Mobile Subthreshold Exercise Program (MSTEP) for concussion: study protocol for a randomized controlled trial

Approximately 1.9 million youth sustain a concussion each year [1], and up to 30% experience persistent postconcussive symptoms (PPCS) such as headache, dizziness, and difficulty focusing that continue for weeks or months [2‐5]. PPCS results in greater utilization of subspecialty care and can impact immediate and long-term social development, cognitive function, and academic success [6]. Individuals with PPCS often have increased symptoms when engaging in physical activity, and such exertional symptoms can lead to avoidance of physical activity and subsequent disability [6]. Research, including our own work, has examined potential benefit of subthreshold (or rehabilitative) exercise [7‐17], with the hypothesis that exercise performed below the threshold that results in symptoms will facilitate recovery. Several randomized controlled trials (RCTs) have been conducted using exercise to treat youth with concussion, showing significant benefit of exercise compared to control [8, 9, 11], but all have required weekly in-person visits, which impedes access for many youth, particularly rural or underserved youth [18‐23]. Delivery of physical activity (PA) interventions virtually can help to transcend geographic barriers, thereby improving access [24, 25].

Prior human and animal research suggests that postconcussion exertional symptoms may result from dysfunction of cerebral autoregulation, causing transmission of systemic hypertension to the cerebral space [18‐20]. Rehabilitative exercise is designed to retrain the body’s physiologic response to exercise, such that an individual is no longer experiencing symptoms with physical activity. In addition, research with rodent models has shown that exercise in the subacute time period following brain injury (> 1 week) increases angiogenesis and concentrations of neurotrophins such as Brain-derived neurotrophic factor (BDNF), which are thought to act as mediators of recovery [26‐28]. Parallel research with humans has found increased cerebral blood flow during exercise [29], and associations between aerobic exercise participation and improved cognition in older adults [30, 31]. In our pilot work [16], we found youth with PPCS had a median moderate-vigorous physical activity (MVPA) per day of only 14 min at baseline (measured using accelerometry with hip-mounted ActiGraph GT3X), despite having previously been athletes. This is in contrast to NHANES (National Health and Nutrition Examination Survey) reported average MVPA for US adolescents of 20–45 min per day [32], and guidelines from the US Physical Activity Guidelines Advisory Committee [33] recommending 60 min of MVPA per day for youth. We thus hypothesize that increases in MVPA in part may mediate the improvement in concussive symptoms seen secondary to a rehabilitative exercise approach (Fig. 1).

We also have noted improvement in psychologic factors such as fear avoidance following subthreshold exercise in our prior trials [16, 17], a unique finding given that research on exercise as a treatment for PPCS has primarily focused on physiologic change rather than psychological factors [18]. Fear avoidance is a concept that has been used to understand the development and perpetuation of chronic pain in adults and youth [34‐36]. According to fear avoidance models, individuals experiencing pain can develop high levels pain-catastrophizing, or a tendency to magnify and ruminate on the potential threat associated with pain and discomfort, leading to feelings of helplessness [37]. Pain-catastrophizing provokes a fear of anticipated pain, avoidance of activities that might cause pain, and ultimately inactivity and dysfunction. We have applied this model to concussion [16, 17], theorizing that individuals with concussion may become fearful that any discomfort during PA is indicative of worsening injury, and begin to avoid all PA. Avoiding all PA then creates functional disability, as youth develop neurophysiologic changes due to lack of movement [38], and secondary mood symptoms due to the disconnection from their normal activities (e.g., school and sports) and peers [6]. Our pilot data indicated declines in fear avoidance that parallel declines in concussive symptoms [16].

We also frame our understanding of the pain experience using the biopsychosocial model, which posits that an individual’s experience of symptoms such as pain is influenced by a complex array of psychosocial factors, particularly individual characteristics (such as gender) and parent and family behaviors [39, 40]. It is thus essential that we explore potential differential response to a rehabilitative exercise intervention in view of these factors. We have made efforts to include sex in our analytic models, as females are diagnosed with concussions more frequently than males playing the same sport [41, 42] and are more likely than males to experience PPCS [43‐46], but males have greater responses (i.e., increases in MVPA) to exercise interventions when utilized in contexts such as obesity [47, 48]. No prior studies of rehabilitative exercise interventions for concussion have examined differential effects by sex [18, 19, 21, 23, 49]. Parental behavior should also be considered, as parents play an important role in the social context of pain and discomfort, and parental concern about the potential danger associated with pain can increase the likelihood of youth concern [50]. Research suggests that parental protective or solicitous behavior (such as paying greater attention to a child due to pain) is associated with impairments in function [51‐53]. It is critical to account for biopsychosocial factors in our analysis and specifically explore whether sex or parental protectiveness result in differential responses to rehabilitative exercise for PPCS.

