Understanding discrepancies between self-reported and objective sleep in adolescents and young adults with subacute concussion

Abstract

Study Objectives

Conduct a multidimensional analysis of sleep perception, objective sleep, and neuropsychiatric wellbeing in individuals with subacute concussion compared to controls.

Methods

Thirty-one recently concussed individuals completed the Pittsburgh Sleep Quality Index, Insomnia Severity Index, and Patient-Report Outcomes Measurement Information System measures of depression, anxiety, stress, and cognitive function. Concussion symptom severity scores (Sports Concussion Assessment Tool) were obtained from participants’ health records. Sleep parameters were derived from at least 7 days of monitoring with the Emfit QS device (total sleep time [TST], time in bed, sleep onset latency, sleep efficiency, wake after sleep onset). Data were compared to 19 controls using parametric or non-parametric tests for independence (α = 0.05). Pearson correlations and linear mixed models assessed relationships between data modalities.

Results

Concussed individuals reported worse sleep and had lower sleep efficiency, longer time in bed, and greater sleep onset latency than controls (p < .05). Patient-Report Outcomes Measurement Information System symptom scores moderated these relationships at significant or near-significant levels. Controls demonstrated agreement between reported and measured sleep (for TST: r = 0.52; p = .023) and a positive relationship between sleep dissatisfaction and wake after sleep onset (p < .05). These relationships were not observed in individuals with concussion. Moreover, individuals with greater discrepancy between reported and measured sleep scored higher on Sports Concussion Assessment Tool concussion symptom inventories (β_TSTdisc = 9.5/h; p_beta = .007; p_model = .022; Total R² = 0.34).

Conclusions

Individuals with subacute concussion exhibited worse self-reported and objective sleep than controls, but showed discrepancy between reported and measured sleep characteristics that correlated with concussion severity at diagnosis. Future assessment of sleep discrepancy may improve understanding of post-concussive sleep disturbance.

Statement of Significance

Previous studies of sleep after concussion base findings primarily on self-reported sleep quality questionnaires. Few include objective data, or adjust for demographic and neuropsychiatric covariates, which limits our current understanding of post-concussive sleep. Furthermore, there is a paucity of knowledge about sleep during the subacute phase of concussion recovery, which can be a critical period for those experiencing protracted symptoms. To help fill this gap, we provide a characterization of sleep in adolescents and young adults with subacute concussion. Our study evaluates the agreement between self-reported and objectively measured sleep after concussion and is the first to identify sleep discrepancy as a potential correlate of injury severity in this population.

Introduction

Traumatic brain injury (TBI) imposes a significant and increasing health risk. Around 80 per cent of all TBIs are mild, characterized by normal imaging, a Glascow Coma Scale score of 13–15, loss of consciousness lasting <30 min, and no more than 24 h of altered consciousness or post-traumatic amnesia [1-3]. Mild-TBI (mTBI) was recently estimated to affect roughly 3 million Americans and 42 million people worldwide each year, though it is likely that these injuries are underreported [3–5]. Commonly referred to as concussion, mTBI can create long-lasting consequences, including headache, mood disorders, and cognitive impairment [2,6–8].

Sleep disturbances affect 30 per cent to 70 per cent of concussed patients and can encompass a variety of problems including circadian dysregulation, insomnia, hypersomnia, frequent awakenings, among other sleep pathologies [6,9–18]. Importantly, individuals who report sleep disturbance during recovery from concussion also experience more severe post-concussive symptoms and protracted recoveries [9,10,16, 19–22]. Treatment of these patients is further complicated by heterogeneity in predominant symptom type, injury severity, and whether patients are in the acute, subacute, or chronic phase of recovery. As a consequence, post-concussive sleep disturbance remains ill-defined both clinically and in research [9,13–18].

A panel of sleep research experts has agreed that sleep latency, efficiency, and wake after sleep onset (WASO) are effective sleep quality parameters in otherwise healthy individuals [23]. However, there is great variability in the technology and methodology used to study sleep after concussion [13,21,22,24]. Our knowledge of post-concussive sleep disturbance is primarily informed by subjective or self-reported sleep quality surveys, which are useful tools but can be limited by reporting bias [9,13,25,26]. Many investigators have cautioned against using self-reported sleep data as a surrogate for objective parameters, arguing that these data capture distinct dimensions of sleep [25–30]. This distinction offers significant clinical implications in instances where post-concussive sleep disturbance is treated based on patient reporting.

To date, few studies of sleep and concussion include both self-reported and objective sleep parameters (derived from polysomnogram or actigraphy). Moreover, no previous study has investigated the consequence of subjective–objective sleep discrepancy during concussion recovery. Recent literature reviews also cite a critical need to include mental health symptoms and psychiatric comorbidities when studying concussion [9]. These factors, such as anxiety, depression, stress, and pain sensitivity, can modify patients’ perception of their symptoms, including sleep quality [6,9,31,32]. In summary, characterizing post-concussive sleep requires a multidimensional approach.

The primary aim of this study was to assess the relationships between self-reported sleep quality, neuropsychiatric comorbidity, and objective sleep–wake parameters in adolescents and young adults with subacute concussion. We hypothesized that these individuals would be different from controls in both subjective and objective sleep. We also hypothesized an association between self-reported sleep quality and duration and objective sleep parameters assessed via a contactless sleep monitoring device. The Emfit QS (Emfit) is a contactless sleep monitoring device that is placed under a participant’s mattress and detects motion during sleep via ballistocardiography (Emfit Corp., Kuopio, Finland). In a previous validation study, our group found an intra-class correlation of 0.99 between Emfit and wrist actigraphy for all sleep parameters assessed except WASO [33]. In this study, individuals within 12 months of a concussion diagnosis completed evaluations of sleep quality and neuropsychiatric outcomes and underwent sleep monitoring for 7–30 days using the Emfit. We ultimately sought to demonstrate a comprehensive approach to post-concussive sleep characterization in these individuals.

