Study design and overview
This prospective cohort study was nested within a larger clinical implementation at two military treatment facilities13,14. Throughout the study, participants continued to receive routine clinical care. The study was approved by the Institutional Review Board at the Walter Reed National Military Center ([WRNMMC]; WRNMMC-2019-0258).
Participants
Patients with sleep problems were recruited from Walter Reed National Military Medical Center (Internal Medicine clinic and the Sleep Disorders Center) and Alexander T. Augusta Military Medical Center (formerly Ft. Belvoir Community Hospital; Family Medicine clinic and Sleep Disorders Center). Inclusion criteria included (1) active-duty military servicemember and/or Defense Eligibility Enrollment System (DEERS) beneficiaries, including retired military or military dependent, (2) age 18–75 years, (3) ownership of a smartphone, and (4) provider or self-referral for sleep problems (including insufficient sleep). Exclusion criteria included (1) pregnancy, (2) untreated and/or uncontrolled major medical or psychiatric illness, and (3) pending retirement or permanent international change of station.
Measures
Subjective sleep diary
Specific sleep parameters included total sleep time (TST), sleep onset latency (SOL), number of awakenings (NOA), wake after sleep onset (WASO), and subjective sleep quality (QUAL, scored from 0-lowest to 10-highest).
Daytime symptoms
Daytime symptoms were measured using twice-daily surveys. Individual items assessed subjective cognition (feeling alert, clear-headed), energy level (feeling refreshed, fatigued), and mood (feeling happy, sad, stressed, relaxed) using an “I feel refreshed” format. Five response options ranged from “not at all” to “very” (scored from 0–4, respectively).
Commercial wearable sleep tracker
Objective sleep data was obtained from the Fitbit Inspire 2 and included TST and SE. This device uses a triaxial accelerometer to measure ambulatory movement to estimate sleep15.
Procedures
Remote study procedures
Participants underwent informed consent and were onboarded by trained study personnel and then received a commercial wearable sleep tracker via priority courier. Next, participants received training and completed a sleep history questionnaire and standard research questionnaires via a smartphone application (WellTap®) with versions for iOS and Android.
Ecological momentary assessment (EMA)
For ten consecutive days, participants completed EMAs, including sleep diaries each morning and daytime symptom assessments 2x/day (i.e., 20 total EMAs over ten days). The ten-day duration was selected to ensure an adequate sample of sleep and daytime symptoms on both workdays and non-workdays. Surveys were completed upon arising and before bed, with specific times personalized based on participants preferred wake/rise times.
Commercial wearable sleep tracker
During onboarding, participants were instructed to remove the sleep tile from the Fitbit application, turn off alerts in the Fitbit app, and to wear the Fitbit device continuously except while bathing or charging the device.
Analytic plan
To test the hypothesis that prior-night sleep is associated with next-day symptoms lagged over 10 days, we developed a series of 40 mixed models (MMs). Within separate MMs, individual prior-night sleep diary (TST, SE, QUAL) and Fitbit (TST, SE) parameters were entered as independent variables, and individual next-day symptoms (feeling alert, clear-headed, refreshed, fatigued, happy, sad, stressed, and relaxed) were entered as continuous dependent variables. All models controlled for age and sex. Because our analytic approach included a large number of statistical tests, the Benjamini-Hochberg procedure for correcting the false discovery rate (BH-FDR) was employed. Statistical significance was set at 0.05. We also examined the association between within-day mood and same-night sleep, using a nearly identical approach.
Participants
Participants (N = 270, 55.2% men, mean age = 45.8 [SD = 13.0] years) included active duty (41.1%), retired military (27.4%), or civilian (24.8%) adults with sleep complaints. Participants self-identified as White (56.3%), Black (23.7%), Hispanic (8.5%), or Asian (7.0%) race/ethnicity. As described elsewhere, based on validated research questionnaires, participants were found to be at high risk for OSA (65.6%) and reported moderate to severe symptoms of insomnia (38.2%), excessive sleepiness (38.5%), depression (20.4%), and anxiety (20.4%)13.
