Protecting the brain from post-stroke cognitive impairment and dementia with multimodal exercise training: study protocol for a Bayesian adaptive trial (PROTECT) Authors: Kevin Moncion, Lynden Rodrigues, Ashanté Bon, Adam Sutoski, Kira Sikorska, Elric Y Allison, Juliano Abreu, Shirin Golchi, Nathalie Arbour, Claudine Gauthier, Caroline Paquette, Pedro Rosa-Neto, Ilana Leppert, Christopher Rowley, Christine L Tardiff, Alexander Thiel, Baraa Al-Khazraji, Ada Tang, Marc Roig Published: 2026 ABSTRACT AbstractIntroductionStroke triggers acute vascular and inflammatory mechanisms that predispose the brain to rapid neurodegeneration. Up to 52% of stroke survivors develop cognitive impairment within 6 months and 20% receive a clinical diagnosis of dementia within 5 years. The subacute phase (<6 months) represents a critical window in which the brain may be most responsive to neuroprotective interventions. Multimodal aerobic and resistance training improves cognition in chronic stroke, but whether it improves cognition, neuroimaging markers and blood biomarkers of dementia risk when delivered during this early window remains unknown. The PROTECT trial will compare the effects of 12 weeks of multimodal exercise (moderate-to-high-intensity resistance and aerobic training) versus a low-intensity exercise comparator on cognition, neuroimaging outcomes, blood biomarkers of cognitive decline and dementia risk in people with subacute stroke.Methods and analysisThe PROTECT trial is a 12-week, Phase 3, assessor-blinded, multisite Bayesian adaptive randomised controlled trial (RCT) following a two-arm parallel group sequential design with 6-month and 12-month follow-up (NCT07445841). Participants will be randomised to multimodal training or the comparator using concealed allocation with permuted blocks of varying sizes. The primary outcome is cognition, measured using the 13-item Alzheimer’s Disease Assessment Scale-Cognitive assessment (ADAS-Cog-13). Secondary outcomes include ADAS-Cog-Plus, structural and perfusion neuroimaging and blood biomarkers of inflammation and neurodegeneration. Tertiary outcomes include cardiorespiratory fitness, functional mobility, muscle strength, body composition, neuropsychological status, patient-reported cognition, quality of life, fatigue and healthcare utilisation. Outcomes will be assessed at baseline, post-intervention (primary endpoint) and at 6-month and 12-month follow-up. Sample size was estimated via 20 000 Monte Carlo simulations using an ADAS-Cog effect size of Cohen’s d=0.63 from a previous exercise RCT. The target was ≥80% power to detect this treatment effect at a one-sided type I error rate of 2.5%, using a weakly informative prior centred at zero with a variance of 100. The minimum required was 45 completers per arm (N=90) and accounting for 25% attrition, up to 120 participants (60 per arm) will be enrolled. Pre-planned adaptive features include: (1) two interim analyses at 50% and 75% of completers; (2) early stopping for efficacy and futility; and (3) sample size re-estimation.Ethics and disseminationEthical approval to conduct this study has been granted by the Centre de recherche interdisciplinaire en réadaptation du Montréal métropolitain (CRIR MP-50-2025-2294) and Hamilton Integrated Research Board (HIREB 19222). Any protocol amendments will be submitted to the appropriate ethics boards. Written informed consent to participate in this study will be obtained from all participants by study coordinators or assistants. Study results will be published and reported in peer-reviewed journal following Adaptive Designs Consolidated Standards of Reporting Trials extension guidelines.Trial registration numberNCT07445841.Keywords: Exercise, Stroke, Dementia, Randomized Controlled Trial, Magnetic Resonance Imaging STRENGTHS AND LIMITATIONS OF THIS STUDY The trial uses a Bayesian adaptive group sequential design with two pre-planned interim analyses, early stopping rules for efficacy and futility and blinded sample size re-estimation for efficiency. The trial uses a Bayesian adaptive group sequential design with two pre-planned interim analyses, early stopping rules for efficacy and futility and blinded sample size re-estimation for efficiency. This is a multisite, assessor-blinded randomised controlled trial with computer-generated concealed allocation stratified by site, sex, stroke type and concurrent rehabilitation with comprehensive cognitive, clinical and mechanistic outcome assessment at baseline, post-intervention and at 6 and 12 months follow-up. This is a multisite, assessor-blinded randomised controlled trial with computer-generated concealed allocation stratified by site, sex, stroke type and concurrent rehabilitation with comprehensive cognitive, clinical and mechanistic outcome assessment at baseline, post-intervention and at 6 and 12 months follow-up. Persons with lived experience of stroke and clinician partners are embedded across the trial, having informed the intervention and comparator design, recruitment and screening pathways and assessment burden and continuing on the steering committee through to dissemination. Persons with lived experience of stroke and clinician partners are embedded across the trial, having informed the intervention and comparator design, recruitment and screening pathways and assessment burden and continuing on the steering committee through to dissemination. Due to the nature of the exercise intervention participants and interventionists cannot be blinded to group allocation. Due to the nature of the exercise intervention participants and interventionists cannot be blinded to group allocation. Introduction and rationale The acute vascular and inflammatory events triggered by a stroke1 predispose the brain to a rapid neurodegenerative process2 that can lead to cognitive impairment and dementia, both of which are very prevalent (15–70%)3–5 and can appear very soon after the stroke event.6 Since 52% of cases will experience post-stroke cognitive impairment within 6 months post-stroke and 20% will receive a clinical diagnosis of dementia within 5 