Skip to main content
Browse Subject Areas

Click through the PLOS taxonomy to find articles in your field.

For more information about PLOS Subject Areas, click here.

  • Loading metrics

Impact of sleep quality and physical activity on blood pressure variability


Increased blood pressure variability (BPV) is linked to cardiovascular disease and mortality, yet few modifiable BPV risk factors are known. We aimed to assess the relationship between sleep quality and activity level on longitudinal BPV in a cohort of community-dwelling adults (age ≥18) from 17 countries. Using Withings home measurement devices, we examined sleep quality and physical activity over one year, operationalized as mean daily step count and number of sleep interruptions, both transformed into tertiles. The primary study outcome was high BPV, defined as the top tertile of systolic blood pressure standard deviation. Our cohort comprised 29,375 individuals (mean age = 58.6 years) with 127.8±90.1 mean days of measurements. After adjusting for age, gender, country, body mass index, measurement days, mean blood pressure, and total time in bed, the odds ratio of having high BPV for those in the top tertile of sleep interruptions (poor sleep) was 1.37 (95% CI, 1.28–1.47) and 1.44 (95% CI, 1.35–1.54) for those in the lowest tertile of step count (physically inactive). Combining these exposures revealed a significant excess relative risk of 0.20 (95% CI, 0.04–0.35, p = 0.012), confirming their super-additive effect. Comparing individuals with the worst exposure status (lowest step count and highest sleep interruptions, n = 2,690) to those with the most optimal status (highest step count and lowest sleep interruptions, n = 3,531) yielded an odds ratio of 2.01 (95% CI, 1.80–2.25) for high BPV. Our findings demonstrate that poor sleep quality and physical inactivity are associated with increased BPV both independently and super-additively.


Increased systolic blood pressure variability (BPV) is associated with adverse effects on multiple organ systems [1, 2]. Prior research has demonstrated that patients with increased BPV have a higher risk of all-cause mortality in disease states ranging from coronary artery disease to diabetes [3, 4]. However, no therapeutic intervention has been proven to lower BPV. There is preliminary data that sleep quality and physical activity level influence BPV. Increased BPV has been associated with poor sleep quality, primarily sleep interruptions and obstructive sleep apnea [57]. The association between physical activity and lower BPV has also been demonstrated using both short-term (hours) and long-term (days, weeks or months) blood pressure data [810]. Prior research has not examined a potential additive or super-additive effect of sleep quality and physical activity on BPV. Using a large multi-national dataset, we hypothesized that better sleep quality and higher levels of physical activity would have a super-additive benefit on BPV.



We obtained a dataset from the Withings corporation (Withings, Inc; Issy-les-Moulineaux, France) [11], which makes health and fitness consumer electronics. Their blood pressure device uses conventional arm-cuff methodology and their actigraphy watch measures all physical activity and converts it to a daily step count [12]. The sleep data is collected using a pneumatic mat under the mattress. Its measurement of sleep interruptions has been previously validated against polysomnography [13, 14].

Our cohort comprises community-dwelling adults (age ≥18) who voluntarily recorded blood pressure, sleep, and step count using devices seen in Fig 1. Subjects were not consented for this research but consented to deidentified data sharing when agreeing to Withings’ terms of service. This limited and de-identified dataset was not subject to IRB regulation. We adhered to the STROBE checklist for observational cohort research.

Fig 1. A sample of Withings devices used to acquire physiologic data.

We included individuals who recorded data on all three devices for at least 40 days during the first year after recording started. This inclusion criterion was based on studies showing that the accuracy of BPV measurement stabilizes after 40 blood pressures [15, 16]. We performed two sensitivity analyses in which we included individuals with data on >2 days and >179 days. Systolic blood pressures <50 and >300 mm Hg were considered non-physiologic and converted to missing, as were step counts of 0 or >50,000 and sleep interruptions of >20. Overall, this led to less than 1% loss of data.

Study outcome, exposure, and covariates

The primary study outcome was BPV. We calculated the standard deviation of daily systolic blood pressure to represent BPV, but as a sensitivity analysis we replicated the primary analysis with two additional BPV methodologies: coefficient of variation and absolute real variability. The main study exposures were step count and sleep interruptions, which is a proxy for poor sleep quality.

Withings-derived covariates included country of use, self-reported age, gender and body mass index, which could either be recorded on a Withings scale or calculated using self-reported data. Study derived covariates included total days of measurement, mean systolic blood pressure, and total sleep duration. To improve model fit, we transformed BPV, sleep interruptions, step count, age, and body mass index into tertiles due to right skew (Fig 2), which is common with physiologic and health science data [1720].