Materials and Methods

Ethical approvals

This paper reports findings from a prospective cohort study of post-concussive sleep disturbances that began in May 2022 at an academic hospital in the Pacific Northwest region of the United States. All study procedures were approved by the Oregon Health & Science University (OHSU) Institutional Review Board.

Study participants

Participants were identified from the OHSU Concussion Clinic, from the community, or through Best Practice Advisory (BPA). BPA is a research recruitment tool used to identify patients who meet study inclusion criteria based on the hospital’s electronic health records. Once a provider added a concussion-related diagnosis into the patient’s electronic health record, a study team member was alerted. Potential participants were contacted by a study team member at their Concussion Clinic visit or by MyChart message, phone, or email. Participants were screened before enrollment into the study and were eligible to participate if they were (1) 12–50 years of age, (2) a student or currently employed, (3) not currently pregnant or incarcerated, (4) had a mailing address serviced by USPS, and (5) had access to a smartphone that could receive text messages and access the internet. Participants were excluded if they were unable to provide consent or were non-English speaking. All participants provided informed consent or assent before study entry; consent was obtained from a parent or legal guardian for all participants under 18.

Participants were allocated to the concussion cohort if they received a diagnosis of concussion given by a medical provider based on the criteria set forth by the American College of Rehabilitation Medicine [1,34]. Control participants were selected from community volunteers and required to have no history of concussion. Eighty-six individuals were consented to participate in the study (Figure 1). Of the consented participants, 36 were excluded from analysis for reasons including: the enrollment questionnaires were not filled out (n = 1), the participant worked night-shift hours (n = 1), the time from injury to enrollment was >1 year (n = 8), <7 nights of data were recorded (n = 16), or the participant denied Emfit participation (n = 10). Based on these criteria, 31 individuals with concussion and 19 controls were included in the final analysis.

Figure 1.

CONSORT diagram of subject selection for final analysis based on study criteria. Diagram of included participants from consent and enrollment to analysis.

Subjective questionnaires

Upon enrollment, participants were asked to provide demographic and injury history information and to complete the Pittsburgh Sleep Quality Index (PSQI), Insomnia Severity Index (ISI), and Patient-Report Outcomes Measurement Information System (PROMIS) questionnaires for depression, anxiety, stress, and cognitive function. Study data were collected and managed using REDCap electronic data capture tools hosted at OHSU [35,36]. The PSQI is a validated measure of sleep quality and disturbances containing 19 self-rated questions, assessing sleep during the previous month. The 19 questions are split into seven components and are each weighted equally with a score of 0–3. These seven components are totaled to yield a global PSQI score ranging from 0 to 21, where a total score >5 significant sleep disturbance [37]. The ISI is a validated measure for the assessment of insomnia “over the last 2 weeks,” which contains seven self-rated questions. Each question is rated on a 0- to 4-point Likert scale, and total scores range from 0 to 28 [38]. A total score of 0–7 indicates the absence of insomnia, 8–14 indicates subthreshold insomnia, 15–21 indicates moderate insomnia, and 22–28 indicates severe insomnia [39]. PSQI data for two participants from the concussion cohort were excluded due to erroneous form completion.

PROMIS is a collection of tools developed by the National Institutes of Health (NIH) for standardized measurement of patient-reported health outcomes in research and clinical practice (freely accessible at https://www.healthmeasures.net) [40]. Since our study included both adolescents and young adults, we used the pediatric-adapted versions for consistency and to allow for comparison between groups. Participants completed the PROMIS Pediatric Short Form v2.0—Depressive Symptoms 8a (8 items), PROMIS Pediatric Short Form v2.0—Anxiety 8a (8 items), PROMIS Pediatric Short Form v1.0—Psychological Stress Experiences 4a (4 items), and PROMIS Pediatric Short Form v1.0—Cognitive Function 7a (7 items) questionnaires [41–45]. These forms are like their adult counterparts in target symptoms and scoring method but are worded more simply. Of note, there currently is no equivalent to the PROMIS psychological stress experiences for adults. Participants reported their symptoms using a 1- to 5-point Likert scale with answers “never” to “almost always” for the depressive symptoms and anxiety questionnaires. On the psychological stress experiences questionnaire, subjects reported their symptoms using a 1- to 5-point Likert scale with answers “never” to “always.” Each question for the cognitive function questionnaire was scored on a 1- to 5-point Likert scale with answers “none of the time” to “all of the time.” These points are summed to produce raw total scores for each PROMIS measure. Higher scores on the PROMIS depression, anxiety, and stress forms indicate greater burden of symptoms, whereas higher scores on the PROMIS cognitive form indicate better cognitive function.

The Sports Concussion Assessment Tool (SCAT) is a standardized measure designed for use by medical providers when assessing symptoms after concussion [46,47]. Within this assessment, patients rate the severity of 22 different symptoms on a scale of 0–6. The SCAT provides a total symptom severity score from 0 to 132, where higher scores indicate more severe concussion symptoms [47]. SCAT symptom severity scores closest to each concussed individuals’ date of injury were obtained from their electronic health records if available. Data were not available for three individuals in this cohort.

