WHAT IS ALREADY KNOWN ON THIS TOPIC
WHAT THIS STUDY ADDS
This study shows that postconcussion recovery in children and adolescents benefits from a dynamic balance of behaviours: more sleep and less sedentary time in the early days, consistent engagement in moderate-to-vigorous activity throughout, and the addition of more light activity as recovery progresses. These findings highlight that it is not one behaviour alone, but their daily distribution over time, that is associated with recovery.
HOW THIS STUDY MIGHT AFFECT RESEARCH, PRACTICE OR POLICY
Clinicians seeking to improve postconcussion recovery in children and adolescents should adopt an integrated movement behaviour approach that balances rest and activity. Specifically, when compared with the sample average, optimal patterns of rest involved more sleep and less sedentary time in the first days, shifting to relatively less sleep with more sedentary time in the first week, and returning to more sleep with less sedentary time in the second week. Optimal activity in the first week postconcussion included lower overall physical activity volume, driven by lower than average light physical activity (LPA), while still achieving higher intensity physical activity through accumulation of above average moderate-to-vigorous physical activity (MVPA), followed by promoting both intensity and volume (ie, both higher LPA and MVPA) of physical activity in the second week.
Introduction
Concussions are a major global health concern, with countries such as Canada reporting them as the most common injury in children and adolescents (0–17 years).1 2 While paediatric postconcussion symptoms typically resolve within 2 weeks of injury, approximately 35% experience persisting symptoms after concussion (PSAC), which can greatly impair their daily functioning.3–5 The risk of PSAC is a strong catalyst for developing and refining best-practice concussion management protocols. For instance, engaging in early physical activity, reducing prolonged sedentary behaviours and promoting quality sleep are strategies that help mitigate postconcussion symptoms.6–9
Concussion management recommendations for physical activity, sedentary behaviour and sleep (collectively termed movement behaviours) in isolation mirror past public health guidelines that treated these behaviours as independent exposures.10 However, modern interpretations understand that movement behaviours are interdependent.10 For instance, when a child spends an extra hour engaged in physical activity, there is one less hour in the day available to be spent engaged in sedentary behaviour or sleep. Not only are movement behaviours an important accounting consideration when determining the distribution of movement behaviours in a 24-hour day, but they also have interacting influences on health.11 For these reasons, paediatric public health guidelines around the world have retreated from isolated behaviour recommendations and towards the integration of all movement behaviours in a day.12–15 Aligned with this new paradigm, compositional data analyses (CoDA) have been applied to appropriately analyse the codependent nature of movement behaviours.16
While CoDA is an analytically appropriate and parsimonious means of understanding collective movement behaviours, only one study has applied it to paediatric concussions.17 Brayton et al found no association between movement behaviour composition and concussion recovery time in 33 adolescents, though the small sample size limits generalisability. Further, their study focused solely on the benefits of improving one movement behaviour relative to the overall composition. However, CoDA can identify the optimal distribution of movement behaviours, providing precise recommendations for balancing all movement behaviour in a 24-hour day. This goes beyond vague recommendations to ‘do more or less’ of an individual behaviour, serving as a powerful tool for clinical concussion management. Accordingly, using data from a large, multisite study, we aimed to determine the optimal distributions of daily movement behaviours associated with lower paediatric postconcussion symptom burden and risk of PSAC classification at 2 weeks postinjury.
Methods
Study design and participants
This study was a secondary analysis of the Pediatric Concussion Assessment of Rest and Exertion (PedCARE, NCT02893969) data.18 The PedCARE study was a randomised clinical trial that recruited participants in three paediatric emergency departments (ED) in Ontario, Canada, from March 2017 to December 2019.6 Eligible participants sustained a concussion within 48 hours of presenting to the ED and were 10.00–17.99 years. The Zurich and Berlin consensus statements for concussions were used to define concussions.7 8 Additionally, concussions were confirmed using an adapted version of the US Centers for Disease Control tiered framework.19 Participants were excluded if they had a Glasgow Coma Scale score ≤13, trauma-related abnormality on brain imaging (if assessed), intensive care unit admission, intubation, neurosurgical intervention, hospitalisation due to a multisystem injury, communication difficulties due to severe neurological development delays, intoxication at ED presentation, absence of trauma history as the primary event, previously enrolled in the study, insurmountable language barriers or inability to complete follow-ups.