Adherence and missing data
Participants completed a mean of 9.3 (SD = 1.3) of ten possible sleep diaries and a mean of 18.6 (SD = 2.5) of twenty possible daytime symptom surveys. (Fig. 1 depicts daytime symptom survey results over ten days.) Fitbit data was available for 94.8% of participants (<5% of data were unavailable due to changes within Fitbit user permissions during the course of the study).
Within each panel, the y-axis represents daytime symptom severity, with higher scores indicating greater levels of a given construct. The x-axis represents days one through ten, i.e., the intensive remote monitoring period. A = clear-headed; B = alert; C = happy; D = sad; E = stressed; F=relaxed; G = refreshed; H=fatigued.
Prior-night sleep and next-day symptoms
As depicted in Table 1, results of lagged MM analyses revealed that all prior-night sleep diary variables were significantly associated with next-day symptoms (all \(p{\rm{s}} < 0.001\) with \({df}=2197\)) over ten days. Specifically, prior-night sleep diary parameters (TST, SE, and QUAL) were positively associated with next-day feeling alert, clear-headed, refreshed, happy, and relaxed; and negatively associated with next-day feeling fatigued, sad, and stressed. Prior-night Fitbit sleep parameters were significantly associated with most next-day symptoms (largest \(p=0.023\) with \({df}=1956\)). Specifically, Fitbit TST and SE were positively associated with next-day feeling alert, clear-headed, refreshed, happy, and relaxed and negatively associated with next-day fatigue. Fitbit TST was negatively associated with next-day stress. These patterns of results were consistent in sensitivity analyses performed separately among civilian (n = 67), retired (n = 74), and active-duty individuals (n = 129).
Within-day mood and same-night sleep
To determine the association between within-day mood and same-night sleep, we performed a series of 10 separate MMs with positive and negative mood as independent variables and sleep diary (TST, SE, QUAL) and Fitbit (TST, SE) variables as dependent variables, again lagged over 10 days while controlling for age and sex. Within-day positive mood demonstrated a significant positive association with same-night sleep diary QUAL (\(\beta =0.206\), t = 3.86, p < 0.001, df = 1858). Within-day negative mood demonstrated a significant negative association with same-night sleep diary QUAL (\(\beta =-0.225\), t = 3.13, p = 0.002, df = 1858). The BH-FDR procedure was again employed.
This is the first study to employ EMA methods to determine the association between prior-night sleep and next-day symptoms among patients with a broad range of sleep problems. It is also the first such analysis among military personnel. Results indicated strikingly robust associations between subjective and objective sleep and daytime symptoms over ten days. Further, patient engagement and adherence was notably high, supporting the potential utility of EMA approaches among patients receiving sleep medicine care.
Mobile health and remote monitoring technologies such as EMA hold great promise for treatment researchers as well as clinicians. For example, EMA reduces recall bias and enhances ecological validity, thus enriching traditional outcomes assessments. Within the context of clinical care, EMA methods can also predict treatment response or identify individuals at risk for relapse for behavioral health conditions1. In our view, future sleep clinical trials should evaluate these potential benefits of EMA, including the application of EMA as a novel outcome measure9.
This study possesses strengths, including our multimethod assessment of subjective and objective sleep and daytime symptoms throughout the ten-day intensive remote monitoring assessment. Further, the constancy of results observed in multiple models increases confidence in a bona fide association between prior-night sleep and next-day symptoms. Concurrently, our study was limited by our convenience sample of smartphone owners from only two military facilities in one geographic region, as well as our observational study design, which is unable to determine causality.
In summary, our findings strongly support the potential utility of EMA methods to enhance the measurement of common daytime symptoms in patients with sleep disorders. Future research should seek to leverage EMA methods to guide personalized care.