News

Sleep is an important component of overall physical and mental well-being, and can predict important health outcome in older adults, including mortality. Sleep is often studied as a single construct represented by duration or the presence/absence of a disorder. However, scientists including Meredith Wallace, PhD, Daniel Buysse, MD, and Martica Hall, PhD, believe that sleep is a multidimensional construct, composed of domains including satisfaction, alertness/sleepiness, timing and continuity, as well as duration—domains that occur simultaneously and interact in complex ways. The team hypothesized that these interactions, when considered in conjunction with social and behavioral contributors, would provide an improved understanding of sleep’s predictive ability and enhance recommendations for measuring public health.

With the dual goals of determining the predictive ability of multidimensional self-reported sleep for mortality (relative to other established risk factors) and identifying what self-reported sleep characteristics are most predictive, the investigators analyzed the sleep data of 8,668 older adults, ages 65–99, from three epidemiological cohorts. They grouped predictors into six domains associated with mortality: sleep, sociodemographic factors, health behaviors, mental health, medications and physical health. The sleep domain included total sleep time, bedtime, wake time, time in bed, sleep efficiency, sleep latency, napping, symptoms of sleep disorders, medications, the Epworth Sleepiness Scale and—where applicable—the sleep quality item from the Pittsburgh Sleep Quality Index 

The team used variable importance metrics from a random survival forest (a machine learning algorithm) to rank the predictive abilities of 47 measures and the domains to which they belong for predicting all-cause mortality. “Unlike more traditional methods, this powerful approach empirically models complex, non-linear relationships among numerous predictors,” said Dr. Wallace, the study’s first author.

The findings, published in Journals of Gerontology: Medical Sciences, determined that the multidimensional sleep domain was a significant predictor of all-cause and cardiovascular mortality. Sleep’s predictive ability was lower than that of physical health, sociodemographic, mental health and medication domains, but higher than that of health behaviors (which include smoking status and alcohol use) and several other well-established predictors. 

 “We know that sleep is important for health. However, Meredith’s sophisticated statistical analyses tell us that sleep is a ‘player’ in health outcomes on par with other, more well-established risk factors,” said Dr. Buysse. “These findings also suggest that we might be able to intervene on multiple aspects of sleep in order to improve health outcomes.”

“The multidimensional sleep health domain was a better predictor of time to all-cause and cardiovascular mortality than was the health behaviors domain,” said Dr. Hall, the study’s senior author. “Health care providers often ask about health behaviors like smoking and physical activity when trying to assess a patient’s health. These data suggest they should also ask about and make recommendations related to sleep.” 

Multidimensional Sleep and Mortality in Older Adults: A Machine-Learning Comparison with Other Risk Factors
Wallace ML, Buysse DJ, Redline S, Stone KL, Ensrud K, Leng Y, Ancoli-Israel S, Hall MH
Journals of Gerontology: Medical Sciences, 2019. Vol 74: 1903-1909.