Counting Sleep

A Risk of Death

Meet Joe Higgins. He was 51 when he started having trouble getting a decent night’s sleep. The quality of his sleep had gradually declined for several years, but Higgins thought it was due to the medicine he was taking for his severe allergies.

“I came to dread the two o’clock meeting in the afternoon, because it was virtually impossible to keep your eyes open in that setting,” Higgins recalls, six years later. “And then you’re nodding off driving home from work.” After being tested at a Hopkins sleep center, he learned that his breathing during an hour. The condition is known as sleep apnea (also called sleep-disordered breathing). About one in ten women and one in four men have sleep apnea, which is caused by the recurring collapse of the upper airway’s soft tissue. As the upper airway muscles relax, airflow decreases and oxygen levels in the blood fall. This briefly rouses the sleeper as they struggle to breathe.

Punjabi, an associate professor of Medicine with a joint appointment in Epidemiology at the Bloomberg School, prescribed an ongoing course of continuous positive airway pressure (CPAP) therapy. Higgins wears a mask during sleep that allows the CPAP machine to keep his upper airway clear. It quickly improved Higgins’ sleeping. But this individual clinical success is just a precursor to a larger public health story. Punjabi enrolled Higgins in a study that is examining sleep apnea and diabetes risk. In Higgins’ case, his glucose tolerance has shown marked improvement. “That was something that was important to me,” says Higgins, whose parents both had type 2 diabetes.

“More and more, we’re recognizing the health consequences of a variety of sleep disorders,” says Punjabi, MD, PhD.

In August, Punjabi and Caffo, an associate professor in Biostatistics, published findings in PLoS Medicine that link sleep-disordered breathing with an increased risk of “all cause” mortality, particularly for men ages 40 to 70. “That’s a very important finding,” Crainiceanu says, one that was long suspected but never before shown. The findings also suggest that the increased risk of death is specifically associated with coronary artery disease, though more studies will be needed to confirm this. Drawing on data from more than 6,000 participants in the national Sleep Heart Health Study coordinated by Johns Hopkins researchers, the paper marks 15 years of data gathering and analysis. “I think it breaks new ground, especially in scientific terms,” says Crainiceanu, an associate professor in Biostatistics. “These results are based on the largest community cohort study of sleep.”

The study points to a surprising fact: Though sleep, as described by Keats, is “more secret than a nest of nightingales,” it generates an enormous amount of data. This plethora of second-by-second information can be both dream and nightmare for researchers.

The Architecture of Sleep

In the atrium of a Johns Hopkins Bayview Medical Center building, a mobile of blue spheres hangs in a surreal constellation. Deeper inside the building is a suite of seven bedrooms. Each room has a bathroom, television, furniture—and a modem-sized electronic unit, which gathers data on the sleeping occupant.

Welcome to the Sleep Study Center.  

This type of research suite marks the gold standard for getting quality data about what happens during sleep. Here, people like Joe Higgins spend a night and have their sleep patterns assessed. Electrodes are affixed to various parts of the body to gather measurements. The sensors in each room send the data to a central hub, where Punjabi points to a page with more than a dozen graphs stacked atop each other. For each subject the scientists gather a dozen or so types of data on eye movement, heart activity, leg activity, oxygen levels, respiration flow and so on. (The Greek roots of polysomnography explain it well: “poly”– many, “somno” – sleep, and “graphy” – write).

Electrodes affixed to various body parts of a sleep subject generate a plethora of second-by-second information that is both dream and nightmare for researchers.

“The thing about [studying] sleep is that it’s a very, very quantitative field,” says Punjabi. “You’re measuring brain activity, and you’re collecting data every second.” He continues, “A hundred to five hundred times per second you’re collecting EEG data. That’s one of the most voluminous things in medicine.”

Yet up to now, the methods for assessing sleep as a series of these discrete stages have been “extremely crude.” That’s why he sees the work of biostatisticians Caffo and Crainiceanu as so important: they’re helping fine-tune the picture from the growing volume of data. “It’s [a matter] of taking a tremendous amount of extracted data and making sense of it,” says Caffo, PhD.

Scientists study our “sleep architecture”— the relative structure of our time sleeping portioned out among five stages, identified by EEG data: stages 1 through 4, plus REM (rapid eye movement) sleep. In The Family That Couldn’t Sleep, D.T. Max describes the typical sequence: Sleep begins with rough alpha waves (stage 1), deepens into longer theta waves (stage 2), followed by sleep spindles, which take the sleeper into the profound sleep of the rolling delta wave of stages 3 and 4. These get interrupted periodically by the jagged lines of REM sleep.