Contactless sleep monitoring

Objective sleep was assessed with the Emfit QS movement monitor (Emfit Corp., Kuopio, Finland). In a previous validation study, Emfit has shown high inter-class correlation with wrist-worn actigraphy commonly used in sleep research [33]. The Emfit device consists of a sensor placed under the subject’s bed at the thoracic level and relies on plethysmography to detect motion during sleep. The methods to extract sleep data from the Emfit monitor have been previously described by our group [33]. Briefly, the monitor collects motion data at a 200 Hz sampling rate. Data are transmitted in real-time to a secure and encrypted server. Continuous data collection allows for capture of bed entry and exit without the need for a sleep diary. Once acquired, raw data are binned into 1-min epochs. Like conventional actigraphy, and as described in previous studies, each epoch’s activity level is used to determine large body movements (bed entry, bed exit) and determine whether the individual is awake or asleep during that epoch [33]. In addition, a low-band filter (0.3–10 Hz) is used to visualize respiratory movements, whereas a high-frequency filter (6–16 Hz) is used to visualize heart rate [48]. Upon enrollment, participants were given a uniquely identified Emfit device and were instructed to place the mat under their mattress at the position where their chest would most likely be during sleep. Devices connected to participants’ home internet, through cellular connection, or by internet with an included hotspot.

Sleep scoring using the Emfit monitor

A detailed protocol outlining the process for sleep scoring using the Emfit monitoring device has been previously published by our group [33]. Briefly, minute-to-minute activity data obtained from the monitor was used to determine bed entry and exit times, sleep start and end times, time in bed (TIB), total sleep time (TST), sleep efficiency (SE), sleep onset latency (SOL), and WASO. Bed entry and exit times were defined as the first and last 1-min epochs showing movement on the Emfit sensor, respectively. Sleep start was defined as the first 10-min period of immobility following bed entry. Sleep end was defined as the last 10-min period of immobility before bed exit. TIB was defined as the difference between bed exit and bed entry. TST was defined as the difference between sleep end and sleep start. SE was calculated as [(TST/TIB) × 100%]. SOL was calculated by subtracting bed entry time from sleep start time. WASO was defined as the number of 1-min epochs spent awake between sleep start and sleep end. The start of each 24-h period was set to 12:00 p.m. on the day of sleep (720 min).

Statistical analysis

Descriptive statistics were used to summarize the demographic characteristics of the cohort, stratified by concussion status. Shapiro–Wilk tests were used to evaluate the normality of each variable’s distribution. Two-sample independent t-tests (for continuous variables) and Fisher’s exact tests (for categorical variables) were used to compare demographic variables between the control and concussion cohorts. One participant from the concussion cohort did not report their height and was excluded from body mass index (BMI) calculation.

Primary analysis

The primary outcomes of this study were reported sleep quality (PSQI, ISI scores), reported neuropsychiatric symptom burden (PROMIS scores), and measured objective sleep parameters (TIB, TST, SE, SOL, WASO). Wilcoxon-rank sum tests were used to compare scores for the PSQI, ISI, and PROMIS evaluations between cohorts. Pairwise Pearson correlations of baseline subjective sleep quality measures and patient-reported neuropsychiatric symptoms were computed, and a Bonferroni p-value adjustment was implemented to account for multiple hypothesis tests. Cross-sectional estimates for objective sleep parameters were derived from 7 to 30 nights of recorded sleep data using regression with mixed effects to account for variability across multiple study nights. To compare average objective sleep parameter values over the study period, a series of regressions using concussion status were fit. To account for inter-individual variability due to repeated measurements within a patient, cluster-robust standard errors were utilized in calculating the p-value of the comparison. Estimates for the effect of concussion on each sleep parameter were then adjusted for age and sex.

A series of linear mixed effects models were also fit to best describe how measures of objective sleep parameters differed with respect to concussion status and self-reported outcomes throughout the study period. Due to collinearity between subjective measures, individual models were developed for each self-reported outcome. First, a random intercept model was fitted and then we examined whether the addition of an autoregressive-1 covariance structure, relating the remaining relationship between sequential measurements after accounting for person, improved model fit via a likelihood ratio test. Finally, we created a series of models testing the fixed effect of concussion, subjective sleep, and neuropsychiatric indices, and their interaction. Significance of fixed effects was determined by a Wald t-test p-value less than α < 0.05. The interaction term was excluded for model parsimony if it was insignificant. Age and sex were included as covariates in all models. A model R² value was calculated to estimate the proportion of total outcome variance explained by the fixed effects by taking the squared correlation of the linear predictor and outcome values.

Secondary analysis

Our secondary analysis explored the agreement between subjective and objective sleep parameters. The PSQI asked participants to estimate their sleep over the last month. Respondents selected from multiple choice answers in 15-min increments in response to questions, “How many hours of actual sleep did you get at night?” and, “How many hours were you in bed?” These answers were used as self-reported TST and TIB, respectively, and combined to calculate SE. Measured TST, TIB, and SE were calculated by averaging data across each individual’s number of study nights. Pearson correlation tests were used to compare the agreement between measured and self-reported TST, TIB, and SE for each cohort. We also constructed Bland–Altman plots by cohort and calculated the mean difference and associated 95% limits of agreement (LOA) for each measure [49].

Subjective–objective discrepancy was calculated by subtracting the value for the self-reported parameter from the measured parameter. Discrepancy was then compared between cohorts using either a Wilcoxon rank-sum test or an independent t-test depending on the result of a Shapiro–Wilk test for normality. Pairwise Pearson correlation coefficients were computed to evaluate the association between subjective–objective discrepancy and SCAT symptom severity scores for all three parameters. We then performed multivariate linear regression to assess the relationship between sleep discrepancy and SCAT scores adjusting for age and sex including analytic weights proportional to the variance of the objective measure [(nights recorded)/(SD_obj)²].

Statistical power and significance

Due to the small sample sizes of each cohort, a post-hoc power analysis was performed via simulation using estimated model parameter values to determine the ability of future studies to detect similar effect sizes of subjective score and concussion status on objective sleep parameters.