In the ED, participants completed questionnaires assessing demographic information, personal health status, mental health history and injury characteristics. The Balance Error Scoring System was used to assess participants’ balance.20
Preinjury and postinjury concussion symptom severity were assessed with the Health and Behaviour Inventory (HBI) and the retrospective HBI. The HBI is a valid and reliable 20-item assessment, rated on a 0 (‘never’) to 3 (‘often’) scale.21 It yields cognitive (range: 0–27), somatic (range: 0–33) and total (range: 0–60) scores, with higher scores indicating a greater symptom burden (ie, more and/or more severe symptoms). Consistent with established paediatric concussion research protocols,22–24 the retrospective HBI was completed by the participants’ parents/guardians in the ED to assess preinjury symptoms, while participants completed the HBI in the ED and at 1, 2 and 4 weeks postinjury using the Research Electronic Data Capture (REDCap)25 database or via phone. This approach reflects developmental considerations regarding recall accuracy and informant reliability.
Participants were randomised to return to physical activity either at 72 hours (intervention) or once asymptomatic (control). Previous analyses showed no differences in moderate-to-vigorous physical activity (MVPA), light physical activity (LPA) or sedentary behaviour between the intervention and control groups.6 Further, preliminary t-tests revealed no group difference for sleep duration (p=0.4). Consequently, we pooled both groups to examine overall movement behaviour patterns. All participants received standard information on concussion education, return to learn and return to physical activity. Participants received a waist-worn Actical accelerometer (Phillips Respironics, Oregon, USA) programmed to record movement in 1 min epochs, with instructions to wear it for 14 consecutive days except during water activities (eg, showering, swimming). Accelerometers are a valid and reliable tool for assessing movement behaviours.26 Participants also completed a daily log via REDCap to record their wake and sleep times.
This manuscript followed the Strengthening the Reporting of Observational Studies in Epidemiology reporting guideline.
Accelerometer data processing
Complete PedCARE accelerometer data processing procedures have been previously published.6 27 Briefly, waking behaviours were classified as MVPA (≥1500 counts/min), LPA (100–1499 counts/min) and sedentary (≤99 counts/min). Sleep was identified via visual inspection and sleep logs. Days with <10 hours of waking wear time or movement values±3 SD were excluded. Only participants with complete data for all four movement behaviours were included.
Outcome measures
Total, cognitive and somatic symptom scores were derived from the HBI completed at the 2-week post-ED follow-up. PSAC was classified by calculating reliable change z-scores. Specifically, the parent-reported retrospective rating of preinjury total symptoms was compared with the child’s self-reported rating of symptoms reported at 2 weeks using a regression formula developed in children with orthopaedic injuries22:
Reliable change z-score=postinjury child total score–(6.352+[0.476×retrospective parent preinjury total score])/9.597.
A reliable change z-score ≥1.65 was defined as PSAC, indicating symptoms increased significantly more than expected compared with retrospective ratings.
Statistical analysis
Participants typically require several days (eg, ≥4 days26) of accelerometer wear to ensure reliable analysis of movement behaviours. However, recent concussions may disrupt movement behaviour consistency over the 13 days; thus, the stability of accelerometer-measured movement behaviours was examined. Specifically, paired Wilcoxon and sign tests were conducted comparing each possible pair of days for each movement behaviour (eg, day 1 sleep compared with day 11 sleep), with Bonferroni-adjusted p values ≤0.00064 (0.05 divided by 78 comparisons) indicating comparisons were significantly different. If tests indicated temporal differences across days (suggesting instability due to concussion), the typical requirement for multiple days of accelerometer data was waived. Otherwise, an accelerometer time-wear of ≥4 days was required.26
After testing this stability assumption, data were used to build compositional regression models. CoDA of movement behaviours involves log-ratio transformations that account for the constrained and co-dependent nature of time-used data to create a set of movement behaviour composition variables. Without this transformation, including all movement behaviours in a single regression model would result in multicollinearity due to the constant-sum constraint. For each day of accelerometer wear (ie, days 1–13) and HBI outcome variables (ie, cognitive, somatic and total HBI score) at the 2-week follow-up, compositional robust regression models28 were built with MM estimation and Koller and Stahel’s29 settings within the robustbase R package30 that included movement behaviour isometric log-ratio transformed compositions. Covariates were selected to align with previous PedCARE studies (ie, age, sex, randomisation group, 5P clinical risk score,5 preinjury comorbidity (ie, anxiety, depression, sleep disorder, other mental health disorder, learning disabilities, attention disorder and other developmental disorder), and the retrospective and ED total HBI scores) and included as fixed effects in all compositional regression models.6 27 In line with previous research,28 31 three separate models were created with compositional movement behaviours: (1) removed, (2) included as linear terms and (3) included as squared terms. The three models were compared by examining the fit of nested regression models with robust analysis of deviance tests.28 When the square or linear models did not improve model fit compared with the no movement behaviour model, optimal compositions of movement behaviours were not examined for that day. Including non-significant models in predictions would have introduced estimates that are not meaningfully different from the sample mean, potentially regressing optimal movement behaviours to the average movement behaviours. Otherwise, linear or square models were selected for subsequent analyses based on their improved fit. For PSAC, generalised linear models were built to conduct logistic regression, mirroring the linear regression models, but excluding retrospective total HBI as a covariate since it is used to determine PSAC. Model fit of nested models was assessed with chi-square tests. Partial R2, or squared semipartial correlation, was calculated for each significant model by comparing R2 values from the base model (no movement behaviour) with the final model (with movement behaviour composition), to estimate the partial effect size attributed to the composition of movement behaviours alone.32 33 Significant regression models were visually inspected to examine linearity assumptions.