In sum, we are undertaking a large multisite RCT of a virtually delivered subthreshold exercise program (MSTEP) to treat youth with PPCS. We will also test an innovative conceptual model to explain the MSTEP treatment effect, including both physiologic and psychological mediators (Fig. 1). Sustained increases in heart rate due to MVPA produce increased cerebral perfusion [54], which might underlie decreased symptoms and improved function. Encouraging youth to exercise despite fears of exacerbating symptoms could also result in lessened activity avoidance, thus improving function [55]. Given the influence of parent and family factors, we will also assess the role of parental behavior in shaping youth response to the intervention.

The goal of this study is to assess the efficacy of a virtually delivered subthreshold exercise program (MSTEP) for treating youth with PPCS. We will test the effect of 6 weeks of MSTEP on trajectory of concussive symptoms and health-related quality of life (HRQoL) 6 months after study entry compared to youth in an active control group (stretching). We will also examine mediation of treatment effects by changes in physiologic mediators (MVPA) and psychological mediators (fear avoidance of concussive symptoms).

The primary hypothesis of this RCT is that MSTEP will improve outcomes (concussive symptoms and HRQoL) in youth with concussion compared to a control intervention. We further hypothesize that improvements in outcomes due to MSTEP will be mediated by increases in aerobic activity (i.e., MVPA) and decreases in fear avoidance of concussive symptoms.

Youth with PPCS will be randomized to 6 weeks of either MSTEP (a virtually delivered subthreshold exercise program) or an active control (stretching). This is a superiority trial, testing whether MSTEP is superior to the active control. Randomization will be allocated 1:1 to MSTEP and control, stratified by severity of concussive symptoms and sex at birth. Data will be collected at baseline, weekly during the intervention and at 6 weeks, 3 months, and 6 months.

Subjects will be recruited from pediatric subspecialty clinics in Sports Medicine and Rehabilitation Medicine in Seattle, WA, and Madison, WI, in the USA. In addition to those two locations, multiple avenues will be utilized for broader national recruitment including clinician referral and online advertisement.

Research coordinators will obtain verbal consent from parents and youth via phone or video conferencing using a standardized interview documenting risks and benefits of the study. Parents and youth will document consent by electronically signing parent permission and assent or consent forms. All procedures have been approved by the Seattle Children’s Institutional Review Board.

Not applicable as biologic specimens are not collected for this study.

We have developed a novel approach to deliver a subthreshold exercise intervention to youth with concussion (Mobile Subthreshold Exercise Program, MSTEP) that utilizes video conferencing software and personal fitness devices such that the intervention can be delivered completely remotely. We will compare this intervention to an active control (stretching) to ensure engagement with the study by control participants.

Youth will be free to discontinue participation at any time. We do not have criteria for investigator initiated discontinuation of treatment.

Participants in both arms will continue to receive usual care from their concussion provider.

No post-trial care is provided.