All statistical analyses in this study were performed using STATA/SE 18 (StataCorp), with an a priori alpha level of 0.05. Simulations were carried out in R version 4.4.1 (R Core Team).

Results

Study population

Demographic and injury information are presented in Table 1. There were no significant differences in age, sex, BMI, race, and ethnicity between the two cohorts. There was a significantly higher prevalence of premorbid head trauma and headaches within the concussion cohort and of attention deficit disorder within the control cohort. There were no significant differences in the frequencies of premorbid anxiety, depression, or sleep medication use between groups. Individuals with concussion were enrolled during the subacute phase of injury (median = 95 days, range = 3–52 weeks). The median time between injury and completed SCAT was 55 days (inter-quartile-range [IQR]: 25–85 days). The mean SCAT symptom severity score was 41.7 (95% CI: 30.5–52.9). Approximately half (52 per cent) of the concussion cohort reported a previous history of one or more concussions prior to the index event that prompted enrollment in our study.

Table 1.

Study population demographics and injury characteristics for individuals with concussion and controls

Variable	Concussion (n = 31)	Control (n = 19)	P-value
Age at enrollment (years)	27.8 (10.8)	25.4 (8.7)	.42
Sex			.55
Male	10 (32%)	8 (42%)
Female	21 (68%)	11 (58%)
BMI (kg/m2)	26.0 (5.6)	23.9 (5.1)	.19
Ethnicity			.63
Hispanic/Latino	2 (6%)	2 (11%)
Not Hispanic/Latino	29 (94%)	17 (89%)
Race			.43
Asian	1 (3%)	1 (5%)
Black/African American	1 (3%)	1 (5%)
White	23 (74%)	16 (84%)
Multiracial	5 (16%)	0 (0%)
Other	1 (3%)	1 (5%)
Premorbid conditions
Head trauma	26 (84%)	1 (5%)	<.001***
Headaches	14 (45%)	0 (0%)	<.001***
Anxiety	8 (26%)	7 (37%)	.53
Depression	9 (29%)	6 (32%)	1.00
ADD/ADHD	0 (0%)	4 (21%)	.017*
Diabetes	1 (3%)	0 (0%)	1.00
Epilepsy	0 (0%)	0 (0%)
Substance abuse	0 (0%)	0 (0%)
Other	3 (10%)	0 (0%)	.28
Reported sleep medication use	8 (26%)	4 (21%)	1.00
Age at injury (years)	25.3 (16.4–37.3)
Time injury to enrollment (days)	95 (51–206)
SCAT Symptom Severity Score	41.7 (29.5)
Time injury to SCAT (days)	55 (25–85)
Previous number of concussions
0	15 (48%)
1	11 (36%)
2 or more	5 (16%)
Cause of trauma
Traffic MVC	4 (13%)
Bicycle MVC	3 (10%)
Water MVC	1 (3%)
Unclassified MVC	1 (3%)
Accidental fall	12 (39%)
Other accident	7 (22%)
Unknown/not reported	3 (10%)

Summary statistics for demographics and injury characteristics by cohort.

^* p < 0.05, ^**p < 0.01, ^***p < 0.001; Mean and standard deviation reported for normally distributed continuous variables; Median and interquartile range reported for non-normally distributed continuous variables; Independent t-test used for comparison of continuous variables and Fisher’s exact test used for comparison categorical variables; BMI = body mass index; ADD/ADHD = attention deficit disorder/attention deficit hyperactivity disorder; SCAT = Sports Concussion Assessment Tool; MVC = motor vehicle collision.

Individuals with concussion reported significantly worse sleep than controls but did not differ in neuropsychiatric symptom burden

We first compared self-reported sleep quality, affective, and cognitive functions in our cohorts upon enrollment in the study. Individuals with concussion scored significantly higher than controls on the PSQI and the ISI (p = .020 and p < .01, respectively; Figure 2, A). Analyses by component for both surveys can be found in Tables S1 and S2. There were no differences in self-reported measures of anxiety, depression, stress, or cognitive function between groups, assessed via the PROMIS questionnaires (Figure 2, B; Table S3).

Figure 2.

Self-reported sleep quality and neuropsychiatric outcomes in concussion and control cohorts. Box plots of raw scores by cohort on the Pittsburgh Sleep Quality Index (PSQI) and Insomnia Severity Index (ISI) questionnaires (A) and the Patient-Reported Outcomes Measurement Information System (PROMIS) depression, anxiety, stress, and cognitive function questionnaires (B). Y-axis ranges are set to total score ranges for each questionnaire. Higher scores on the PSQI and ISI signify worse perceived sleep. Higher scores on the PROMIS depression, anxiety, and stress questionnaires signify greater severity of these symptoms, whereas lower scores on the PROMIS cognitive function questionnaire signify poorer cognitive function. Wilcoxon rank-sum tests were used to test for significant differences in scores between concussed and control cohorts. Stars represent p-values for these tests at ^*p < .05 and ^**p < .01 significance levels.

Subjective sleep quality correlated with neuropsychiatric symptoms independent of concussion status.

We next performed pairwise Pearson correlations to quantify the associations between our subjective variables. Figure 3 depicts the Bonferroni-adjusted Pearson pairwise correlation coefficients between the PSQI, ISI, and PROMIS questionnaires. A series of tests comparing two correlations using Fisher’s Z-transformation showed that the correlations between these measures did not differ significantly based on group status; thus, the groups were combined for this analysis. PSQI and ISI scores were associated with each other and with greater burden of depression, anxiety, and stress, and poorer cognitive function (Figure 3). Furthermore, the correlations between PSQI, ISI, PROMIS depression, and PROMIS anxiety scores remained statistically significant after adjusting for age and sex (Figure S1).