General steps to calculate the optimal movement behaviours in a 24-hour day include: (1) using significant regression models to predict the outcome values for a range of hypothetical movement behaviour combinations, (2) filtering the predicted outcome values to retain a proportion representing the optimal outcome values and (3) describing the movement behaviours associated with this range of optimal outcome values. Specifically, the range of all possible values within 2 SDs of the compositional mean of each movement behaviour was calculated in 10 min increments, and combinations of movement behaviours adding to 1440 min (or a full 24-hour day) were used to predict the effect on the outcome for each composition in the significant models. The bottom 5% of the predicted outcomes were retained to identify the predicted movement behaviour compositions associated with the greatest reductions in post-concussion symptoms—or the optimal movement behaviours.28 31 The optimal movement behaviours for each day were visualised with smoothed lines added for ease of interpretability. Smoothed highlight lines were not in predictions, only to illustrate trends. All analyses were conducted in RStudio V.4.3.3,34 using key packages, including robustbase
30 and compositions.35 Sample size was determined by the parent study, which was powered to detect a clinically meaningful change in HBI and group differences across timepoints.6 The current secondary analyses were powered to detect medium effect sizes.
Patient and public involvement
No patients or members of the public were involved in the design or interpretation of this study.
Equity, diversity and inclusion statement
Our study included children and adolescents who had sustained a concussion, regardless of any sociodemographic characteristics (eg, sex, gender, race/ethnicity). The authorship group consists of three women and five men, including junior, mid-career and senior researchers.
Results
The analytical sample included 259 participants, and was 45% female, with a median age of 13.1 (IQR: 11.7, 14.7) years, and 10.0 (IQR: 6.0, 12.0) days of accelerometer wear (online supplemental file 1, eFigure 1; table 1). The number of participants for each day ranged from 160 (62% of analytical sample; day 12) to 192 (74% of analytical sample; day 9; online supplemental file 1, eTable 1). Generally, over the 13 days, average sleep and sedentary behaviour decreased, while LPA and MVPA increased. While some discrepancies were observed between Bonferroni-adjusted paired Wilcoxon and sign tests (eg, day 3 and 4 MVPA differed only in the sign test), both generally showed similar patterns. Specifically, MVPA demonstrated the strongest lack of stability based on the number and pattern of significantly different days. Notably, the earlier days of MVPA (eg, day 2) were significantly different from all other days of data, whereas days 6–13 were not significantly different (figure 1). The demonstrated lack of stability warranted not using standard accelerometer wear definitions (eg, ≥4 days26 of data needed for each participant) used in research among healthy individuals.
Figure 1
Stability of movement behaviours over 13 days. Red indicates no difference when comparing a movement behaviour duration from 1 day to another. Teal indicates a significant difference for the paired sign test alone, purple indicates a significant difference for the paired Wilcoxon test alone and green indicates both the paired sign and Wilcoxon tests demonstrated significant differences between movement behaviour durations occurring on different days. All analyses applied post hoc Bonferroni adjustments; recalculated p values ≤0.00064 (0.05 divided by 78 comparisons) were considered significant. LPA, light physical activity; MVPA, moderate-to-vigorous physical activity.