Caffo and colleagues have managed to recast this conventional categorization of sleep as a sequence of five stages and uncover a new pattern that allows clearer correlations to a person’s health. In a 2008 paper in the Journal of Clinical Sleep Medicine, Caffo and Punjabi showed that two groups that seemed to have the same sleep architecture (meaning they spent similar proportions of sleep in the various stages according to conventional interpretation) were in fact very different in their vulnerability to disease.

Research on sleep-disordered breathing has traditionally been based on “composite” sleep-stage summaries, notes Caffo. These were high-level views that could only show the percentage of time spent in each stage of sleep. Caffo and Punjabi wanted to explore the data more precisely. So Caffo and his colleagues tapped two biostatistical tools—log linear and multi-state analysis—to analyze the multiple types of sleep data and tease out previously unknown differences in the transitions between the various sleep stages.

These differences in transitions overlooked by traditional measures,” says Caffo, “may produce new clinical indices for measuring sleep disturbance.

Subjects with sleep-disordered breathing had a greater number and higher rates of sleepstage transitions than those without sleepdisordered breathing, the authors note. All these transitions are chopping up the duration spent in each sleep stage. Overall, they may get the same amount of REM sleep, but the REM sleep is composed of more, shorter episodes. Researchers believe people who experience rapid transitions may get the same number of minutes of REM sleep but it may not be the same quality.

“This new way of looking at the data essentially gives you a more complete picture,” says Caffo, “and a better predictor of disease."

Bad sleep is clearly linked with poor health, but researchers now want to know if they can prevent cognitive and functional decline by treating sleep problems.

Crainiceanu, PhD, agrees. “You can see things that were impossible to think about before,” he says, such as differences in EEG data for smokers versus non-smokers or among different age groups. He describes the advances from the biostatistical viewpoint in terms of multi-resolution analytical tools: “This allows researchers to investigate multiple sub-groups while quantifying differences all the way from the fine detail of an individual’s raw EEG data (125 observations per second) to the heavily summarized individual’s sleep architecture.”

At the population level, these analytical methods are now allowing researchers to examine data from previous cohort studies in much greater detail than before.

This comes at a time when there is finally a rich trove of sleep data to be explored.

After nearly two decades, several longterm studies have amassed the nocturnal patterns of thousands of people. One, the Osteoporotic Fractures in Men Study, known as MrOS or more informally “Mister Os,” began in 2000 with 5,995 men ages 65 and older and continued until this year. Another, with 9,700 women in the same age range, began in 1986 in Minneapolis, Portland, Oregon and Pittsburgh and continued over 20 years. And the Sleep Heart Health study coordinated by Johns Hopkins recruited more than 6,400 people in cities across the U.S.

With the greater computational capacity gained in the last two decades, researchers can now perform more fine-grained analysis of that EEG data and can ask, Do any disturbances during segments of a person’s sleep correlate with their medical conditions? Answering that question with quantitative input “is extremely vital for moving this field forward,” says Punjabi. For him, the PLoS Medicine paper linking sleep disruption to mortality is more a prelude than a finale.

“Instead of saying, ‘The story is done here,’” he observes, “I think the story is just starting.” He calls for intervention studies that explore whether reversing or minimizing the physiological effects of sleep apnea and other sleep disorders can decrease the risk of mortality.

The Case for Treatment

Adam Spira, an assistant professor in Mental Health, sees opportunities for improving treatment for sleep disorders in psychology and medicine. Research fascinates him, and what really interests him is not simply the evidence that bad sleep may be linked to health problems, but the potential to ameliorate those problems if there is indeed a causal link. 

"We have good treatments for bad sleep in older adults, whether it’s insomnia or apnea,” he says. Behavior therapy and cognitive-behavioral therapy (CBT), for instance, is effective in reducing insomnia, and CPAP therapy is useful for addressing apnea. “So,” he says, “it will be really good to know if treating these problems can improve those outcomes.” If so, some problems typically accepted as a product of aging may be treatable.

Spira, PhD, came to Hopkins in 2008 after a fellowship in geriatrics and geropsychiatry at the University of California-San Francisco (UCSF), where he looked at the sleep of older people in epidemiological terms. He received an invitation to come to Hopkins while he was working on studies of sleep in MrOS and the Study of Osteoporotic Fractures (SOF) with Katie Stone, a scientist at the California Pacific Medical Center Research Institute. Spira respects the team of Hopkins scientists working on sleep research, and was drawn by the impressive range of interdisciplinary collaborators. “You’ve got a biostatistician, you’ve got Naresh who is a pulmonologist and critical-care medicine doc, you’ve got a professor in epidemiology here, and then my PhD is in clinical psychology.”

