Persistent postconcussion symptoms (PPCSs) may develop in some individuals after mild traumatic brain injury (mTBI) [1,2], and mTBI can be used interchangeably with “concussion,” as defined by the American Congress of Rehabilitation Medicine in 2023 [3,4]. The symptoms represent a complex phenomenon causing a broad spectrum of issues, including musculoskeletal and balance disturbances, cognitive impairments, visual and auditory dysfunctions, fatigue, and headache-related problems [5]. Adults with PPCSs frequently demonstrate increased sensitivity to physical, cognitive, and perceptual stimuli, which can substantially impair their capacity to engage in physical activity, participate in social interactions, sustain full-time employment, and pursue or complete higher education [6].

In Denmark, approximately 30,000 to 40,000 adults experience traumatic brain injury (TBI) annually, with more than 80% of cases being classified as mTBI [6,7]. However, this number is likely underestimated owing to underreporting and the fact that only hospital contacts are included in the epidemiological statistics. Within the subset of approximately 25,000 individuals with mTBI, 40% experience symptoms persisting beyond 4 weeks, and up to 15% continue to experience symptoms up to a year or longer after the initial concussion [6]. A nationwide cohort study identified 19,732 adults aged 18‐60 years with hospital-diagnosed mTBI over 5 years from 2003 to 2007, and found that 43% of the individuals were not engaged in full-time employment 5 years after mTBI [6].

TBI involves damage to brain tissue, hemorrhage, and neurometabolic disorders [8]. In contrast, mTBI is characterized by normal imaging findings, despite the presence of complex and disabling symptoms [9]. Symptom aggravation in patients with PPCSs is unpredictable in terms of timing and severity, often impeding the ability to lead a normal everyday life. In some cases, it may even be impractical or futile to pursue basic activities. There appears to be a threshold beyond which postconcussive symptoms worsen, potentially due to intolerance to various stimuli and activities. Symptom aggravation may be caused by microtraumas in the brainstem and corpus callosum [10,11]. Symptoms of exercise intolerance are often part of the clinical presentation, and symptom-limited exercise cessation serves as an indicator of this intolerance, reflecting the underlying pathophysiology of PPCSs. Concussions are thought to disrupt autonomic nervous system regulation of cerebral blood flow (CBF), producing abnormal cardiovascular and cerebrovascular responses to increasing metabolic demands that are not detectable at rest [12,13]. A graded exercise test is capable of provoking this dysregulation in a controlled and systematic way [5,12]. By adopting the same operational definitions as Leddy et al [13] and Haider et al [14], a direct comparison with prior work using both the Buffalo Concussion Treadmill Test and the Buffalo Concussion Bike Test (BCBT) is possible.

National clinical guidelines for the nonpharmacological treatment of persistent symptoms following mTBI recommend graded aerobic physical exercise as an important part of rehabilitation; however, these guidelines also underscore the lack of evidence in the existing literature [2]. This highlights the need for further clinical controlled studies to better understand this field, including the pathophysiology of brain metabolism and blood-brain barrier (BBB) permeability [15] in association with functional ability, as well as research into patient perceptions of PPCSs.

The body of literature on PPCSs consists mainly of a limited number of randomized controlled trials (RCTs), with a disproportionate focus on the recovery of elite youth athletes following sport-related concussions [1,16-19]. This research bias significantly limits a comprehensive understanding of the manifestation of PPCSs in the general population. Furthermore, literature often overlooks the daily challenges and symptoms reported by patients in their routine lives. The evaluation of therapeutic interventions frequently lacks the integration of patient perspectives. Thus, an interdisciplinary approach that combines patient-experienced knowledge with individualized therapeutic exercise programs may provide deeper insights into postconcussion cerebral functioning. This approach has the potential to substantially advance the understanding of the field. This could lead to more informed development, implementation, and longitudinal evaluation of intervention methodologies, as well as highlight patient perceptions throughout the complex trauma of PPCSs.

