Participants were police officers examined during the first follow-up of the Buffalo Cardio-Metabolic Occupational Police Stress (BCOPS) longitudinal study. The objectives and design of the BCOPS study have been described previously.²² Briefly, the BCOPS study aimed to examine associations of psychosocial measures, lifestyles, and physiological biomarkers of stress with subclinical metabolic and vascular disease markers among police officers. The first BCOPS follow-up study was conducted among 281 police officers and all examinations were performed between 2011 and 2014 at the State University of New York at Buffalo in Buffalo, New York, USA. Retinal examination was performed during the first follow-up period.^23,24 Among the 281 participants, 46 officers were excluded due to missing information on retinal examination (19 officers were recruited after the completion of the retinal examination and 27 declined the retinal examination). Further exclusions included 29 officers who had missing information on any of the sleep parameters, one officer who had missing information on retinal arteriolar measurement, and three officers who had influential sleep duration (<3 hours), resulting in a final sample of 202 officers for analyses. The tenets of the Declaration of Helsinki were followed, and the study was approved by the State University of New York at Buffalo Institutional Review Board and the National Institute for Occupational Safety and Health Human Subjects Review Board. All participants from the follow-up study signed the consent form.

Objectively measured sleep duration and sleep quality were derived from actigraphy data. Each participant was instructed to wear a sleep watch on his/her wrist of the non-dominant hand for 15 days and completed a sleep diary for the same period, which was used for the actigraphy data editing. The actigraphy data were collected using the Micro Motion Logger Sleep Watch™ (Ambulatory Monitoring Inc., NY) and were analysed using the Action-W software with Cole Kripke algorithm for sleep scoring. Numerous parameters that describe sleep quantity and quality were then derived. In the present analyses, we decided to use total sleep time, sleep efficiency, sleep onset latency, WASO, and number of awakenings as independent variables for they have been shown to be associated with health outcomes and wellness among general population.^20,21 The definition of each sleep parameter was consistent with previous studies.^20,21 In short, total sleep time was the number of hours slept during time in bed and did not include napping during work. Sleep efficiency was the percentage of time spent asleep during the time spent in bed, and sleep onset latency was the time it took a participant to fall asleep after getting to bed. To mark time in bed, each participant was instructed to press the event marker button of the actigraph when first lying down in bed each night and when getting out of bed at the end of a sleep period. In addition, participants were also instructed to record times the actigraph was off wrist (e.g., when taking shower). The participants also kept a sleep diary for each day which consisted of time in bed (in bed time and get up time) as well as start time of work and work hours. Data from sleep diary were used to cross-check time in bed information obtained from the actigraph. Sleep onset latency was then derived as the number of minutes from the time the subject first lay down in bed to the time the subject was first scored as asleep by the sleep-scoring algorithm.

WASO was the total duration of wakefulness after the first persistent sleep, i.e., 20 continuous minutes of sleep with no more than one-minute of wakefulness. The number of awakenings referred to the total number of awake episodes with each lasting at least 5 minutes, excluding the final wakening. In addition to the above sleep parameters, longest wake episode (i.e., the duration of the longest wake episode during time in bed) was used because we hypothesized that the longest wake episode would be a stronger indicator of poor sleep quality than WASO. The estimate for each sleep measure was the average across the 15 days the sleep watch was worn.

Retinal microvascular measurements were obtained from fundus photography. Retinal imaging was performed by research associates who were certified through trainings provided by the Ocular Epidemiology Reading Centre (OERC) at the University of Wisconsin-Madison. The CR-2 Digital Non-Mydriatic Retinal Camera (Canon Inc., Lake Success, NY, USA) equipped with a Canon EOS 60D (18 megapixels) camera back and the Canon Retinal Image Control Software was used for imaging. The imaging procedure followed a standardized written protocol. Before imaging, each participant was seated in a windowless room with the lights turned off to allow the pupils to dilate naturally at least 4 mm in diameter. After a 5-minute waiting period, if an eye did not dilate to the desired size, the small pupil feature was used in imaging the eye. Two 45° colour retinal images were taken per eye, with the right eye always first for each participant. One image was centered on the optic nerve (field 1) and the second on the macula (field 2). After images were taken for each participant, the photographer transmitted the quality images to the OERC reading centre. The OERC staff monitored the image quality and provided feedbacks for improvements. An image that had fewer than four acceptable measurements of either vessel type was considered ungradable.

All the fundus images were graded using the Interactive Vessel Analyzer (IVAN) (V1.0). The vessels coursing through the grid between 0.5 to 1.0-disc diameters (diameter of 1800 μm) from the disc margin were manually identified as arterioles or venules. The IVAN takes the six largest arterioles and the six largest venules from the disc nerve centred image to calculate the CRAE and CRVE using the Parr-Hubbard-Knudtson’s²⁵ revised formula: CRAE=0.88∗w12+w221/2 and CRVE=0.95∗w12+w221/2, in which w₁ is the diameter of a larger arteriole or venule, and w₂ is the diameter of a smaller arteriole or venule. The iteration procedure of pairing up the six arterioles or the six venules has been described in detail by Knudtson et al.²⁵ The image scale factor for the present study is 4500 μm. If one of the six largest vessels was ungradable, either CRAE or CRVE was not computed.