With a major grant from the National Institute on Aging recently approved, Spira and colleagues at the Bloomberg School and the School of Medicine aim to delve further into the rich material of the Mister Os and SOF studies and detect links between poor sleep and functional decline. The five-year career development award will allow Spira to learn from mentors including Punjabi, Bloomberg School Mental Health Professor George Rebok, PhD, and his postdoc mentor Kristine Yaffe, MD, Psychiatry and Epidemiology professor at UCSF. The study will explore that data in more detail for connections between sleep disturbance and decline in performing daily functions such as housework and grocery shopping.

Punjabi sees that work as critical. “What’s really necessary—and this is what Adam’s starting to work on with the Mister Os study,” he says, “is objective, fine-grained monitoring of actigraphy, which is a nice way to look at sleep duration objectively.”

Spira first worked with actigraphy studies in California, measuring sleep disturbance using an actigraph, a small device that looks like a wristwatch with a blank face. Worn on the wrist, the actigraph contains a tiny accelerometer that measures movements as an indication of sleep and sleep disturbance. While less comprehensive than the sensors in a sleep center, actigraphy has the advantage of continuous collection for up to two weeks. "The new models are waterproof so you don't ever have to take them off," says Spira. Actigraphy data gets uploaded, run through a set of algorithms, "and that gives you all sorts of quantitative values representative of sleep parameters."

Spira pulls up sample data on the computer in his office. "These are some of the sleep parameters," he says. "Sleep efficiency, that's the percent of the time in bed that you're actually spending asleep. Here, this example is quite high—it's 95 percent basically." Sleep efficiency of under 80 percent is considered poor sleep.

Spira expects to devote the next five to 10 years to examining the link between latelife sleep disturbance and both cognition and function (which includes basic tasks such as selffeeding and bathing, as well as grocery shopping and preparing food). Knowing which aspects of bad sleep—short duration, fragmentation, etc.—lead to poor daytime functioning would help researchers develop treatment studies aimed at improving poor outcomes. "Can we prevent cognitive and functional decline? That's the hope—that we'll be able to change these unfortunate trajectories," Spira says.

Those negative trends could possibly be pushed further toward the end of life to maximize quality of life measures, such as the ability to think clearly, the capacity to take care of yourself and others, and the ability to play an active role in the community.

"We may be closer to understanding how negative consequences emerge from bad sleep," says Spira, and so, "from a public health perspective, we might better understand how to protect the health of this growing segment of our population."

Perchance to Dream

Spira has launched a field-based actigraphy study, this one looking at the role sleep might play in quality of life among older African-American adults in Baltimore. His small project is piggybacking on a larger study of the Experience Corps, a national program that has recruited 2,000 older people to tutor and mentor elementary school students in 22 cities across the country, providing literacy coaching, homework help, role models and attention.

"I wanted to see the relationship between their sleep, as measured by actigraphy, and function, and whether over the course of four months, the sleep of those in the Experience Corps program improves more than the sleep of those in the control group," he says.

It may be true, for example, that the physical activity involved in volunteering in schools every week could lead to better sleep, perhaps by increasing both physical activity and improving mood. That study will also expand the range of populations for which a sleep profile exists, notes Spira, whose project won him an award last spring through the JHSPH Faculty Innovation Fund.

Epidemiological studies such as these, coupled with ongoing advances in biostatistics, hold the potential for dramatically improving health on a wide scale, says Crainiceanu.

"Every month there's a new very large data-set study that has similar problems that we are able to address," he says. "All of them are gaining strength from the current methodological developments. Studies that include images, such as MRIs, are one of the big areas that will benefit from this research, but there are many, many other areas as well." Better ways to summarize large amounts of imaging data will help public health professionals sift through data, identify key trends and patterns more quickly, and test solutions.

"The research we have is oriented toward population level imaging and functional analysis," Crainiceanu adds, noting that
other groups are interested in their work. The challenge is to develop faster and better algorithms and a better infrastructure for these huge data sets.

The public health field is gaining new tools from the collaboration on sleep research, says Crainiceanu. The sleep findings have gained new robustness from the breakthroughs in analytical methods and led to fresh perspectives on how to look at vast data sets that are increasingly common across the field of public health. These broader benefits will be highlighted at a conference that the Bloomberg School's Department of Biostatistics is planning for 2011, about statistical methods for very large data sets, such as those generated by sleep research, genomics research or observational studies.

"New technologies and modern computing are letting us better measure health and biology," says Caffo. To more effectively use this new information, he says, we need new statistical methods.

Their collaboration, in Punjabi's poetic words, "brings new insights to old problems and helps unravel some of the complex tapestry that weaves sleep with our medical health."