Figure 3.

Associations between subjective measures of sleep quality, affect, and cognitive function in study participants. Heat plot of Pearson correlation coefficients between total scores for Pittsburgh Sleep Quality Index (PSQI), Insomnia Severity Index (ISI), and Patient-Reported Outcomes Measurement Information System (PROMIS) depression, anxiety, stress, and cognitive function questionnaires. A difference in correlations test showed that the correlations between these measures did not differ significantly by cohort, thus both concussed and control cohorts were included in the calculations. Stars represent Bonferroni-adjusted p-values for each pairwise correlation at ^*p < .05, ^**p < .01, and ^***p < .001 significance levels.

Contactless sleep monitoring revealed differences in objective sleep parameters in individuals with concussion compared to controls

There was no significant difference in the number of included study nights per subject between groups (concussion_median = 18 nights [IQR: 14, 24]; control_median = 21 nights [IQR: 11, 25]; Wilcoxon rank-sum test, p = .73; Table S5). Group means for each objective sleep parameter were calculated using linear regression models with mixed effects for repeated measures and are described in Table S5. On average, individuals with concussion had significantly higher TIB than controls (p = .024), but there were no significant group differences in any of the other parameters. We repeated these models with age and sex included as covariates and found that individuals with concussion experienced longer SOL, longer TIB, and lower SE than controls (p < .05; Table 2). The beta coefficients for the group effect of concussion on SOL, TIB, and SE were 19.9 min, 42.3 min, and –3.1 per cent, respectively. There was no significant effect of concussion on any of the other objective sleep parameters (Table 2).

Table 2.

Age- and sex-adjusted beta coefficients for the effect of concussion on objective sleep parameters

Variable	β ± SEE	P-value (group contrast)
Time of bed entry (min)a	−26.45 ± 20.79	.203
Sleep onset (min)a	−6.74 ± 21.37	.752
Sleep end (min)*	11.95 ± 21.37	.576
Time of bed exit (min)*	15.46 ± 21.07	.463
Total sleep time (min)	18.71 ± 15.64	.232
Sleep onset latency (min)	19.91 ± 8.79	.023*
Time in bed (min)	42.33 ± 17.04	.013*
Sleep efficiency (%)	−3.21 ± 1.61	.047*
WASO (min)	6.99 ± 9.40	.457
WASOnorm (min)†	0.001 ± 0.018	.956

Estimated effects of concussion (β) on objective sleep parameters using linear mixed effects models with age and sex included as covariates; p-values are for the predicted difference in marginal means by group; ^*p < .05; SEE = standard error of estimate; WASO = wake after sleep onset.

^*Times of bed entry, sleep, wake, and bed exit were corrected for and aligned to a 24-h clock (0–1440 min).

^†WASO was normalized to TST.

Associations between self-reported outcomes and objective sleep parameters differed by cohort

To meet our primary aim, we created separate age- and sex-adjusted mixed effects models with participants’ PSQI, ISI, or PROMIS scores added as independent predictors (detailed results for all regression models can be found in Tables S6–S8). Adjustment for PSQI and ISI scores did not meaningfully change the effect of concussion on measured SOL and TIB but did attenuate the effect of concussion on SE to near zero (Table S6). Furthermore, a relationship was observed between sleep perception and WASO, such that a unit increase in subjective sleep score was associated with an expected increase of 5 WASO minutes for the control group only (p < .05; Figure 4; Table S6).

Figure 4.

Effect of self-reported sleep quality scores on wake after sleep onset (WASO) by cohort. Predicted effects of sleep questionnaire score on expected minutes of wake after sleep onset (WASO) by cohort (left: Pittsburgh Sleep Quality Index [PSQI]; right: Insomnia Severity Index [ISI]). Lines represent population-averaged predictions with 95% CIs for control and concussion, limited to the range of PSQI, and ISI scores observed in this population. Point sizes are proportional to sample size and standard deviation of objective measurement: [nights per subject/SD_WASO].

TIB remained significantly higher in concussed individuals by 37.5–41.0 min regardless of subjects’ scores on the PROMIS questionnaires (Table S7). However, the effect of mental health symptoms on SOL was less clear. Symptom burden appeared to be generally associated with increased SOL independent of concussion (Table S7). Furthermore, PROMIS-adjusted models predicted increased SOL by 15.4–18.4 min in the concussion group, but these associations were not always significant at p < .05 (Table S7). Of note, controlling for PROMIS anxiety scores revealed a significant group difference in WASO, such that the concussion group had an estimated 38.2 more minutes of WASO than controls at the minimum reported anxiety score (p = .02; Table S7). However, the effect of anxiety on WASO and the degree to which it is moderated by group status could not be ascertained at the prespecified significance level (Table S7). Again, the effect of concussion on SE was negligible after adjusting for PROMIS scores (Table S7).

Our models also revealed an effect of anxiety, depression, stress, and cognitive function on sleep times and wake times that did not differ significantly by group. Specifically, each point increases in raw scores on the PROMIS anxiety, depression, and stress questionnaires resulted in later expected sleep and wake times by an average of 3–7 min (Figure S2; Table S6). Each point decrease in raw score on the PROMIS cognitive function questionnaire, indicating poorer cognitive function, resulted in later expected sleep and wake times by 3.80 and 3.71 min, respectively (Figure S2; Table S6). There was no significant effect of any of these measures on TST (Table S6). Furthermore, no significant relationship was found between SCAT symptom severity scores and any of the objective sleep parameters (Table S8).