The potential confounders were selected through the following steps. First, a list of potential confounders was selected from the literature. Then we removed any variables that were known to be in the causal pathway between poor sleep quality and CRAE or CRVE from the list. The final list of potential confounders included age, sex, race/ethnicity, body mass index (BMI), physical activity, alcohol consumption, smoking status, hypertension, sleep apnoea, police stress, and shift work. Sex and race/ethnicity were also selected as potential effect modifiers in the associations of sleep parameters with CRAE and CRVE based on these three questions: 1) is there a difference in the prevalence of the outcome between groups? 2) is there a difference in the prevalence of the exposure between groups? and 3) does the relationship between the exposure and outcome differ between groups?²⁶

Information on the identified confounders and effect modifiers were objectively measured during the interview, obtained from self-reported and interviewer-administered questionnaires, or obtained from payroll records. Weight, height, and blood pressure were measured following the standardized protocol described previously.²⁷ BMI was computed using the standard formula. Systolic blood pressure (SBP) and diastolic blood pressure (DBP) were measured in seated position for each participant who had rested for 5 minutes.²⁴ Information on antihypertension medications was collected through a questionnaire that asked all the prescribed and over-the-counter medications taken in the past 30 days as of the examination. Each participant’s hypertension status was derived and was categorized as controlled hypertension (i.e., having SBP < 140 mmHg and DBP < 90 mmHg and taking antihypertensive medications), uncontrolled hypertension (with SBP ≥ 140 mmHg or DBP ≥ 90 mmHg regardless of medication use), or normotensive (i.e., having SBP < 140 mmHg and DBP < 90 mmHg and free from antihypertensive medications) as described in the previous BCOPS study.²⁴ The measurement of fasting serum glucose,²⁷ glycated hemoglobin (HbA1 c), white blood cell count,²⁸ and C-reactive protein (CRP)²⁹ have been described previously.

Physical activity referred to the hours spent on sports, work, and household-related activities in the previous seven days and were collected using the 7-Day Recall questionnaire,³⁰ replacing weekdays by days worked and weekends by days off duty to integrate the participants’ shift work characteristic. Total hours of physical activity were calculated for each participant by summing the time spent on moderate, hard, and very hard activity during the seven days. Alcohol intake per week was derived from the information collected on usual food choice and usual food use through a food frequency questionnaire. One drink was defined as a 12-ounce can or bottle of beer, one medium glass of wine, or one shot of liquor. Smoking status was self-reported as current, former, or never.

Information on sleep apnoea was obtained using the Survey Screen for Prediction of Apnoea questionnaire³¹ asking the participants three questions 1) “Do you know or has someone told you that you snort or gasp while sleeping?”; 2) “Do you know or has someone told that you snore loudly while sleeping?”; and 3) “Do you know or has someone told you that breathing stops or you choke or struggle for breath while sleeping?”. Each question was rated as: never, rarely, 1–2 times a week, 3–4 times a week, and 5–7 times a week. Symptoms frequency index was computed by averaging the no missing values for the responses to the three questions. In the present analysis, we categorized the frequency index into three levels as 1) no symptom, if the index score was equal to zero; 2) moderate, if the score was between zero and two; and 3) severe, if the index was equal to or greater than two.

Police stress was assessed using the 60-item (stressor) Spielberger Police Stress Survey.³² The means of how to derive stress index has been described previously.³³ Briefly, each participant reported the frequency of each stressor occurring in the past year and provided a rating score for the intensity of the stressor on the scale from 0–100. Then, a total stress index was calculated for each participant by multiplying the frequencies of each stressor in past year by the rating score and then summing these products over all 60 items. A higher score indicated a higher stress level.

The details of shift work derivation can be found in a previous BCOPS publication.³⁴ Briefly, the electronic work history records provided detail information on work schedule, start, and end time on the daily bases. A 10-hour permanent non-rotating shift work system was implemented by the police department since 1994. Day (if started between 4:00 AM to 11:59 AM), afternoon (if started between 12:00 PM to 7:59 PM) or midnight shift (if started between 8:00 PM to 3:59 AM) was derived for each participant from the work history data.

Diabetes status was defined as yes if meeting any of the following conditions 1) fasting plasma glucose ≥ 7.0 mmol/L or non-fasting glucose ≥11.0 mmol/L or HbA1 c ≥ 7.5%; and 2) taking antidiabetic medications.³⁵

Descriptive statistics for the identified potential confounders, effect modifiers, and sleep parameters were compared by racial/ethnic background and sex. Age-adjusted associations of each potential confounder and effect modifier with CRAE and CRVE were examined using partial Pearson’s correlation coefficient. Unadjusted and multivariable-adjusted associations of each standardized sleep parameter (i.e., sleep duration, sleep efficiency, sleep onset latency, WASO, number of awakenings, and longest wake episode) with CRAE and CRVE were examined using simple and multiple linear regression model controlling for age, sex, race/ethnicity, BMI, physical activity, alcohol consumption, smoking status, hypertension status, apnoea symptoms frequency index, log-transformed police stress, and shift work.

