How to Measure Sleep
If you have a sleep problem, sleep doctors have some basic measures to monitor it. But researchers publishing in the 1 May Physical Review Letters propose a new way of looking at sleep data that they hope will give more insight into the problems and treatments. It measures the likelihood of a person’s sleep state changing from one time interval to the next. In a proof-of-concept test, the team used it to correctly distinguish between people before and during treatment for obstructive sleep apnea (OSA), a chronic obstruction of the airway.
As we sleep, our brains pass through a series of stages, like shifting between gears in a car. The most famous is REM sleep, characterized by fluttering eye movements. But other categories include light sleep (stage S1), stable sleep (S2), slow-wave sleep (SS), and even being awake (W), each identified by a signature from polysomnography (PSG), in which electrodes record from the scalp, eyelids, and heart.
For people with apnea, who are constantly waking up during the night, the longest continuous duration of restful sleep is an important measure, so some clinical researchers have looked at the durations of each state in search of patterns. Fewer studies have considered the probabilities for switching from one state to another, says Jong-Won Kim of the University of Sydney, Australia. Kim and his colleagues took a new approach to calculating these probabilities, hoping to derive a quantitative technique that might provide more detailed information about an individual’s sleep and perhaps relate to what’s really happening in the brain.
The team tabulated data from 113 subjects undergoing treatment for sleep apnea at Seoul National University Hospital in South Korea. The PSG readings came in 30-second chunks, assigned to a sleep state according to standard rules used in sleep research.
The researchers calculated the likelihood, after every 30-second interval, that a sleep state would have ended and switched to a new state. For REM sleep, the likelihood of one 30-second REM interval being followed by a non-REM interval was 14 percent for nearly the first 10 minutes of the state and then dropped to 4 percent, implying that longer REM intervals were more stable. They saw the same pattern in the other states: the transition probability was relatively high for the first 2-10 minutes, but after that, switches became less likely. The data were consistent with a modified “Markov” model–common in other areas of biophysics–which allowed the team to compute a complete matrix containing the probabilities of switching from each state to every other state. Although the Markov model is mathematical and not physiological, “our results strongly constrain any such model that might be developed,” Kim says.
To see if their new sleep measures would reflect actual changes in the patients’ experiences, the team repeated the analysis on the apnea patients as they underwent continuous positive airway pressure (CPAP) treatment, in which a mask blows air down the person’s windpipe. For the largest component of sleep, S2, the probability for not switching after each 30-second interval increased by 28 percent overall, and plots of individual patient durations vs. time were statistically different in 70 percent of cases. Kim says he and his colleagues are currently studying additional hypnograms from 5,000 patients at the Seoul hospital to see if their method can diagnose other sleep disorders.
Drawing conclusions from this study about underlying sleep mechanisms is probably not a good idea, says Dennis McGinty, a neurophysiologist at the University of California in Los Angeles, because apnea patients are “a special case.” But as a clinical tool it may have merit, according to David Rapoport, a sleep researcher at New York University. The CPAP effect is “a good first step,” he says. The next goal would be to demonstrate that CPAP helps those suffering from mild apnea or another disorder where it’s harder to tell if the treatment is working.
JR Minkel is a freelance science writer in New York City.