The purpose of this interprofessional RCT is to investigate the effect of a 12-week graded aerobic physical exercise program combined with usual activities and/or care for the treatment of adult patients with PPCSs. Our approach provides an opportunity to triangulate subjective and objective patient data with regard to novel combinations of patient perception, physiology, and neurobiology. The overall study perspective is to establish a tailored, evidence-based active treatment for PPCSs that benefits patients and supports clinical decision-making, and to explain the pathophysiology of PPCSs.

This RCT evaluates the effect of an individualized graded aerobic exercise intervention in adults with PPCSs from the general population. Participants undergo an intensive battery of physical and magnetic resonance imaging (MRI) examinations, alongside personal interviews and process evaluations, throughout the intervention period. The project is divided into the following 4 subprojects:

This study protocol describes the design of the Rehabilitation in Postconcussion Symptoms (REPCon) project, an assessor-blinded, parallel-group, randomized controlled superiority trial. Adult patients with PPCSs showing symptoms for more than 4 weeks after the initial mTBI were randomized in a 1:1 ratio to either a control group (CG) or a TG. Participants in both groups continued their usual daily activities and received usual care as delivered in routine clinical practice. In accordance with the pragmatic design of the trial, usual care is not standardized and reflects current real-world management. Usual care may include follow-up by a general practitioner or medical specialist, symptom monitoring, pharmacological treatment (eg, for pain or sleep disturbances), advice regarding activity modification and return to activity, and referral to allied health services such as physiotherapy or psychology, as clinically indicated. Participants allocated to the TG additionally received a supervised graded physical exercise program, as described below. Concomitant care and co-interventions were discouraged in both groups, and any changes were reported to the study group. The only protocol-mandated difference between groups was the addition of a supervised exercise intervention in the TG. The study includes the following four subprojects: (1) qualitative evaluation, (2) physiological evaluation, (3) neurofunctional evaluation, and (4) comprehensive evaluation. The primary outcome is improvement in exercise tolerance measured as time to symptom exacerbation during the graded aerobic exercise testing protocol 12 weeks after baseline [14]. The secondary outcomes include differences in symptom complexity, dynamic gait, fatigue, headache, and the findings of a range of other neurophysiological tests and functional abilities between the TG and CG at 12 weeks after baseline, as well as the normalization of quantitative MRI measures in the TG and the between-group differences in these measures. The tertiary outcome is the triangulation of data from the subprojects to provide a nuanced and integrated understanding of biology, physiology, and patient perceptions of PPCSs, thereby potentially improving clinical recommendations. The trial has been registered in the ClinicalTrials.gov repository (NCT05785000). The findings will be reported in accordance with the SPIRIT (Standard Protocol Items: Recommendations for Interventional Trials) guidelines [20,21]. The SPIRIT checklist is provided in Checklist 1.

The REPCon project involved 2 settings. The first setting was the House of Practice & Innovation at the University College Copenhagen, Denmark, where data collection for qualitative interviews, physical tests, the exercise intervention of the TG, and process evaluations took place. The qualitative interviews and process evaluations were performed by the project leader (BWL) and the Norwegian research partner (HS). The physical tests were conducted by physiotherapists (JV, FFØ, KDC, IF, and DAS) and guided by the project leader (BWL). All exercise sessions were led by personal trainers with academic backgrounds in physiology (BTH) and physiotherapy or by physiotherapy students and were guided by the project leader (BWL).

The second setting was Rigshospitalet, Copenhagen, Denmark, where the imaging procedure (magnetic resonance [MR] scans) was conducted under the supervision of the principal investigator (HBWL), physicians (SPC and DT), a physicist (ET), engineers (MBV and UL), and radiographers.

The eligibility criteria for trial participation are presented in Textbox 1.