Statistical methods

We report descriptive statistics for the entire cohort and those with high BPV versus those without. We then fit a logistic regression to the top tertile of standard deviation, which we call “high BPV”. This model’s goodness of fit was confirmed using the Hosmer-Lemeshow test, and the absence of multicollinearity was confirmed by a variance inflation factor of less than 10. We initially fit a linear regression model but, despite transforming the independent and dependent variables, it did not meet the assumptions of linear regression.

After multivariable logistic regression, we report odds ratios and to determine if the exposures were super-additive, we derived the RERI (relative excess risk due to interaction) [21]. We then modeled the exposures as an interaction (steps*interruptions) and used marginal effects to derive the predicted probability of having high BPV for optimal exposure status (highest tertile of step count and lowest of sleep interruptions). We also show the probability of high BPV for a model with continuous values of the exposures in a 3D plot.


The Withings dataset contained 96,826 individuals, of which 915 were excluded for having no data or non-physiologic data and 66,536 were excluded for not having data on at least 40 days during the year. The remaining 29,375 individuals comprised our primary cohort, which on average had an age of 58.6±12.0 years, measurements on 127.8±90.1 days, systolic blood pressure of 126.5±10.9 mm Hg, daily step count of 5,818±3,370, sleep interruptions of 2.3±1.2, standard deviation of 9.4±3.1, was 82.5% male, and was from 17 different countries. Additional demographics are seen in Table 1.

Table 1. Baseline demographics and outcomes in the cohort and after stratification by the highest tertile of BPV versus the others.

Fig 2 shows the distribution of the study’s exposures and outcomes, as well as the right skew that necessitated transforming them into tertiles for accurate modeling. The proportion of individuals in the tertile strata of sleep interruption and step count was well balanced (Table 2), but there is a significantly higher proportion of individuals in the highest tertile of sleep interruptions for those in the lowest (vs. highest) tertile of step count (13.1% vs. 10.1%, p<0.001).

Table 2. Distribution of tertiles of activity (step count) and sleep quality (interruptions).

In the adjusted logistic model, the odds ratio of having high BPV for those in the top tertile of sleep interruptions (poor sleep) was 1.37 (95% CI, 1.28–1.47), and for those in the lowest tertile of step count (physically inactive) was 1.44 (95% CI, 1.35–1.54). In the RERI model, the excess relative risk was significant at 0.20 (95% CI, 0.04–0.35, p = 0.012), which confirms that the exposures are super-additive. Comparing those with the optimal exposure status (highest tertile of step count and lowest of sleep interruptions, n = 3,531) to those with the worst exposure status (lowest tertile of step count and highest of sleep interruptions, n = 2,690), the odds ratio for being in the top tertile of BPV was 2.01 (95% CI, 1.80–2.25).

Near identical results were seen for the alternate measures of BPV, which is seen in Table 3. When we include individuals with >2 days of data, the cohort is larger (n = 90,540) and there is a slight reduction in the effect size, but all elements of the analysis remain the same, while in individuals with >179 days of data (n = 7,077) there is an increase in the effect size (Table 4).

Table 3. Multiple methodologies to measure high BPV, including standard deviation, coefficient of variation, and absolute real variability.

Table 4. Effect of changing inclusion criteria to include more or less individuals based off number of days with available data.

The predicted probabilities for high BPV are seen in Fig 3. For those with optimal exposure status the probability of having high BPV is 26.9% (95% CI, 25.4–28.4) while for those with the worst exposure status the probability is 40.6% (95% CI, 39.1–42.1) (p<0.001 for difference). The super-additivity is even more apparent when the exposures are treated as continuous variables (Fig 4).

Fig 3. Predicted probability of being in the highest tertile of BPV by tertiles of step count and sleep interruptions.

* Adjusted for age, gender, country, body mass index, days of measurements, mean blood pressure, and total time spent in bed.

Fig 4. Predicted probability of being in the highest tertile of BPV by continuous values of step count and sleep interruptions.

* Adjusted for age, gender, country, body mass index, days of measurements, mean blood pressure, and total time spent in bed.


In an international cohort of community-dwelling adults who purchased and used home devices to measure blood pressure, sleep quality, and physical activity, we show that poor sleep quality and physical inactivity are associated with increased BPV both independently and super-additively. This novel finding has not been previously demonstrated because of the lack of datasets with simultaneous home monitoring of blood pressure, sleep quality, and physical activity. Given their novel nature, the findings warrants further study in more generalizable cohorts with additional demographic data and information on medical comorbidities and medications [22]. Despite the possibility of unmeasured confounding, measurement bias, and selection bias in our analysis, the large cohort and frequent and concurrent measurements of the study outcome and exposures are strengths afforded by this unique dataset.