We plan to recruit 200 subjects. Given the RCT design and repeated assessments, we calculate power for our primary outcomes (HBI) and (PedsQL) using two-way repeated analysis of variance (ANOVA) with two factors: study group is the between-subject factor, and time point (baseline, 6 weeks, 3 months, and 6 months) is the within-subject factor. To address our main hypothesis of differences in trajectories, we will test the between-within interaction effect, i.e., the interaction between study group and time. If we were to see similar differences in means between the two study groups on HBI and PedsQL as we did in our pilot STEP RCT [16], our planned sample size of 200 would provide > 99% power to detect a significant group-by-time interaction (i.e., differences in trends over time between the groups) in both HBI and PedsQL. This estimate is based on 4 time points, similar but slightly larger variances as we observed in pilot data, and an ICC = 0.5 for moderate within-subject correlations. We also wanted to ensure we would have sufficient power to assess mediation. For this power calculation, we used the regression model approach for single mediator. Vittinghoff [98] showed that for the linear regression, \( {y}_i={b}_0+{b}_1{x}_i+{b}_2{m}_i+{\epsilon}_i,{\epsilon}_i\sim N\left(0,{\sigma}_e^2\right), \) testing the mediation effect is equivalent to testing the null hypothesis b₂ = 0 versus the alternative hypothesis b₂ ≠ 0. The sample size depends on power, type I error rate, regression coefficient b₂, standard deviation of the mediator σ_m, standard deviation of the error term in the above regression σ_e, and the correlation between the predictor x and the mediator m, ρ_xm. Using residualized change scores from our pilot data for the outcome (HBI) and mediators (MVPA and FOPQ) [16], we found that if we were to observe similar mediation effects from MVPA levels and FOPQ as in our pilot data, our proposed total sample size of 200 would provide 78% power to assess mediation by MVPA and 95% power to assess mediation by FOPQ.

Patients will be recruited from both local clinic populations and nationally via contact with colleagues who care for youth with concussion and online advertisements.

We will utilize block randomization, stratified by severity of concussive symptoms (HBI ≥ 20) and sex at birth, with computer-generated random number sequences.

The randomization table will not be visible to the research staff.

Allocation will occur via REDCap.

We will describe the study as a comparison between exercise programs, in order to blind participants as to whether they are in the intervention or control arm. The statistician will be blinded to intervention status by labeling the intervention variable non-specifically in the dataset (i.e., 1 and 2). The PI and research coordinators will not be blinded.

Procedure for unblinding if needed {17b}

Not applicable as this is not a pharmaceutical trial.

We will collect two types of data: [1] survey data and [2] physical activity data via accelerometry. Please see Table 1 for time points of data collection.

Survey data will be collected online using REDCap and self-report.

Accelerometry data will be collected using the ActiGraph GT3X Link (see section “Outcomes {12}”).

Please see section “Strategies to improve adherence to interventions {11c}” as this section was written to pertain both to adherence and retention.

Survey data will be stored online in REDCap. When possible, data fields will be limited to ensure data quality (e.g., utilizing ranges for dates). Accelerometry data will be downloaded from actigraphs and merged with other data using unique identifiers.

Survey data will be entered and stored in a HIPAA-compliant online database (REDCap) accessed only by study staff. Data from accelerometry will be stored on a server at the Seattle Children’s Research Institute with strictly controlled physical access both at the perimeter and at building ingress points by professional security staff utilizing video surveillance, state of the art intrusion detection systems, and other electronic means for ensuring confidentiality. After data is collected, information which would identify the subjects will be removed and code numbers used instead with the translation between the code and PHI stored in a separate file. All analyses will be conducted with deidentified data.

These are not applicable as these types of data are not being collected.

The primary outcome analyses will be assessed using intention-to-treat with the predictor of interest being experimental group regardless of adherence with assigned treatment. We will examine the impact of adherence on treatment effects in secondary analyses. Pre-identified potential confounders such as demographics (age, sex, and body mass index [BMI]), baseline biopsychosocial factors (standardized measures of depression and anxiety [PHQ-9 and GAD-7]), baseline clinical factors (history of prior concussion, severity of symptoms, duration of symptoms at entry into the study, outcome measures at baseline), and parental protective behavior (Adult Responses to Children’s Symptoms, ARCS) will be summarized using descriptive statistics. In addition to stratification factors (age, sex, and severity of symptoms), we make the a priori decision to control for biopsychosocial factors, particularly parental protective behavior, as well as any other unbalanced confounders from the list above as independent predictors in all subsequent regression analyses. We will model and compare trajectories of outcome measures using mixed effects models. We choose mixed effects modeling for its ability to account for variation at different levels, flexibility regarding time trend and unequally spaced time points, and ease with management of missing data.

The first part of the primary analysis will examine differences in HBI and PedsQL (primary outcomes) between study arms post-intervention (6 weeks) and follow-up (6 months) respectively. The main analytic tool will be based on covariate-adjusted regression models. The two primary outcomes can be considered as continuous outcomes thus linear regression models will be applied, in which group assignment will be the predictor of interest and other pre-identified confounders will be adjusted as covariates.

