; Drake et al. 2001). Furthermore, fatigue-inducing pro-inflammatory cytokines (interleukin-6 and tumor necrosis factor alpha) are negatively influenced by the quantity and quality of sleep (Vgontzas et al. 1999). CVD is a disease state stimulated and exacerbated by systemic inflammation. Prior research has also shown that insomnia with objective short sleep duration is associated with a higher risk for hypertension (Vgontzas et al. 2009a,b) and for type 2 diabetes mellitus (Vgontzas et al. 2009a,b), both major risk factors for CVD.

The Berlin Questionnaire focuses on an aspect of sleep quality. It is a validated instrument to quantify high vs. low risk for sleep apnea. The high-stress group with substantially higher BMI also has much higher odds of having sleep apnea. This finding is consistent with prior research that correlates increasing BMI with higher risk for sleep apnea (Newman et al. 2005). Explanations of these associations may include alternative theoretical models. Stress may stimulate maladaptive eating, leading to weight gain and subsequent development of sleep apnea. Alternatively, sleep apnea may disrupt the restorative functions of sleep (experienced as higher stress levels) and simultaneously disrupt hormonal regulation of hunger leading to greater calorie consumption and weight gain. These pathways toward greater risk of CVD warrant corrective attention at a time early in the cycle to preclude a downward spiral of health indicators.

Worse sleep quality as measured by the PSQI correlated with higher stress levels (Figure 1). Similarly, the ESS and fatigue scale, consequences of the impact of poor sleep quality, correlated with higher stress levels. The finding that different tools showed worse scores with higher stress levels gives credibility to the observation linking poor sleep quality with high stress. Of course the challenge will be finding effective ways to improve sleep quality and consequent daytime symptoms, translating to improvements in CVD risk.

A novel aspect of our research is the use of the PSS and PSQI as tools to measure stress and sleep quality. There are few other studies that link perceived stress with poor sleep quality. There is one publication that utilized both the PSS and the PSQI in the same study (Strange et al. 2009). These coauthors investigated 220 pregnant women and found that PSS did not predict preterm birth and that preterm births were associated with lower daytime dysfunction scores on the PSQI. A PSS–PSQI connection was not reported in the study.

CVD is the leading cause of death in women, despite the cardiovascular protection afforded by their endogenous hormones and increased levels of HDL cholesterol (Wasserthiel-Smoller 2010). In our study, women were found to have significantly higher perceived stress scores than men. This finding may indicate that stress levels, specifically the measured PSS score, may be an important gender-relevant risk factor to survey, especially as a preventive strategy for improving women's health.

What cannot be determined in a cross-sectional study is causality. It cannot be inferred whether or not perceived stress causes deranged sleep or if poor sleep habits cause increases in perceived stress. It is possible that both perceived stress and sleep habits are worsened by another stimulus and that they respond in parallel to that stimulus. The relationship of perceived stress and disturbed sleep deserves further clarification, perhaps with a study providing an intervention aimed at stress or at sleep alone.

One limitation of the current study is that several indices were measured using subjective self-reports. Self-reported data included perceived stress levels, sleep quality, daytime sleepiness, and fatigue. However, the tools utilized to gather these indices were validated instruments with known performance characteristics and some of the data sought have no alternative ways of being measured. It may be useful in future studies to utilize objective measures such as a polysomnogram instead of the Berlin Questionnaire for sleep apnea and actigraphy as an objective measure of sleep quantity. Furthermore, strength of the current study is that actual measurements of height, weight, and waist circumference were used in place of self-reported values.

Our finding of correlation of perceived stress levels with sleep disruption adds to the growing body of evidence that stress may play an important role as a risk factor for CVD. Certainly the evidence to date is worthy of follow-up studies. A justifiable next study could examine the impact of stress management strategies and sleep improvement on incident CVD. Assessing maladaptive behaviors and physiological abnormalities associated with stress may allow for targeted intervention to promote vascular health.

Acknowledgments

This study was supported by the Henry M. Jackson Foundation for the Advancement of Military Medicine.

Declaration of interest: The authors report no conflicts of interest. The authors alone are responsible for the content and writing of the paper.