Our findings regarding sleep quality are consistent with prior research that examined 183 patients with obstructive sleep apnea who discontinued treatment and had a subsequent increase in BPV by 1.14 mm Hg [6] and a second study of 3,565 individuals undergoing overnight polysomnogram that found associations between BPV and obstructive sleep apnea severity as well as total sleep time/REM duration after adjusting for sleep apnea severity [5]. Likewise, prior research in small cohorts has demonstrated an association between physical activity and lower BPV, particularly in individuals who are physically fit [810]. However, what has been lacking from these prior studies is an examination of the possible synergistic effects of sleep quality and physical activity on BPV. This was an important knowledge gap because prior research has shown an independent connection between activity level and sleep quality [23, 24], which we also showed in our study (Table 2).

A major impediment to developing clinical trials focused on BPV has been that effective treatments to lower BPV are not established. Using a treatment effects simulation of a clinical trial, we previously showed that dihydropyridine calcium channel blockers reduce BPV ∼2 mm Hg [22]. The question of how much BPV would have to be reduced to improve outcomes is not known but it is possible that interventions would have to be combined. Our hypothesis-generating analysis shows that targeting activity and sleep quality may have the potential to lower BPV and could theoretically be used in addition to pharmaceutical approaches.


This study is subject to unmeasured confounding, a limitation that must be recognized due to its potential to result in incorrect interpretations of causal relationships. Unmeasured confounding also compromises the reliability of our findings and their applicability in guiding clinical practice. Consequently, the outcomes of this research should be viewed with caution and not be directly implemented to alter clinical practices or patient care strategies.

We used a convenience sample of primarily male individuals who voluntarily purchased the commercial Withings devices, used them consistently, and sent data to the Withings app. This introduces a selection bias, although the large international sample provides an element of diversity. Nonetheless, the results should not be considered generalizable because disadvantaged socioeconomic groups, older adults, and those with cognitive impairment struggle with implementing mobile health technology [25, 26]. In addition, home devices typically require a Bluetooth connection and internet or cellular transmission of data. More than 25% of Americans over the age of 65 do not have an internet connection [27] and there is a well known digital divide for underrepresented groups [28, 29].

There is also measurement bias. Although home measurement devices offer convenience and the potential for more frequent BP readings, they may differ in accuracy and reliability compared to clinical measurements. We lacked information on medical comorbidities, in particular sleep apnea and cardiovascular diseases. Finally, our study could not account for participants’ medication regimens (e.g. antihypertensive drugs), which could significantly influence the outcome and exposures. Future research should investigate these findings in groups who may not be well represented in convenience samples like this and collect more detailed demographic, medical, socioeconomic, and medication data.


In summary, our findings demonstrate that poor sleep quality and physical inactivity are associated with increased BPV both independently and super-additively in a convenience sample of primarily male individuals who voluntarily purchased commercial home measurement devices. Given these limitations, our findings need to be replicated in a more generalizable cohort.