Table 1
Sample characteristics
Day-by-day regressions for days 1, 3, 9 and 11 showed no associations between movement behaviours and HBI. Thus, these days were excluded from predicting the optimal composition of movement behaviours. Across the significant models, the optimal compositions of movement behaviours were associated with concussion symptoms (table 2; figure 2) as low as 2.9 for cognitive HBI symptoms (day 6 movement behaviours), 1.3 for somatic HBI symptoms (day 10 movement behaviours), 4.5 for total HBI symptoms (day 6 movement behaviours) and 0.0 probability of PSAC (all significant days). Generally, the optimal pattern of MVPA was above average for all days (figure 3; table 2). Sleep was above average on day 2, then dropped below average for the remaining days, except for days 8 and 10 where it was again above average. Sedentary behaviour appeared to follow the inverse pattern of sleep, starting below average at day 2, rising above average for most other days except days 8 and 10, where it again dropped below average. For LPA, patterns were less clear, with levels possibly responding to the optimisation of other behaviours. However, on days 8 and 10, LPA was well above average. Partial R2 values for significant analyses ranged from 0.07 to 0.09 (small-to-medium effect)32 33 for cognitive HBI symptoms, 0.01–0.08 (small effect)32 33 for somatic HBI symptoms, 0.07–0.09 (small-to-medium effect)32 33 for total HBI symptoms, and 0.09–0.22 (medium effect)32 33 for probability of PSAC (online supplemental file 1, eTable 2).
Figure 2
Predicted Health Behaviour Inventory (HBI) scores at follow-up associated with optimal movement behaviours. Solid dots represent the predicted HBI scores (red=cognitive, green=somatic, blue=total) at 2-week follow-up for the optimal movement behaviours on that day. Solid lines form when many solid dots overlap. The dashed line represents the total sample’s mean HBI score at 2-week follow-up, regardless of movement behaviours. Only days with statistically significant associations between movement behaviours and HBI scores are displayed (n=7). Cog, cognitive; Soma, somatic, Tot=total.
Figure 3
Optimal movement behaviours from days 1 to 13. Movement behaviours are compositions scaled to add to 24 hours. Movement behaviours are categorised as light intensity physical activity (LPA, red), moderate-to-vigorous intensity physical activity (MVPA, green), sedentary behaviour (sedentary, blue) and sleep (purple). Dots represent the arithmetic mean (with movement behaviours scaled to sum to 24 hours), bars represent the full range of potential optimal values, dashed lines are the full sample arithmetic means (with movement behaviours scaled to sum to 24 hours), and highlight lines are smoothed lines from general additive models (GAM) using the geom_line(stat=‘smooth’, method=’gam’, formula=y (hours/day) ~ s(x (day), bs=‘cs’, k=5), …) function from the ggplot2 R package. Smoothed highlight lines were not used during predictions and were only used to help visualise trends in data. Only days with statistically significant associations between movement behaviours and postconcussion symptoms are displayed (days 1, 3, 9 and 11 did not show significant associations and are, therefore, not shown).
Table 2
Optimal movement behaviours and predicted postconcussion symptoms at the 2-week follow-up
Discussion
This study showed that specific daily movement behaviour compositions were associated with optimal concussion symptom recovery at 2 weeks postinjury in children and adolescents. However, these optimal behaviours did not follow a simple pattern, which is not surprising given the observed lack of temporal stability in postconcussion movement behaviours. Generally, in the first days post-paediatric concussion (ie, day 2), patients should benefit from rest characterised as above-average sleep (ie, 11.5 hours/day) and below-average sedentary behaviour (ie, 8.5 hours/day). While rest remains important over the 13-day period, sleep and sedentary behaviour trade off over time. In the first week, above-average sedentary behaviour (eg, 11.7 hours/day, day 4) and below-average sleep (eg, 8.0 hours/day, day 4) are optimal. In the second week, this pattern reverses, with above-average sleep (eg, 10.9 hours/day, day 10) and below-average sedentary behaviour (eg, 6.6 hours/day, day 10) being optimal. For physical activity, above-average MVPA (eg, 0.6, 1.5, 1.1 hours/day, days 2, 7, 13) was generally optimal throughout, whereas above-average LPA (eg, 5.5 hours/day, day 10) showed the most benefits in the second week. Together, these results suggest clinicians should promote lower-volume (ie, less LPA) and higher-intensity (ie, MVPA) physical activity in the first week postconcussion, followed by promoting both intensity and volume of physical activity in the second week.
The recommendation for initial rest (<24–48 hours) is standard practice for concussion symptom management.9 However, some argue that early resumption of physical activity (>24–48 hours) is optimal for symptom management.6 27 These two perspectives are not at odds. In fact, the optimal distribution of movement behaviours for reducing postconcussion symptoms in this sample included both early rest (combined sleep and sedentary behaviour) and higher-than-average levels of MVPA throughout. Further, the benefits of these daily optimal movement behaviours underscore the importance of considering all movement behaviours within a 24-hour day. For instance, optimal movement behaviours were associated with total HBI scores of 4.5–8.3 compared with the average of 15.5. Additionally, the probability of PSAC classification in the overall sample was 0.12, which is considerably higher than the 0.00–0.02 associated with the optimal movement behaviours in six of the 13 days. While these findings are a promising first step towards tailoring daily movement behaviour recommendations for postconcussion symptom management, the optimal patterns of movement behaviours presented in this study should be reevaluated in further large and diverse samples.
