Detecting stress in college freshman from wearable sleep data
PsyArXiv - Bloomfield - mental health, sleep, wearables, respiratory rate, college
This was a study of 507 first semester college students enrolled at a public univeristy who wore an Oura ring for three weeks, taking weekly surveys which were compared against sleep biometrics collected from the Oura ring device.
The study noted that 66% of all survey responses (participants * number of weeks) indicated moderate-to-high stress. Additionally, the relationship between stress and nightly respiratory rate was the most robust compared to other biometrics like skin temperature and heart rate.
Abstract:
“Sleep plays an important role in health and functioning, yet little is known about how it is linked to stress in young adults. Consumer wearables have been particularly successful at quantifying sleep and may be useful in identifying changes in mental health. The transition to college has a notable impact on both stress and sleep behaviors. Students from a public university provided continuous biometric data and answered weekly surveys for their first semester of college (N=507). Longitudinal models showed that increased average respiratory rate during sleep was associated with higher weekly perceived stress scores, effects that persisted after controlling for gender, race, first-generation status, mental health history, trauma history. Specifically, for every increased breath per minute, the odds of experiencing moderate-to-high stress were 1.25 times higher (a 25% increase), holding demographic and psychological history variables constant. Notably, relationships with stress were specific to respiratory rate, but were not found with other sleep measures previously linked to stress, such as heart rate variability, total sleep hours, sleep efficiency, sleep onset latency, and wake up count. Consistent with previous work, female gender, a previous mental health diagnosis, and previous exposure to multiple traumas were significantly associated with self-reported stress. These findings point to respiratory rate as a potentially important factor to measure due to its robust association with stress among college students. Wearable data may help us identify, understand, and to better predict stress, a strong signal of the ongoing mental health epidemic among college students”
The authors note that there is a not a uniform correlative response between stress and respirator rate, meaning that average respiratory rate (ARR) was had the strongest correlation with moderate-to-high-stress (PSS) for students with “average” ARR.
It was also noted that stress correlates with sleep latency (how long it takes you to fall asleep), total sleep time, HRV (time between heart beats), resting heart rate (RHR), and ARR.
Three models were tested to assess the odds ratio for PSS, controlling for different covariates. Model 1 — controlling for gender — found that for every +1 breath / minute PSS odds were 1.28x (increased 28%). Model 2 — controlling for prior mental health diagnosis and traumatic exposure — found that for ever +1 breath / minute PSS odds were 1.31x (increased 31%). Model 3 — controlling for gender, race, first-generation status, mental health diagnosis, and traumatic exposure — found that every +1 breath / minute PSS odds were 1.25x (+25%).
Limitations:
The authors noted that their study did not account for medication used (prescription or other sleep aids) or sleep disorders. However, their sample appeared to get notably more sleep / night than other college-aged groups. It was also noted that the authors did not have baseline data for the sample prior to entering college.
All of that means that this sample could have been, but weren’t necessarily, particularly good sleepers or heavily medicated.
What We Can Learn:
By far the biggest risk factors for stress were being female (OR 3.43, SE 0.37) and having a prior mental health diagnosis (OR 6.99, SE 0.36). These aren’t necessarily surprising figures, though we do need to keep an eye out for gender bias, which has been a noted concern for the PSS-10 assessment that was used (ref.).
Specifically, the helplessness subscale (the 6 negatively phrased items) are prone to being scored more highly by females than males. Similar confounding factors could be said of specific mental health disorders (e.g. feelings of helplessness and hopelessness are symptoms of depression).
The data set for the original reference isn’t publicly available, but it would have been interesting to see a fourth model accounting for the helplessness and self-efficacy sub-scales of the PSS-10.
Additionally, it would have been helpful to know which mental health diagnosis were included and which were excluded and by what criteria (e.g. were students simply asked or were there clinician referrals).
Lastly, the strongest predictability of ARR for PSS was for people with “average” respiratory rates. This is fascinating and highlights the importance of noting that people can have both a hypo- and hyper-arousal response to stress. That is, in Dan Siegel’s Hand-Brain-Model, we have both “lid flipping” (cessation of rational / frontal cortex activity) both up (hyperarousal: fight, flight) and down (hypoarousal: freeze, feint).
Bloomfield, L., Fudolig, M. I., Dodds, P. S., Kim, J., Llorin, J., Lovato, J. L., … Danforth, C. M. (2023, September 19). Detecting stress in college freshmen from wearable sleep data. https://doi.org/10.31234/osf.io/eu896