  1. 1. Gosmanova EO, Mikkelsen MK, Molnar MZ, Lu JL, Yessayan LT, Kalantar-Zadeh K, et al. Association of Systolic Blood Pressure Variability With Mortality, Coronary Heart Disease, Stroke, and Renal Disease. J. Am. Coll. Cardiol. 2016;68:1375–1386.
  2. 2. Muntner P, Levitan EB. Visit-to-visit variability of blood pressure: current knowledge and future research directions. Blood Press. Monit. 2013;18:232–238. pmid:23676615
  3. 3. Yano Y, Reis JP, Lewis CE, Sidney S, Pletcher MJ, Bibbins-Domingo K, et al. Association of Blood Pressure Patterns in Young Adulthood With Cardiovascular Disease and Mortality in Middle Age. JAMA Cardiol. [Internet]. 2020 [cited 2020 Feb 10];Available from: pmid:31968050
  4. 4. Basson MD, Klug MG, Hostetter JE, Wynne J. Visit-to-Visit Variability of Blood Pressure Is Associated With Hospitalization and Mortality in an Unselected Adult Population. Am. J. Hypertens. 2018;31:1113–1119. pmid:29860426
  5. 5. Liu X, Logan J, Kwon Y, Lobo JM, Kang H, Sohn M-W. Visit-to-visit blood pressure variability and sleep architecture. J. Clin. Hypertens. Greenwich Conn. 2021;23:323–330. pmid:33492762
  6. 6. Lettau F, Schwarz EI, Stradling JR, Kohler M. Blood Pressure Variability in Obstructive Sleep Apnoea: Data from 4 Randomised Controlled CPAP Withdrawal Trials. Respiration. 2017;93:311–318. pmid:28351060
  7. 7. Marrone O, Bonsignore MR. Blood-pressure variability in patients with obstructive sleep apnea: current perspectives. Nat. Sci. Sleep. 2018;10:229–242. pmid:30174467
  8. 8. Xu X, Meng X, Oka S. Long term habitual vigorous physical activity is associated with lower visit-to-visit systolic blood pressure variability. Am. J. Hypertens. 2020;34.
  9. 9. Leary AC, Donnan PT, MacDonald TM, Murphy MB. The influence of physical activity on the variability of ambulatory blood pressure. Am. J. Hypertens. 2000;13:1067–1073. pmid:11041160
  10. 10. Tomitani N, Kanegae H, Kario K. The effect of psychological stress and physical activity on ambulatory blood pressure variability detected by a multisensor ambulatory blood pressure monitoring device. Hypertens. Res. 2023;46:916–921. pmid:36522422
  11. 11. Withings | Smart Scales, Watches and Health Monitors [Internet]. [cited 2022 Jul 18];Available from:
  12. 12. Pulse HR ‐ Which activities are detected by my watch? [Internet]. Withings Support. [cited 2022 Dec 26];Available from:
  13. 13. Edouard P, Campo D, Bartet P, Yang R-Y, Bruyneel M, Roisman G, et al. Validation of the Withings Sleep Analyzer, an under-the-mattress device for the detection of moderate-severe sleep apnea syndrome. J. Clin. Sleep Med. JCSM Off. Publ. Am. Acad. Sleep Med. 2021;17:1217–1227. pmid:33590821
  14. 14. Yang R-Y, Bendjoudi A, Buard N, Boutouyrie P. Pneumatic sensor for cardiorespiratory monitoring during sleep. Biomed. Phys. Eng. Express. 2019;5:055014.
  15. 15. Lim HM, Chia YC, Ching SM, Chinna K. Number of blood pressure measurements needed to estimate long-term visit-to-visit systolic blood pressure variability for predicting cardiovascular risk: a 10-year retrospective cohort study in a primary care clinic in Malaysia. BMJ Open. 2019;9:e025322. pmid:31005918
  16. 16. Mena LJ, Maestre GE, Hansen TW, Thijs L, Liu Y, Boggia J, et al. How many measurements are needed to estimate blood pressure variability without loss of prognostic information? Am. J. Hypertens. 2014;27:46–55. pmid:23955605
  17. 17. Taming of the Skew in Healthcare Data | Insurance Thought Leadership [Internet]. [cited 2024 Feb 1];Available from:
  18. 18. Kallioinen N, Hill A, Horswill MS, Ward HE, Watson MO. Sources of inaccuracy in the measurement of adult patients’ resting blood pressure in clinical settings: a systematic review. J. Hypertens. 2017;35:421–441. pmid:27977471
  19. 19. Delucchi KL, Bostrom A. Methods for Analysis of Skewed Data Distributions in Psychiatric Clinical Studies: Working With Many Zero Values. Am. J. Psychiatry. 2004;161:1159–1168. pmid:15229044
  20. 20. Nandy A, Basu A, Ghosh A. Robust inference for skewed data in health sciences. J. Appl. Stat. 49:2093–2123. pmid:35757589
  21. 21. Richardson DB, Kaufman JS. Estimation of the Relative Excess Risk Due to Interaction and Associated Confidence Bounds. Am. J. Epidemiol. 2009;169:756–760. pmid:19211620
  22. 22. de Havenon A, Petersen N, Wolcott Z, Goldstein E, Delic A, Sheibani N, et al. Effect of dihydropyridine calcium channel blockers on blood pressure variability in the SPRINT trial: a treatment effects approach. J. Hypertens. 2021.
  23. 23. Wang F, Boros S. The effect of physical activity on sleep quality: a systematic review. Eur. J. Physiother. 2021;23:11–18.
  24. 24. Dubinina E, Korostovtseva LS, Rotar O, Amelina V, Boyarinova M, Bochkarev M, et al. Physical Activity Is Associated With Sleep Quality: Results of the ESSE-RF Epidemiological Study. Front. Psychol. 2021;12:705212. pmid:34447339
  25. 25. Anderson GO. Getting Connected: Older Americans Embrace Technology to Enhance Their Lives [Internet]. AARP Research; 2018 [cited 2021 Apr 16]. Available from:
  26. 26. Wang S, Bolling K, Mao W, Reichstadt J, Jeste D, Kim H-C, et al. Technology to Support Aging in Place: Older Adults’ Perspectives. Healthcare. 2019;7:60. pmid:30974780
  27. 27. NW 1615 L. St, Suite 800Washington, Inquiries D 20036USA202-419-4300 | M-857-8562 | F-419-4372 | M. 7% of Americans don’t use the internet. Who are they? [Internet]. Pew Res. Cent. [cited 2021 Apr 17];Available from:
  28. 28. Saeed SA, Masters RM. Disparities in Health Care and the Digital Divide. Curr. Psychiatry Rep. 2021;23:61. pmid:34297202
  29. 29. Mullangi S, Kaushal R, Ibrahim SA. Equity in the Age of Health Care Information Technology and Innovation: Addressing the Digital Divide. Med. Care. 2019;57 Suppl 6 Suppl 2:S106–S107. pmid:31095047