Strengths and limitations
A major strength of this study was the examination of the stability of post-concussion accelerometer data patterns in children and adolescents. Findings indicated a lack of stability in mean movement behaviour patterns postconcussion. Future research on movement behaviours and concussion should consider this by not over-relying on the standard criteria, developed for healthy populations, that require several days of accelerometer data. Another strength of this study is the relatively large sample size. While the sample size helps overcome the limitations of previous research,17 current findings still require replication in larger and more diverse samples.
One limitation of this study is the use of different reporters for pre-injury versus follow-up HBI assessments. While this approach follows established paediatric concussion research practices,23 24 it may introduce reporter-related variability given the moderate parent–child agreement typically observed for concussion symptoms.21 36 37 The cross-sectional nature of our movement behaviour data is a limitation. Although optimal movement behaviours on isolated days were independently associated with lower concussion burden at 2 weeks, the cumulative effects over multiple days could not be assessed. CoDA application to movement behaviour research is a developing area, and no study has yet used time-series CoDA to analyse 13-day behaviour patterns. Future research should develop techniques to examine both daily (eg, early morning vs late afternoon activity) and longitudinal (e.g, 13-day sleep trends) temporality in movement behaviours. Another limitation was the use of complete case analysis rather than techniques capable of handling missing data, such as multiple imputation. Based on the lack of temporality in our data, imputation was not attempted. Future research should explore approaches for handling missing data in longitudinal/time-use compositional datasets. Additionally, our accelerometer data processing decisions may have affected generalisability. Our analysis intentionally deviated from standard wear-time definitions due to instability in postconcussion movement behaviours, but other data reduction procedures may also have influenced results (eg, excluding <10 hours of wear may omit valid high-sleep cases or water-based activities; requiring complete data may introduce selection bias). Future research should systematically compare data reduction procedures, as wear-time thresholds, to assess their impact on inclusion criteria and resulting compositional estimates. The current study aligned with previous CoDA movement behaviour research28 31 and compared linear and squared movement behaviour terms for model fit. Future research could compare this standard modelling approach with methods capable of examining non-linear associations (eg, fractional polynomials, regression splines). Further, examining subgroup-specific models based on age and sex would be an important modelling consideration for future work. While a comprehensive set of covariates was included in analyses, additional unmeasured factors (eg, cognitive load, emotional symptoms, social factors) may interact with movement behaviours and influence symptom burden. Additionally, while certain patterns of movement behaviour durations showed benefits for postconcussion symptoms in this study, the types of movement behaviours contributing to these durations are unknown. For instance, the 9 hours of optimal sedentary time observed on day 2 could reflect time spent watching TV, using social media, talking with friends or staring blankly at the wall. Therefore, future paediatric concussion research should incorporate comprehensive time-use surveys that capture specific sedentary behaviour categories to better understand how different activity patterns influence recovery outcomes.
Conclusions
This study highlights the complex interplay and benefits of all movement behaviours in the management of postconcussion symptoms among children and adolescents. Clinicians seeking to improve postconcussion recovery in children and adolescents should adopt the integrated movement behaviour approach by emphasising the importance of balancing both rest and activity. Specifically, optimal rest included higher than average sleep and lower than average sedentary behaviour in the first days and second week postconcussion, and lower than average sleep and higher than average sedentary behaviour in the first week postconcussion. While optimal activity included lower volume (ie, less LPA) and higher intensity (ie, MVPA), physical activity in the first week post-concussion was followed by promoting both intensity and volume of physical activity in the second week.
Data availability statement
Data are available on reasonable request.
Ethics statements
Patient consent for publication
Ethics approval
This study involves human participants and main ethics approval was obtained from CHEO Research Ethical Board (16/80X). All participating institutional ethics committees approved the PedCARE study. Parents provided informed consent and participants provided informed assent or consent to participate in the study before taking part.
Acknowledgments
We thank and acknowledge the research coordinators and research assistants across the three sites responsible for patient recruitment, enrolment and follow-up. Student volunteers at all three sites provided assistance in patient screening at the emergency departments. We appreciate the collaboration and assistance of all the treating physicians of the emergency departments across the sites.