Patients were recruited through a network of multidisciplinary professionals with an established referral flow of adults with PPCSs, including general practitioners, physiotherapy clinics, municipal rehabilitation centers, and the Danish Patient Association for Concussion. Additional recruitment occurred via internal bulletins at the University of Copenhagen and University College Copenhagen, as well as through online postings on the recruitment platform TrialTree. Interested individuals received comprehensive written information about the study and were invited to participate in an initial telephone screening interview completed by the project leader (BWL) and the physiotherapists (JV and KDC). During this interview, a structured questionnaire was administered to assess eligibility based on the predefined inclusion and exclusion criteria. Eligible candidates were subsequently invited to an in-person clinical screening examination at University College Copenhagen, conducted by a medical doctor (HBWL, SPC, or DT) and a physiotherapist (BWL or JV). This session included verification of medical documentation of mTBI, a clinical neurological examination, completion of screening questionnaires, and provision of detailed verbal and written information about the study, including information about how to contact the medical doctor and project leader throughout the project. All patients provided written informed consent prior to enrollment into the trial.

Randomization to either the TG or the CG occurred after completion of the clinical screening examination and followed a 1:1 allocation ratio. Because PPCSs are more prevalent among women than men, an equal sex distribution was not anticipated; however, the allocation ensured an equal number of patients with PPCSs in each group. The randomization sequence was generated by MKZ, who was not involved in any other aspect of the trial, including patient recruitment, assessment, intervention delivery, data management, or analysis, thereby maintaining independence and minimizing the risk of allocation bias. The sequence was produced a priori using a computer-generated randomization list created through an online tool provided by Sealed Envelope. Allocation was implemented according to this pregenerated list, ensuring that those responsible for enrolling patients and assigning them to groups had no access to the randomization sequence prior to assignment, which is in line with SPIRIT recommendations for allocation concealment.

For all baseline assessments in subproject 2 (physiological evaluation), the tester was blinded to group allocation. The tester was also blinded to group allocation for all 12-week follow-up assessments. In addition, different testers conducted the baseline and 12-week follow-up assessments. Furthermore, the personal trainers supervising the exercise sessions were different from all the testers.

In subproject 3 (neurofunctional evaluation), which involved collecting and analyzing MRI data, a blinding protocol was strictly followed. The project staff handling and analyzing the MRI data were unaware of the participants’ group allocations. Similarly, the project leaders and statisticians were blinded to group allocations in the dataset used for the physiological and neurobiological analyses.

All participants completed a multimodal battery of physiological tests (Table 1) at baseline and the 12-week follow-up. Since all participants had completed a subjective symptom profile assessment and a basic neurological test at screening, the purpose of the physical test was to cover relevant physical, oculomotor, balance, autonomic, and cervical aspects of PPCSs through objective measures prior to the validated BCBT [14]. The BCBT determined the individual HR threshold at which concussion-specific symptom exacerbation occurred and outlined the structure for the 12-week training program in the TG.

Participants with PPCSs randomized to the TG participated in an individualized 12-week graded aerobic exercise program performed on an ergometer bicycle. The intervention consisted of 2 supervised sessions per week (24 sessions in total) and was delivered in addition to participants’ usual care and regular activities. In line with procedures applied in the CG, participants were instructed to maintain all ongoing treatments and regular activities throughout the 12-week intervention period and were not permitted to initiate any new treatments or activities during this time. This requirement was clearly communicated at screening visits and prior to the blinded random allocation as a condition for study participation. Each exercise session lasted 45 minutes (20 minutes for the specific exercise session on the bike and 25 minutes for filling out questionnaires and practicalities), with at least 1 day of rest between sessions. All sessions were conducted at the House of Practice & Innovation, University College Copenhagen, Denmark, and were supervised by exercise physiologists, physiotherapists, and physiotherapy students, all of whom had received standardized training in the delivery of the graded aerobic exercise protocol. At the beginning of each session, patients reported symptoms experienced since the previous session, rated their current symptom severity using a Visual Analog Scale (VAS) [14], and completed the Rivermead Post-Concussion Symptoms Questionnaire (RPQ) [22]. Sessions commenced with a 2-minute warm-up at light resistance (60 revolutions per minute [rpm]), corresponding to 50% of the heart rate at symptom threshold (HRt), as determined by the BCBT [14] conducted at baseline. The BCBT was repeated at weeks 4 and 8 to reassess the HRt and allow individual adjustment of exercise intensity. The main training phase consisted of ergometer cycling at 60 rpm for 20 minutes, with intensity set at 70%‐80% of the HRt. HR, Borg Rating of Perceived Exertion (RPE) [14,23], VAS symptom rating, pulse oximetry results, and watt output were recorded at 0, 7, 14, and 20 minutes during each session. Sessions concluded with a 2-minute cool-down at 50% of the HRt, followed by a brief evaluation of the perceived energy level and symptom response. Exercise intensity and physiological responses were monitored continuously to assess symptom provocation. Training was individually adjusted based on subjective symptom reporting (VAS) and perceived exertion (Borg RPE). As part of a standardized protocol, cervical posture, shoulder girdle tension, and breathing patterns were monitored and corrected to ensure an appropriate cycling environment.

