Data Source and study population
This cross-sectional study used data from the 2015–2016 and 2017–2018 cycles of the National Health and Nutrition Examination Survey (NHANES). NHANES employs complex multistage sampling procedures to survey the civilian, noninstitutionalized USA population. The participants are interviewed at home, followed by physical examinations and further interviews at a mobile examination center (MEC)22. The NHANES protocol was approved by the National Center for Health Statistics Ethics Review Board23. Informed consents were obtained for all eligible subjects before they participated in the NHANES interviews and physical examinations. All methods were performed in accordance with relevant guidelines and regulations of the NHANES. We used de-identified and publicly available data that did not meet human subjects research criteria and thus was exempt from institutional review board oversight. The datasets used are available at https://wwwn.cdc.gov/nchs/nhanes/.
Figure 1 illustrates the criteria used for identifying the study sample. Of the 19,225 adults (≥ 18 years old) in the 2015–2018 NHANES24, 5,712 participated in interviews and physical examinations and met at least one of the following criteria for hypertension: current use of BP-lowering medication (n = 3,706), average SBP ≥ 130mmHg (n = 3,744), or SBP ≥ 80mmHg (n = 2,323)2. We excluded participants who were pregnant or missing sleep duration or BP data. To minimize the potential for unmeasured confounding25, we excluded those with severe disease, including congestive heart failure and severe (stage 4–5) chronic kidney disease (defined as a history of undergoing dialysis in the previous 12 months or a Chronic Kidney Disease Epidemiology Collaboration estimated glomerular filtration rate < 30 mL/min/1.73 m2)26, 27,28.
Habitual Sleep Duration. The NHANES calculated the amount of sleep usually obtained in a night or main sleep period during weekdays or workdays from two survey questions: “What time do you usually fall asleep on weekdays or workdays?” and “What time do you usually wake up on weekdays or workdays?” We categorized habitual sleep duration into < 6, 6 – <7, 7–9, and > 9 hours/night or main sleep period5,11,12.
Hypertension Control. A standardized protocol was used to obtain three BP readings taken one minute apart at the MEC29. Hypertension control was defined as average SBP < 130mmHg and DBP < 80mmHg2.
Confounders. Sociodemographic, sleep, and health-related factors associated with sleep duration and hypertension control were identified as potential confounders9,11,12,30−35. The factors include gender (male, female), age (< 18–39, 40–59, 60–79, ≥ 80 years), nativity (US.-born, not US.-born), education level (less than high school, high school graduate, some college, college graduate), employment status (parttime < 35 hours, fulltime 35–44 hours, fulltime ≥ 45 hours, not working), health insurance (insured, uninsured), and annual household income (<$55,000, ≥$55,000). The $55,000 NHANES income category was used as a cut-point because it is close to the US median household income for the 2015 to 2017 period36. Race/ethnicity was self-reported and classified as non-Hispanic White, non-Hispanic Black, Hispanic, and Other.
History of sleep apnea symptoms was defined as a history of; (1) snoring ≥ 3 times/week; (2) snorting, gasping, or stopping breathing while sleeping ≥ 3 times/week; or (3) being excessively sleepy during the day ≥ 16 times/month despite sleeping for ≥ 7 hours per night37,38. Help-seeking for sleeping difficulty was defined as a history of ever telling a health care professional that one had trouble sleeping. Depressive symptoms were screened using the 9-item Patient Health Questionnaire (PHQ-9, range of 0–27) and categorized as minimal or none (0–4), mild (5–9), moderate to severe (≥ 10–14)39.
The number of healthcare visits in the past 12 months (excluding home visits, phone consultations, overnight hospitalization, and emergency room visits) was grouped into none, 1–2, and > 2 visits. Body mass index (BMI) was analyzed as a continuous variable. Cardiovascular disease was defined as a history of being told by a health professional that one had coronary heart disease, angina, a heart attack, or stroke. Diabetes was defined based on either a history of being told by a health care professional that one has diabetes or having a blood glycohemoglobin level of 6.5% or higher40. Moderate chronic kidney disease was defined as an eGFR of 30 - <60 mL/min/1.73 m2 26.
Cigarette smoking was categorized as never smoker (never smoked at least 100 cigarettes in their lifetime), former smoker (smoked at least 100 cigarettes in their lifetime but not smoking currently), and current smoker. Physical activity was self-reported and included leisure, work, and transportation (commuting by walking or bicycling) activities. Moderate-intensity and transportation-related physical activities were assigned four metabolic equivalents of task (MET) scores/minute, and vigorous-intensity physical activity eight MET scores/minute41. Weekly physical activity levels were categorized as none (0 MET-minutes), low (< 600 MET-minutes), sufficient (600–1200 MET-minutes), and high (> 1200 MET-minutes)42. Alcohol intake was classified as none (never had at least 12 alcoholic drinks in a lifetime or any alcohol in the past year), moderate (not more than one drink for women and not more than two drinks for men in a day), and heavy (more than one or two drinks in a day for women and men, respectively)43.
Statistical Analysis & Missing Data
Data distribution across various variables was analyzed for the total study sample and across habitual sleep duration categories. The percentage of observations with complete data for all covariates was 85%. The covariates with missing data (unweighted number and weighted percentage) included education level (n = 6, 0.1%), annual household income (n = 413, 6.0%), health insurance (n = 10, 0.2%), alcohol intake (n = 333, 5.3%), BMI (n = 67, 1.0%), depressive symptoms (n = 357, 5.6%), and healthcare visits in past year (n = 10, 0.1%). Missing data were imputed using multiple imputation with chained equations 44. A total of 20 imputation datasets were generated, and the results pooled to generate estimates of the multiply impute model using STATA IC’s multiple imputation (MI) estimate procedures 45.
The variables used in the imputation model included all the variables in the analysis model, the NHANES cluster, strata, and weights variables, and auxiliary variables significantly associated with missing data in some covariates (age and education level of the reference person, homeownership, number of rooms in house of residence, and household size).
We compared the observed data to the complete imputed data to check for differences in data distribution in the covariates with missing data. The distribution remained similar for depressive symptoms, BMI, healthcare visits in the past year, education level, health insurance, and alcohol intake. In the annual household income, the proportion of the < $55,000 group increased from 48.5–49.3% while the ≥ $55,000 group reduced from 51.5–50.7% after imputation (Appendix Table 1).
Models to analyze the association between habitual sleep duration and hypertension control were fit using the complete imputed data. Logistic regression models (unadjusted and adjusted) were fit to analyze the association between habitual sleep duration and hypertension control, with the crude and adjusted odds ratios for hypertension control and their corresponding 95% confidence intervals presented. Effect modification by age and gender was assessed separately by adding an interaction term (sleep duration × age or sleep duration × gender) to the adjusted logistic regression model. All data were analyzed using STATA 1C software (Version 15, StataCorp LLC, College Station, Texas, 2017). Survey commands were used to apply sample weights to account for the NHANES complex sampling design. The level of significance for all analyses was set at a p-value < 0.05.