Individuals with concussion showed discrepancy between subjective and objective sleep characteristics

As a secondary analysis, we sought to characterize the agreement between subjective and objective sleep in this population. The PSQI instructs subject to answer questions based on their usual sleep habits over the previous month. Responses to questions 4a and 4b from our participant-facing version of the PSQI were used as reported measures of each individual’s TST (4a), TIB (4b), and SE ([4a/4b] × 100%). Figure 5 compares average objective TST measured by the Emfit device with the subjective TST reported on the PSQI. For controls, measured and reported TST were strongly correlated (Pearson correlation test, r = 0.52, p = .023; Figure 5, A). However, there was no significant agreement between measured and reported TST for the concussion cohort (r = 0.21, p = .273; Figure 5, A). Similarly, measured and reported TIB and SE were strongly and significantly correlated for controls (TIB: r = 0.58, p < .01; SE: r = 0.60, p < .01), but not for individuals with concussion (TIB: r = 0.35, p = .07; SE: r = −0.09, p = .64; Figures S3 and S4).

Figure 5.

Agreement between reported and measured total sleep time (TST) in concussion and control cohorts. (A) Scatter plot of average total sleep time (TST) measured by the Emfit device versus self-reported TST on the Pittsburgh Sleep Quality Index (PSQI) by cohort with a 1–1 reference line in gray. Point sizes are proportional to the standard errors of the objective measurement for each subject: [SD_TST/(nights recorded)^1/2]. (B) Box plot of TST discrepancy (measured TST − reported TST) by cohort. Independent t-test revealed a significant difference between groups (^*p < .05). (C and D) Bland–Altman plots comparing measured and self-reported TST for (C) controls and (D) individuals with concussion.

We next calculated sleep discrepancy by subtracting reported from measured TST, TIB, and SE. There was a significant difference in the magnitude of TST discrepancy between cohorts by independent t-test (p = .048; Figure 5, B) and by multivariate linear regression with adjustments for age, sex, and variability in study nights per subject (Table S8). Furthermore, Bland–Altman analysis that participants in the concussion group slept on average 0.86 h more than the amount reported on their PSQI ([95% LOA: −1.5, 3.2]; Figure 5, C), whereas controls had a mean difference of 0.20 h ([95% LOA: −1.7, 2.1]; Figure 5, D). In contrast, the mean difference between measured and reported TIB was 0.85 h for controls (95% LOA: −1.1, 2.8) compared to 0.53 h for individuals with concussion (95% LOA: −2.0, 3.1); thus, there was no significant difference in TIB discrepancy between groups (independent t-test, p = .36; Figure S3). Lastly, individuals with concussion underestimated their SE, whereas controls overestimated their SE (Wilcoxon rank-sum test, p < .01; Figure S4). The mean difference between measured and reported SE was −6.3 per cent for controls ([95% LOA: −18.4%, 5.8%]; Figure S4) compared to 3.5 per cent for individuals with concussion ([95% LOA: −25.5%, 32.5%]; Figure S4). Like TST, this group contrast was significant after adjusting for age, sex, and number of included study nights per subject (Table S9).

Given these group differences, we assessed the relationship between sleep discrepancy and injury severity within concussed individuals. A Pearson correlation test revealed an association between TST discrepancy and SCAT concussion symptom severity scores at time of injury (r = 0.39, p = .047) that remained significant after adjusting for covariates age and sex and including analytic weights for the number of study nights in a multivariate linear regression model (β_TSTdisc = 9.5/h, p_beta = .007, p_model = .022, Total R² = 0.34; Figure 6). SE discrepancy was also positively associated with SCAT symptom severity score by a Pearson correlation test (rho = 0.39, p = .047; Figure S5). However, it was not a significant predictor of scores after adjusting for the above confounders (β_SEdisc = 0.40/%, p_beta = .25). There was no association between TIB discrepancy and concussion symptom severity scores (r = −0.10, p = .636).

Figure 6.

Association between concussion symptom severity and discrepancy in reported and measured total sleep time (TST). Scatter plot of Sports Concussion Assessment Tool (SCAT) symptom severity score at time of injury versus TST discrepancy (measured TST − reported TST) for concussion cohort only. Point sizes are proportional to the standard errors of the objective measurement for each subject: [SD_TST/(nights recorded)^1/2]. Linear prediction with 95% CI was obtained via multivariate linear regression with age and sex as covariates and with the following analytic weights: [(nights recorded)/(SD_obj)²]; (β = 9.5/h, p_beta 0.007, p_model = 0.022, Total R² = 0.34).

Following our analysis of the association between concussion status and the discrepancy between subjective and objective sleep parameters, we performed post-hoc simulations using varying sample and effect sizes (as given by the standardized regression coefficient) to estimate the power to detect significant effects of these associations. These simulations revealed that the current sample size would be well-powered (power > 0.95) to detect a small-to-medium effect size (standardized regression coefficient of 0.35) on objective–subjective parameter discrepancy, and almost sufficiently powered (power = 0.78) to detect medium effect sized of concussion status (standardized regression coefficient of 0.50).

Discussion

Sleep–wake disturbances persist in individuals with subacute concussion

Our study compared self-reported sleep quality and neuropsychiatric symptoms with ballistocardiography-based sleep parameters in adolescent and young adult patients with subacute concussion. Individuals with concussion reported worse sleep than controls via the PSQI and ISI questionnaires. This finding parallels a robust body of literature that cites worse perceived sleep after TBI [3,9,21,22,24]. However, few studies include objective data, and those that do find varied results [9,13,25,26]. In our study, objective assessment with the Emfit revealed that concussed individuals spent more time in bed, took longer to fall asleep, and had poorer SE than controls. Furthermore, predicted effects of concussion on TIB were relatively stable (β = 35–50 min) across all adjusted and unadjusted models. In contrast, group effects for SOL and SE were predicted with less certainty, and the size of the effect of concussion on SE became negligible after adjusting for patient-reported health outcome scores. The potential effect of concussion on sleep latency and efficiency is likely still clinically significant as TIB is related to both SOL and SE. However, at least in this population, the effect of concussion on TIB was most compelling.