Exercise was terminated immediately if any of the following criteria were met: (1) exacerbation of PPCSs, defined as a ≥3-point increase on the VAS; (2) a HR exceeding 90% of the HRt; or (3) a desire or need to discontinue the session.

Adherence to the study protocol was monitored throughout the 12-week intervention period. Participants in both the TG and CG reported their perceived well-being and exercise activity twice weekly using mail or SMS text messages. In these messages, participants specified the type and duration of their rehabilitation and physical activities. Automated reminder messages were sent in cases of missing responses.

Subprojects 1 and 4 were epistemologically informed by phenomenological [24] and constructivist [25,26] perspectives, aiming to capture participants’ lived experiences of PPCSs while highlighting the process of participating in a clinical study. During the 2-year phase of the overarching data collection, qualitative interviews were performed with 6 participants, and data were generated through open-ended interviews that encouraged rich first-person descriptions and paved the way for a second round of qualitative interviews [25]. Following the 2-year phase of data collection, all participants were invited for a second round of individual interviews. A total of 44 participants completed these interviews, and data were generated through a semistructured interview guide. Analytically, data will be examined using the Charmaz constructivist grounded theory methodology [25]. The analysis will proceed through iterative cycles of initial and focused coding, constant comparison, and memo writing. In vivo codes will be used where appropriate to retain participants’ own language, supporting the development of interpretive categories grounded in the empirical data and leading to the development of a conceptual model of living with PPCSs and participating in a clinical trial.

The participants completed 2 MRI sessions (scan sessions A and B) at baseline and after the intervention period. An overview of MRI data collection is shown in Table 2. The MRI scans were performed at the Department of Clinical Physiology and Nuclear Medicine, Rigshospitalet, using a Philips 3T dStream Achieva MRI scanner with a 32-channel phased-array head coil.

Scan session A involved the acquisition of high-resolution structural brain images using a 3D T1-weighted turbo field echo sequence. The total brain volume, gray matter volume, white matter volume, and cortical thickness were estimated through the segmentation of the brain using standard software packages such as Freesurfer. A fluid-attenuated inversion recovery (FLAIR) sequence was used to acquire brain images for the assessment of the white matter lesion burden. Susceptibility-weighted images were obtained for the assessment of possible microbleeds or larger hematomas. Diffusion-weighted imaging was performed with b-values of 500, 1500, and 4000 s/mm² in 6, 6, and 15 different diffusion encoding directions, respectively, allowing restricted spectrum analysis of brain tissue integrity. Finally, the BBB was measured with dynamic contrast-enhanced (DCE) T1-weighted MRI, using a standard dose of a conventional MRI contrast agent. BBB permeability was calculated using the Patlak method, where the slope of the Patlak plot (Ki) is an estimate of the unidirectional influx constant of the contrast agent as a marker of BBB permeability and can be interpreted as the permeability surface area product of the BBB with reference to the contrast agent. An in-depth description of the methods and calculations has been provided in previous publications [27,28].