Effect modifications of sex and race/ethnicity on the associations of each sleep variable with CRAE and CRVE were tested using stratified analysis by constructing simple and multiple regression models which included interaction terms. In the multiple regression model, adjustments for the effect modification of sex were age, race/ethnicity, BMI, physical activity, alcohol consumption, smoking status, hypertension status, apnoea symptoms frequency index, log-transformed police stress, and shift work. While the adjustments for the effect modification of race/ethnicity were the above potential cofounders, except that race/ethnicity was replaced by sex. The type I error rate was set at 5% for the associations of sleep parameters with CRAE and CRVE, and 20% for the interaction tests to account for reduced power of testing interaction terms. All analyses were performed using the SAS software, Version 9.4 (SAS Institute, Cary NC).

In the present study, sex modified the associations of sleep onset latency and longest wake episode with CRVE; and the associations were significant only among men. These results were consistent with that from the Multi-Ethnic Study of Atherosclerosis¹⁵ reporting a significant association between obstructive sleep apnoea and wider CRVE among men but not among women. The significantly higher level of glucose and higher prevalence of diabetes might explain the gender differences in the associations. Another possible factor contributing to the gender difference may be the protective effect of oestrogen among female officers³⁶ who are mostly under the age of menopause or on oestrogen-related medications. Finally, the small sample size of female officers might reduce the statistical power of testing a potential association.

The present result showed that the positive association between longest wake episode and CRVE was significant among whites but was not evident among African Americans. We need to be cautious when interpreting these results. First, the small sample of African Americans might reduce the power of detecting a potential association. Second, the significant effect modification of race/ethnicity might be biased by residual confounding. Therefore, future epidemiological studies with a larger sample size of African Americans and longitudinal studies are needed.

The unique finding in the present study is that compared to the WASO, the longest wake episode had a stronger association with CRVE. The mean time of WASO includes all wake episodes, some of which may not be caused by disordered sleep but rather are due to family needs such as waking up to take care of dependents. However, the longest wake episode may reflect the true inability to return to sleep.

We were not surprised by the result that none of the sleep parameters were associated with CRAE in the present study, for previous studies have found that CVD risk factors have different impact on retinal arterioles and venules. For example, hypertension is associated with narrower CRAE and inflammatory markers are associated with wider CRVE.³⁷

The mechanisms underlying the relationship between poorer sleep quality and wider CRVE are not yet well explored. The most common explanation is that normal sleep is manifested by a cyclic alternation between non-rapid eye movement (NREM) and rapid eye movement (REM) sleep lasting 7–9 hours. REM is associated with activated sympathetic nerve activity³⁸ and NREM sleep is characterized by parasympathetic predominance.³⁹ During sleep, various physiological changes vary across stages of NREM and REM sleep.³⁹ Chronic disruptions of the sleep stages, such as insomnia, is shown to be a state of hyperarousal due to abnormal modulation of the autonomic nervous system.⁴⁰ The increased sympathetic nervous system activity has major effects on systemic conditions such as impaired glucose metabolism^41,42 and increased level of systemic inflammation⁵ which may contribute to damages of endothelium function.

Biological studies have demonstrated that the retina and brain have the highest density of pericytes in the body.⁴³ Increased blood glucose concentrations could contribute to pericyte loss.⁴⁴ When blood vessels lose pericytes, they will become dilated.⁴³ In the exploratory analysis, we also observed that the levels of white blood cell count and CRP were positively associated with the longest wake episode and were significantly higher among men and white officers, which might explain the modification role of sex and race/ethnicity. However, the challenge of this explanation is that poor sleep was not associated with CRAE although pericytes are abundant both in retinal arterioles and venules. A study has shown the vasodilation in the porcine retina ex vivo after exposure to glucose concentration,⁴⁵ but it is not clear in human retina in vivo.

Several limitations need to be considered when interpreting the findings from the present study. First, the cross-sectional study design precluded us from making any causal inferences. Second, our study participants were relatively young. Retinal abnormalities (CRAE narrowing and CRVE widening) are more common in older people.^46,47 Therefore, our findings may not be generalizable to older populations. Second, the small sample size of female officers or African Americans might reduce the statistical power to detect potential associations among the two groups. Third, unmeasured confounding could contribute to the observed associations or could have even masked the true association. Despite the limitations, the present study tested the relationship of several sleep parameters from actigraphy data with retinal vascular diameters which extended previous research that was limited to sleep apnoea and retinal microvasculature. Another strength was the increased reliability and reduced measurement error using an average of 15-day of actigraphy monitoring that captured the sleep variations over a 15-day of shift work cycle.

In conclusion, the present cross-sectional study found that sleep quality measured by actigraphy, indicated by sleep onset latency, was positively associated with CRVE among male officers, and the longest wake episode was positively associated with retinal CRVE among male or white police officers. These associations were independent of the selected potential confounders. None of the sleep parameters were associated with CRAE. Future longitudinal studies are warranted to examine the potential causal link between poor sleep quality and retinal microvasculature change.