The second part of the primary analysis will examine trajectories of concussion symptoms. Given that concussive symptoms (HBI) will be assessed 9 times, we believe the data will provide sufficient resolution to model time as a continuous variable when examining change in HBI over time. Let Y_ij denote the HBI measured for ith subject at time t_j, x_i denote the group assignment for ith subject, and Z_i denote the vector of covariates; we will consider the following general model specification: \( (1)\ {Y}_{ij}=f\left({\boldsymbol{\beta}}_i^{\left({x}_i\right)},{t}_j\right)+\boldsymbol{\gamma} {\boldsymbol{Z}}_i+{\epsilon}_{ij};{\boldsymbol{\beta}}_i^{\left({x}_i\right)}\sim N\left({\boldsymbol{b}}^{\left({x}_i\right)},\boldsymbol{\Sigma} \right);{\epsilon}_{ij}\sim N\left(0,{\sigma}^2\right) \). Function f(β, t) can take flexible functional forms to capture different shapes of trajectories. For example, f(β, t) = β₀ + β₁t represents a linear trend over time; f(β, t) = β₀ + β₁t + β₂t² represents a quadratic trend over time; or more generally, we can assume a restricted cubic spline for f(β, t) to represent more flexible trajectory shapes. Based on our pilot study and prior studies in the literature, we have observed that concussive symptoms and other health outcomes exhibit faster rates of improvement in the acute phase post-injury (< 3 months), with rates then decreasing and leveling off. For these reasons, a more parsimonious and accurate function for time trend could be exponential decay function in the form f(β, t) = β₂ + (β₀ − β₂) × exp(−β₁t),where t is time, β₀ is the initial value at t = 0, β₂ is the asymptote as t goes to infinity, β₁, constrained to be positive, is the rate of change or how quickly the process decays from the initial value to the asymptote [99]. Parameters b⁽¹⁾ and b⁽²⁾ are the group specific shape parameters, and equality of these parameters will be tested using multivariate Wald test to assess the hypothesis that concussive symptoms (HBI) decline more quickly in the MSTEP group compared to control. Note the rate of change β₁ could also be influenced by other biopsychosocial variables. We can address this by considering a more advanced hierarchical model in which: [2] log(β₁(x_i, Z_i)) = α₀ + α₁x₁ + α₂Z_i, the logarithmic transformation ensures positivity of the change rate and parameter α₁captures the difference in rate of change between study arms.

Health-related quality of life as measured by PedsQL will be assessed at 4 time points. The smaller number of time points limits options for trajectory shapes; thus constraining us to examine changes over time between groups by modeling time as a discrete predictor: (3) Y_ij = β_0i + b₁x_i + b_2jt_j + b_3jx_i × t_j + ϵ_ij; β_0i~N(b₀, τ²); ϵ_ij~N(0, σ²).

With this specification, the random intercept β_0i captures within-subject correlation, and coefficients for the group-by-time interactions b_3j estimate the differences in changes over time between groups. Individual coefficients can be tested using Wald test, and overall group-by-time interaction can be tested using likelihood ratio test.

This is not applicable as this is not a pharmaceutical trial.

We will examine potential mediation of the intervention effect by [1] MVPA and [2] fear avoidance of concussive symptoms We plan to conduct two sets of mediation analyses. The first set of mediation analyses will focus on the changes between baseline and 3 months. The main motivation for this choice is that this analysis amounts to a conventional single mediator model which is easy to implement and has a simple interpretation, directly addressing questions such as, “How much of the effect of MSTEP on improved concussive symptoms is mediated through increased MVPA or decreased fear avoidance?” We will use residualized change scores instead of the difference scores in part because they adjust for baseline differences and avoid some of the problems with the reliability of difference scores. The residualized change score is the difference between the observed score at 3 months and predicted score at 3 months, where baseline score is used to predict 3-month score. With residualized change scores calculated for mediators (MVPA and FOPQ) and outcomes (HBI, PedsQL), we can proceed to the standard single mediator model as depicted in the following figure. The mediation model can be represented using the three ordinary-least-squares (OLS) regression equation: Y_i = i₁ + cX_i + ϵ_1i; M_i = i₂ + aX_i + ϵ_2i; Y_i = i₃ + c^′X_i + bM_i + ϵ_3i, where c is the total effect of X on Y, c’ is the direct effect of X on Y controlling for M, a is the effect of X on M, b is the effect of M on Y controlling for X, and ab is the mediated effect. Significance of mediated (indirect) effect can be tested via Sobel test or bootstrap methods [100].