Scan session B focused on a hypoxic stress test. CBF, cerebral metabolic rate of oxygen (CMRO2), and brain lactate concentration were obtained during normoxic conditions, and measurements were repeated during inhalation of hypoxic air containing 12%‐14% fractional oxygen. The hypoxic condition was created by connecting the participant through a 1-way valve and wide-bore tubing to a face mask covering both the nose and mouth and to an AltiTrainer system (SMTEC) that delivered hypoxic air. Throughout the scan, HR, blood pressure, arterial saturation (SaO2), inspired and expired oxygen partial pressures, and end-tidal carbon dioxide were continuously monitored using a Veris 8600 vital signs monitor (MEDRAD Inc). The hypoxic state lasted about 20 minutes. The initial 6‐7 minutes were used for ensuring a hypoxic steady state, and this was followed by repeated MRI measurements.

CBF was determined by measuring the blood flow through the feeding cerebral arteries using velocity-sensitive phase-contrast mapping MRI. Blood flow in the cerebral arteries was calculated by multiplying the mean blood velocity by the cross-sectional area of regions of interest defining each vessel. Total CBF was normalized to individual brain volume for each participant to yield CBF values in mL/100 g/min.

CMRO2 was calculated using the Fick principle as follows:

The venous oxygen saturation (SvO2) of the blood leaving the brain in the sagittal sinus was measured using the susceptibility-based oximetry MRI technique. SaO2 was obtained by finger pulse oximetry and is known to be close to arterial oxygen saturation. An in-depth discussion of the sequences and postprocessing methods has been reported previously [29-33].

Cerebral lactate concentrations were measured using magnetic resonance spectroscopy (MRS) with a water-suppressed point-resolved spectroscopy pulse sequence. The MRS voxel was placed in the precuneus, and the sequence was optimized to measure lactate. Postprocessing and quantification were performed using LCModel (Version 6.3‐1F). The water peak measured in the spectrum was used to quantify the lactate concentration. The water concentration in the voxel was calculated based on the proportions of gray matter, white matter, and cerebrospinal fluid in the voxel identified by tissue segmentation of the anatomical images [33,34].

A 2D gradient-echo, dual-echo pseudo-continuous arterial spin labeling (pCASL) sequence with an echo-planar imaging readout was used. This sequence involved 2 echo times, 54 dynamics, and a total duration of 8 minutes for visual stimulation that consisted of a flickering black-and-white checkerboard presented to the participant inside the MR scanner. The flickering checkerboard was presented 4 times in 54-second blocks, and a 54-second period with a black screen was present before and after the session and in between each stimulation block. The flickering frequency was set to 8 Hz, which is known to result in the highest metabolic response [35]. CBF maps were obtained by subtracting blood-labeled and nonlabeled control images, and blood oxygenation level–dependent (BOLD)-weighted maps were calculated from the nonlabeled images acquired at the second long echo. The methodology has been described previously [35].

Adverse events, adherence, and patient safety were monitored continuously throughout the trial. Withdrawal from the trial was considered if an adverse event occurred, adherence deteriorated substantially, or a serious illness arose such that the patient’s safety or the scientific integrity of their continued involvement could no longer be ensured. Decisions regarding withdrawal were made on a case-by-case basis, considering the nature, severity, and potential impact of the event on both patient safety and trial validity. All events relevant to these evaluations, including symptom exacerbations, missed sessions, safety-related concerns noted during supervised training sessions, and medical issues reported by the patient, were documented and revised. The project leader (BWL) was responsible for assessing each situation and determining the appropriate course of action, including involvement of medical doctors and withdrawal of the patient from the trial.