Our study design is most comparable to a prospective study of subjective and objective sleep in adolescents within 3–12 months of a concussion diagnosis conducted by Tham and colleagues in 2015 [50]. Like ours, participants in this study reported significantly worse sleep and were found to have poorer SE than controls. In contrast, their objective assessments revealed decreased SE by means of decreased TST, a discrepancy likely explained by study factors such as target population age, use of wrist actigraphy for sleep measurements, and prevalence of comorbid depressive symptoms.

In general, comparison across studies of post-TBI sleep is limited by variability in study duration, population demographics, and methodological considerations such as injury timeline, injury severity, and type of sleep monitoring technology [21,22,24,51,52]. We narrowed our sample to individuals within 1 year of injury given the paucity of data on sleep in subacute concussion and found that sleep disturbance persists beyond the initial weeks post-concussion, likely after individuals have sought initial care. However, there remains a lack of consensus regarding which objective sleep parameters, if any, best capture this disturbance [24,51,52]. More research in the subacute period is needed to improve our definition and treatment of post-concussive sleep disturbance.

Neuropsychiatric comorbidities modulate objective and subjective markers of sleep quality

We included a battery of neuropsychological symptom questionnaires to better understand the role of mental health in post-concussive sleep. There were no significant group differences in levels of anxiety, depression, stress, and cognitive function reported by the PROMIS evaluations at enrollment. Instead, increased burden of these symptoms was significantly associated with worse perceived sleep quality, regardless of concussion status. The relationship between sleep and mental health is well documented, even in the absence of concussion [15,31]. And while this could explain the observed associations between PSQI, ISI, and PROMIS, it is also possible that data from subjective questionnaires may be more representative of individual characteristics rather than group characteristics. In other words, certain participants may be more likely to score higher on all subjective measures, a phenomenon described as “response style effects” by Dowling and colleagues in 2016 [53].

As a result, we hypothesized that mental health symptoms might confound the observed group differences in objective sleep. Inclusion of PROMIS scores as covariates in our mixed effects models revealed several avenues through which neuropsychiatric symptoms and objective sleep may be related. Interestingly, adjustment for these scores had no significant effect on time spent in bed. Instead, we found an association between depression, anxiety, stress, and cognitive function and later sleep and wake times assessed objectively by the Emfit device, independent of group status. This agrees with the existing literature and suggests that circadian rhythm disorders might be more significantly affected by these symptoms alone and less related to concussion history [54,55]. Increased symptom burden also predicted longer SOL and poorer SE, though these associations were weak and we were unable to differentiate the effect of mental health and injury status on these sleep parameters with statistical certainty. Additional studies are needed to further clarify the nature of these relationships and how they moderate recovery after concussion.

This report underscores the independent relationships between mental health burden and both subjective and objective sleep after concussion. Furthermore, we’ve shown that these symptoms can influence objective sleep in situations where concussion doesn’t. As a result, though neuropsychiatric symptoms are well-known sequelae to brain injury, their effects may modulate or even obscure interpretation of post-concussive sleep data. Unfortunately, studies of post-concussive sleep often exclude people with comorbid mental illness [56–58]. Our study offers methodological improvements that will benefit the utility and validity of future studies.

Sleep discrepancy is a potential marker of concussion severity

Our mixed effects models revealed no significant relationship between perceived sleep quality and sleep time, wake time, SOL, TST, TIB, or SE. Furthermore, concussion status remained a significant predictor of SOL and TIB regardless of sleep perception scores. In contrast, we observed an interaction between self-reported sleep quality and concussion status in determining total minutes of WASO. When controlling for perceived sleep, the concussion cohort had an average of 37–39 additional minutes of WASO (Table S6). Interestingly, the interaction was such that WASO was positively correlated with sleep dissatisfaction in controls, but not in concussion. A possible interpretation is that restlessness during sleep contributes to worse perceived sleep quality in healthy individuals, whereas there may be a different mechanism behind self-reported sleep disturbance in individuals with concussion that cannot be captured by actigraphy. In fact, current literature suggests that subjective and objective sleep data measure independent dimensions of sleep and may be difficult to compare [25–30,59]. Our findings support subjective and objective sleep quality as unique outcomes, but also postulate concussion status as a nexus between these dimensions.

To explore this further, we compared sleep parameters calculated from answers in the PSQI to those measured objectively via the Emfit monitor. We found that self-reported and objective TST, TIB, and SE were correlated in controls, but not in individuals with concussion. Moreover, discrepancies between measured and reported SE and TST were significantly higher in the concussion group and correlated with concussion symptom severity scores. After adjusting for age, sex, and variability in total study nights, we saw a positive association between underreported TST at the present time point and higher SCAT scores at the time of injury. We are unsure why we did not observe a group difference in TIB discrepancy or why SE discrepancy could not predict SCAT scores in our adjusted models. However, it could be that individuals with concussion can accurately recall TIB but are discrepant in their reporting of TST, and thus, with less certainty, SE.

