Association between sleep and blood pressure in midlife: the CARDIA sleep study

Abstract

Background: Epidemiological studies have reported an association between self-reported short sleep duration and high blood pressure (BP). Our objective was to examine both cross-sectional and longitudinal associations between objectively measured sleep and BP.

Methods: This study is ancillary to the Coronary Artery Risk Development in Young Adults (CARDIA) cohort study. Blood pressure was measured in 2000 and 2001 and in 2005 and 2006. Sleep was measured twice using wrist actigraphy for 3 consecutive days between 2003 and 2005. Sleep duration and sleep maintenance (a component of sleep quality) were calculated. Analyses included 578 African Americans and whites aged 33 to 45 years at baseline. Outcome measures were systolic BP (SBP) and diastolic BP (DBP) levels, 5-year change in BP, and incident hypertension.

Results: After we excluded the patients who were taking antihypertensive medications and adjusted for age, race, and sex, shorter sleep duration and lower sleep maintenance predicted significantly higher SBP and DBP levels cross-sectionally as well as more adverse changes in SBP and DBP levels over 5 years (all P < .05). Short sleep duration also predicted significantly increased odds of incident hypertension (odds ratio, 1.37; 95% confidence interval, 1.05-1.78). Adjustment for 16 additional covariates, including snoring and daytime sleepiness, slightly attenuated the associations between sleep and BP. Sleep duration appeared to mediate the difference between African Americans and whites in DBP change over time (P = .02).

Conclusion: Reduced sleep duration and consolidation predicted higher BP levels and adverse changes in BP, suggesting the need for studies to investigate whether interventions to optimize sleep may reduce BP.

Figure 1

Distribution of participants in sleep duration and sleep maintenance categories (n=578).

Figure 2

Predicted 5-year change in SBP and DBP by sleep duration and sleep maintenance from regression analyses adjusting for age and race-sex and excluding those taking anti-hypertensive medication at either baseline or follow-up (n=505). Error bars represent the standard error of the predicted change (the beta coefficient).