All data, including questionnaire data, were managed securely and in full compliance with the General Data Protection Regulation (GDPR) and the Danish Data Protection Agency’s guidelines for the protection of personal data. All data were collected electronically and entered directly during physical testing and training, as well as during MRI scans. Data collected through mail correspondence were entered directly by patients when they submitted responses via secure mail and were stored on secure drives at the trial sites. For qualitative interviews and process evaluations, data were recorded electronically and transcribed by project assistants, and the original recordings were subsequently deleted. All data were pseudonymized, and the decryption key was stored separately from the dataset. Access to the data was controlled and provided by Rigshospitalet and University College Copenhagen. All data were stored on secure drives at the trial sites. Written consent forms were stored in locked cabinets within a secured room, and the forms will be retained for 5 years following the completion of the RCT.

Based on previous research [14], a clinically meaningful change in exercise tolerance on the BCBT has not been formally defined using a minimal clinically important difference. Therefore, a clinically meaningful difference was prespecified based on a combination of the empirical variability and functional interpretation of the test. In the original BCBT validation study, patients demonstrated a mean time to symptom provocation of 14.6 (SD 6.0) minutes, whereas healthy controls showed a mean time of 27.2 (SD 5.2) minutes, indicating a disease-related impairment of approximately 12‐13 minutes. A between-group difference of 6 minutes was therefore defined as clinically meaningful, corresponding to approximately half of this observed deficit, 1 SD of patient performance, and completion of 3 additional workload stages on the BCBT, which progresses in 2-minute increments.

Assuming a common SD of 6 minutes, a between-group difference of 6 minutes corresponds to a standardized effect size of 1.0. With a 2-sided type 1 error rate of 5% and a statistical power of 90%, at least 22 participants were required in each group.

Relevant uncertainty measurements for BBB permeability, spectroscopy, lactate, and diffusion were already available for subproject 3, for both paired and unpaired data. Specifically, for BBB permeability, we set the standard difference to 1 (effect size) based on our previous investigation of the reproducibility of BBB permeability [45], and setting the significance level to 1% and the power to 90%, 27 participants were required in each group. Considering a tentative dropout rate of 30%, we aimed to recruit a total of 70 participants in the study.

The primary outcome of this RCT study is an increase in time to symptom exacerbation during the BCBT from before to after the 12-week intervention. To evaluate this effect while avoiding sensitivity to false-positive error from repeated measurements on the same participant, we will use a linear mixed model. To assess the effect of the intervention, a regressor indicating either pre- or postintervention (Time) and a regressor indicating group allocation (Group) will be included as fixed effects. Participant identification will be included as a random effect (u) to account for repeated measurements. Sex and age may influence these results, and they will be included as fixed effects. The additional effect of group allocation and pre- and postintervention on the results of the BCBT will be included as an interaction term (Group_i and Time_ij). This interaction term will provide the main result of the RCT study. The model is formulated as follows:

where Y is the duration (in minutes) of exercise on the bicycle until symptom provocation and βs are regression coefficients, with β₃ representing the primary result of the RCT. The subscript i denotes the subject, and the subscript j denotes pre- or postintervention measurement.

For MRI parameters, 2 main tests will be performed for the baseline data. These analyses will be repeated after the 12-week intervention period and will include the intervention as a regressor, as explained in the following text. At baseline, the CMRO2 level (overall oxygen consumption of the brain) between healthy control participants and patients with PPCSs will be tested using ordinary least squares regression adjusted for age and sex, with heteroscedasticity-consistent SEs to account for possible nonconstant error variance. To investigate whether the physiological response to hypoxic exposure differs between patients with PPCSs and healthy controls, we will test whether the change in CMRO2 from hypoxic exposure differs between the groups. Initially, we will assess differences at baseline (preintervention) using a mixed linear regression as follows:

where Y_ij denotes CMRO2, Group refers to either healthy controls or patients with PPCSs, and SaO2 is the actual oxygen saturation level in blood measured during the experiment (ie, during normoxia and hypoxia). Thus, we take the actual individual level of blood desaturation into account. The regression value β₃ of the interaction term provides an inference regarding the patient’s ability to maintain CMRO2 under hypoxic stress conditions. After the 12-week intervention period, only patients will be tested for a possible change in their ability to maintain CMRO2 during the hypoxic exposure. Patients will be split according to group allocation (CG or TG), and the model is formulated as follows:

where Time_ijk codes for pre- or postintervention (k) and Group codes for either the CG or TG, with the other parameters being as described above. The interaction term infers information on whether time (pre- or postintervention) has any impact on the group-specific CMRO2 response. Comparison of BBB permeability between patients with PPCSs and healthy controls will be performed using a linear regression model adjusted for age and sex, with heteroscedasticity-consistent SEs to account for possible nonconstant error variance. For exploratory regional analysis of BBB permeability across multiple brain parcels, Benjamini-Hochberg false discovery rate correction will be applied to control for multiple comparisons.

Structural equation modeling (SEM) and latent class analysis (LCA) will be used to identify latent variables and subgroups associated with PPCSs. Both approaches will be applied in an exploratory framework to generate hypotheses for future research. In the SEM analyses, observed variables will be treated as manifestations of an underlying normally distributed latent construct representing PPCS severity. The model will include the summed Rivermead score, time cycled and maximal workload during the BCBT, MRI-based CMRO2 measurements under normoxia and hypoxia, and BBB permeability. The latent variable will be scaled to the summed Rivermead score to facilitate interpretation. By integrating information across multiple domains, we aim to obtain a more comprehensive and sensitive representation of symptom severity than can be obtained with any single measure alone. Conceptually or procedurally related indicators (eg, fatigue-related items, BCBT variables, and MRI measures) will be modeled to account for residual correlations by including terms describing local dependence. This is performed to avoid inflating latent factor loadings or identifying spurious subgroups driven by unmodeled shared variance. In the LCA analyses, observed variables will be treated as indicators of an underlying categorical latent variable. Patients will thus be grouped according to their multivariate response profiles. The number of classes will be estimated using information criteria (Akaike information criterion and Bayesian information criterion) and by excluding solutions with very small or clinically uninterpretable classes. Given the limited sample size, we will restrict models to a maximum of four classes. For each class, we will estimate the expected response pattern and assess whether the resulting classes have meaningful clinical interpretations. Associations between latent outcomes and external covariates, such as age and sex, will be examined post hoc, with covariates treated as external variables rather than incorporated directly into the latent variable models. This includes assessing whether patients are exercise-intolerant, defined by both clinical criteria and self-reported experience. Finally, we will extend the models to incorporate both baseline and follow-up data. In the SEM analyses, we will estimate the effect of the intervention on the latent PPCS construct, while in the LCA analyses, we will test whether the intervention effect differs across latent classes.

Statistical modeling will be handled externally at the Institute for Biostatistics, University of Copenhagen. In this large study, we have included other a priori defined measures of distinct types, and therefore, further testing will be performed. We treat these tests as hypothesis-generating for further studies and consider that corrections for multiple testing are irrelevant, as discussed previously [46,47].

For the primary outcome, an intention-to-treat analysis including all patients with at least baseline test data as a repeated observation of the dependent variable will be performed using a linear mixed model. Missing data will be assumed to be missing at random, and there will be no imputations [48]. A complementary per-protocol analysis will also be performed, with the inclusion criteria reported in detail. For the primary outcome analysis, participants will be included regardless of the availability of secondary outcomes, including MRI data.

This study has been conducted in accordance with the principles of the Declaration of Helsinki, and the protocol was approved by the Research Ethics Committee in the Capital Region of Denmark (reg. H-21047857; January 2022) and cannot be amended without reapplication. Furthermore, this study has been conducted in adherence to the study protocol and International Council for Harmonisation Good Clinical Practice guidelines, following applicable Danish and European law. The study does not fall under the scope of external monitoring, as it does not involve medical products or medical devices. Patients were covered by insurance through Rigshospitalet and the Danish Patient Compensation Association. Participation was voluntary, and inclusion required informed consent that could be withdrawn at any time by the participant.

The ethics protocol was extended for 1 year to accommodate the inclusion of 2-3 patients per week until December 2025.