From these results, we hypothesize that concussion status plays a role in the distinction between subjective and objective dimensions of sleep observed here and in previous studies. It could be that a mechanism of concussion contributes to concussed patients feeling like they sleep less than what is measured by actigraphy. More importantly, this mechanism may be exacerbated by increased severity of injury. Winer and colleagues proposed a similar concept in 2021 after finding an association between “sleep misperception,” executive functioning, and burden of amyloid beta in a population of cognitively healthy adults [60]. While we did not examine the relationship between sleep discrepancy and executive function in our population, we did find a correlation between concussion symptom severity scores at time of injury and self-rated cognitive function scores at enrollment. In this context, and given the findings of Winer et al., sleep discrepancy might be a link between brain injury and the trajectory of cognitive health over a lifespan.

Our findings do not negate self-reported sleep disturbance as an important and valid symptom of concussion that requires clinical attention. However, if the relationship between subjective and objective dimensions of sleep provides insight into concussion severity, clinical treatment based solely on self-report sleep questionnaires might consequently be underinformed. Ultimately, sleep discrepancy is a new concept that warrants validation but offers potential insights into the physiology of sleep after concussion.

Limitations

While our study is novel in its comprehensive approach to sleep phenotyping, it was exploratory in nature and therefore has considerable limitations. First, our concussed cohort was recruited from both health record advisory tools as well as from a local concussion clinic. Recruitment from the clinic may have created selection bias that we did not adjust for. For example, it is possible that sleep discrepancy after concussion is limited to those that seek care for persistent symptoms. For these reasons, our statistical analyses cannot be used for hypothesis testing. Future studies should adjust for recruitment mechanism as well as time from injury to enrollment.

Furthermore, it is worth noting that while our cohorts were comparable in age and sex, our sample was small and homogenous in other demographic characteristics such as race and ethnicity. There was also a significantly higher prevalence of attention deficit disorder in controls compared to individuals with concussion. However, if we assume that stimulant medication use reduces objective sleep quality, it is more likely that this led to an underestimation of the effect of concussion on sleep parameters as opposed to an overestimation. Still, future studies should address this potential confounder and others such as sleep medication use, daytime activity level, fatigue, occupation, and socioeconomic status.

Roughly, half of this cohort had a history of previous concussion before enrollment. Since we did not stratify our cohort by this history, we do not know if disturbed sleep is mediated by concussion history and cannot speak to the effect of multiple concussions. We also must acknowledge the variability within a TBI diagnosis and that our findings are likely not generalizable to populations of moderate and severe TBI. Even within concussion alone, it is difficult to capture or control for heterogeneity in injury presentation, predominant symptom cluster, and intraindividual variability in sleep patterns. We chose to address this variability by use of mixed effects modeling. Finally, there are some limitations inherent to our sleep scoring device. Like actigraphy, the Emfit allows for easy in-home estimation of objective sleep parameters (TST, SOL, SE), but it ultimately cannot characterize sleep architecture with the same depth and precision as polysomnogram.

These challenges are common to most studies of sleep and concussion and likely limited the amount of variance explained by the fixed effects in our models. Thus, even complex multilevel modeling leaves much left to be understood about sleep after concussion. We focused our study on subacute concussion because of its increasing health risk and because there is limited data regarding sleep during this phase of injury, but more research is needed to understand how sleep disturbances evolve over time within this population. When possible, baseline pre-concussion sleep data should be included to identify changes in sleep patterns and quality that occur as a result of injury.

Conclusion

We demonstrate a multidimensional study of subjective and objective sleep in individuals with subacute concussion. Our results suggest that patient-reported sleep perception may not equate to objective sleep–wake disturbances in this population. Instead, reported and measured sleep discrepancy may provide insight into the initial severity of injury. This hypothesis requires further validation, but our methodology underscores considerations for future research. Ultimately, post-concussive sleep disturbance will be best defined by longitudinal studies that include self-reported sleep quality, objective sleep parameters, and measures of psychiatric comorbidity.

Institution where work was performed: Oregon Health & Science University.

Author contributions

Caitlyn Emma Wong (Data curation [supporting], Formal analysis [lead], Visualization [equal], Writing—original draft [lead], Writing—review & editing [lead]), Madison Luther (Conceptualization [equal], Data curation [lead], Project administration [lead], Writing—original draft [supporting], Writing—review & editing [supporting]), Avery Scatena (Data curation [supporting], Project administration [supporting], Writing—review & editing [supporting]), Seiji Koike (Formal analysis [supporting], Writing—original draft [supporting], Writing—review & editing [supporting]), Melissa Novak (Conceptualization [equal], Methodology [supporting], Resources [supporting], Writing—review & editing [supporting]), Jonathan E. Elliott (Conceptualization [supporting], Methodology [supporting], Writing—review & editing [supporting]), Jeffrey Iliff (Conceptualization [supporting], Methodology [supporting], Writing—review & editing [supporting]), Miranda M. Lim (Conceptualization, Methodology, Writing—review & editing [supporting]), Emily Kosderka (Conceptualization, Methodology, Resources, Writing—review & editing [supporting]), Juan Piantino (Conceptualization, Funding acquisition, Investigation, Methodology, Project administration, Resources, Supervision [lead], Writing—original draft [supporting], Writing—review & editing [supporting]). Juan Piantino takes full responsibility for the data, the analyses and interpretation, and the conduct of the research; has full access to all the data; and has the right to publish all data separate and apart from any sponsor.

Funding

Research reported in this publication was supported by the National Heart, Lung, and Blood Institute (K23HL150217-05) as well as the National Center for Advancing Translational Sciences of the National Institutes of Health (TL1TR002371). The content is solely the responsibility of the authors and does not necessarily represent the official views of the National Institutes of Health.

Disclosure statement

Financial disclosure: Drs. Piantino, Iliff, and Lim serve on the Scientific Advisory Board for Applied Cognition, Inc.

Non-financial disclosure: None.

Data availability

The data underlying this article will be shared on reasonable request to the corresponding author.