The second set of mediation analyses will be longitudinal mediation using a latent growth curve modeling (LGCM) framework, allowing us to utilize the repeated assessments and therefore provide more insight into the mediation process, while allowing more accurate functional forms. Step 1: The growth trajectories of mediator and outcome are modeled separately to determine the trajectory shapes and estimate the overall changes in the mediator and the outcome. In the following, we assume exponential decay trajectories for both mediator process and outcome process, although different forms can be assumed. The measurement models can be specified as: Mediator Process : M_ij = β_2i + (β_0i − β_2i) exp(−β_1it_j) + e_ij

Step 2: Mediational process is then modeled by combining the two growth curve models and relating the growth factors (random effects β’s and γ’s) and the independent variable X (group assignment) in the structural model expressed as follows:

These equations posit that [1] the initial state of outcome process (γ₀) affects rate of change (β₁) and asymptote (β₂) of the mediator process [2]; initial state, rate of change, and asymptote of moderator process affect both rate of change and asymptote of the outcome process. The single mediator model and longitudinal mediation model using LGCM will be implemented estimated using M-Plus and R statistical software.

Multiple imputation (MI) will be used to adjust model estimates that include incomplete data. Specifically, we will generate multiple imputed datasets with missing values filled in using chained equations (MICE). Baseline characteristics will be imputed using predictive mean matching and missing outcomes will be imputed using joint linear mixed models. Regression models will be fitted to each imputed dataset and the results will be pooled to produce the final estimates for inference.

We will upload the final protocol to clinicaltrials.​gov once all data has been collected. Deidentified data and statistical code will be made available to qualified investigators upon request.

The PI (SC) will oversee the research team and ensure all aspects of the protocol are followed. The PI will also run monthly meetings with all co-investigators (FP, TP, CZ, JM, MAB) to provide additional oversight regarding any issues that arise. A Data Safety Monitoring Board (DSMB) has also been developed consisting of a pediatrician specialized in Adolescent Medicine, a rehabilitation medicine scientist with expertise in concussion clinical care and research, and a senior research scientist with expertise in behavior change research regarding physical activity.

The PI (SC) will oversee data monitoring to assess issues with missingness and ensure quality. All analysis will be completed by the study statistician (CZ). A DSMB has also been developed as described above.

Adverse events will be reported as per IRB and NIH guidelines. We will also utilize a data safety monitoring board (DSMB) who will address any safety or ethical concerns. Prior to the start of protocol use, the DSMB reviewed the study procedures and plans for safety monitoring. Over the course of the trial, the board will review the data collection procedures at multiple time points (approximately twice per year, depending on recruitment and issues that arise) and will monitor the occurrence of any adverse events. All potentially adverse events will be reported to the DSMB within 48 h of occurrence. These adverse events include deaths, suicide attempts, severe side effects from physical activity, study dropout, hospitalizations, and/or clinical deterioration defined as the development of new suicidal or homicidal behaviors.

Each year, the DSMB will receive a report that summarizes all serious and unexpected adverse events and the DSMB will produce a report that reviews and summarizes [1] the committee’s opinion as to the safety, confidentiality, and privacy of the study and if study investigators are adequately assuring safety, confidentiality, and privacy [2]; a progress summary towards recruitment and follow-up goals [3]; any concerns relating to serious or unexpected adverse events. This yearly summary will be forwarded to the principal investigators, who will submit it to the Seattle Children’s Research Institute IRB.

The PI will lead study meetings each week with the research staff and will monitor for any concerns. The DSMB will meet approximately biannually, or as needs dictate.

Changes in protocol will be communicated to the public via clinicaltrials.​gov. If participants are actively engaged in the study, they will be notified of any changes that would affect their participation.

Results from the proposed project will be disseminated widely through traditional pathways, first through conference presentations targeting academicians, primary care providers, and concussion specialists, and then via peer-reviewed publications in academic journals. Dissemination to community stakeholders will occur through local and national presentations. We will also work with Seattle Children’s Research Institute media channels to promote publications and presentations that arise from this work.