years6, preserving neurodegeneration and promoting cognitive recovery as early as possible is crucial. The subacute phase post-stroke (<6 months) constitutes a critical window of opportunity to implement interventions that promote neural repair7 and cerebrovascular remodelling.8 Interventions introduced later in the chronic phase of recovery (>6 months) may miss this window and lose part of their effectiveness, leading to suboptimal long-term recovery in mobility,9 motor function10 and cognition.11 Targeted interventions that stimulate neuroprotection during the early phases of recovery may reduce post-stroke disability12 13 and lower the disproportionally high incidence of dementia observed after stroke.11 Physical exercise improves some aspects of cognition14 and cerebrovascular health,15 prevents neurodegeneration16 and reduces dementia risk in non-stroke populations.17 In patients with chronic stroke (>6 months), multimodal training interventions that combine both aerobic and resistance exercises may be effective at improving cognition and markers of cerebrovascular health.18 19 However, there is no research examining the protective effect of this type of training on dementia risk in the earlier subacute phase of stroke recovery (<6 months), a period of heightened neuroplasticity7 in which the brain could be more responsive to the protective effect of exercise than during chronic phases of recovery. The PROTECT trial will compare the effects of 12 weeks of moderate-to-high intensity multimodal training versus a low-intensity exercise comparator on cognition as measured by the 13-item Alzheimer’s Disease Assessment Scale-Cognitive assessment (ADAS-Cog-13), neuroimaging outcomes, and blood biomarkers of cognitive decline and dementia risk in people with subacute stroke (<6 months post-stroke). Methods and analysis Design PROTECT is a 12-week, Phase 3, assessor-blinded, multisite Bayesian adaptive randomised controlled trial (RCT) following a two-arm parallel group sequential design with 6-month and 12-month follow-up (figure 1). This registered RCT (NCT07445841) has the following pre-planned adaptive features, (1) two interim analyses at 50% and 75% of current maximum completers; (2) early stopping rules for efficacy and futility; and (3) sample size re-estimation. This protocol follows the Standard Protocol Items: Recommendations for Interventional Trials (SPIRIT) guidelines20 and results will be reported following the Adaptive Designs Consolidated Standards of Reporting Trials (CONSORT) extension guidelines.21 The project began in summer 2025 with administrative planning, the development of ethics applications and standard operating procedures, resource and staffing allocation, the identification of recruitment sources and training workshops for interventionists and assessors. Participant identification, screening, recruitment and enrolment began in winter 2026, of which the first participant was enrolled in the study was 9 March 2026. The project is expected to be completed by winter 2030. Study CONSORT flow chart. CONSORT, Consolidated Standards of Reporting Trials. Patient population Participants will be recruited from hospital, rehabilitation and community sites in Quebec and Ontario, Canada. Inclusion criteria include participants who are medically stable, aged 40–80, within 6 months following a first-ever stroke confirmed by MRI/CT, who can walk independently at least 10 m (assistive devices permitted) and capable of following instructions. Participants will be excluded if they have a pre-existing diagnosis of dementia, use medications that impact cognition, have significant disability as determined by the modified Rankin scale score of >3, are engaged in another structured exercise programme outside of rehabilitation, have any absolute contraindications to exercise, or have pain or behavioural issues that will prevent safe exercise participation, have contraindications to MRI scanning.22 23 All participants will undergo comprehensive assessments by blinded assessors to characterise their medical history, stroke history and severity, comorbidities and medications. Recruitment Participants will be recruited from clinician referrals and access to participant databanks from patients who have consented to be contacted for participation in research project. Recruitment procedures have been co-developed with patient and clinical partners, who have reviewed screening pathways and advised on facilitators or barriers to participation. Participants for the sites in Quebec will be recruited from the Jewish Rehabilitation Hospital (JRH) in Laval and the Jewish General Hospital in Montreal. The JRH admits 190 patients with subacute stroke each year and has an exercise programme that enrols 5–6 new subacute patients per month. The stroke programme at the JGH admits 260 patients with subacute stroke each year. We will also recruit participants from clinical sites in the greater Montreal area, including West Central Montreal (Richardson, Lethbridge-Layton-MacKay) and Villa Medica. Participants for the site in Ontario will be recruited through the Ontario Central South Regional Stroke Network, the Regional Rehabilitation Centre (Hamilton Health Sciences) and Joseph Brant Hospital as primary sites. Collectively, all sites across Quebec and Ontario admit approximately 1800 patients with subacute stroke per year, and our recruitment rates (five to six participants/month) in previous studies suggest that achieving the target sample size is feasible.24 25 Randomisation Participants will be randomised 1:1 to multimodal training or the comparator using a computer-generated sequence permuted variable block sizes, stratified by site, sex (male/female), stroke type (ischaemic/haemorrhagic) and whether they are receiving concurrent structured in/outpatient rehabilitation (yes/no). The principal investigator (MR), who is not involved in recruitment, consent or data collection or assessments, generated the random allocation sequence and kept it confidential until after the completion of baseline assessments to ensure concealed allocation. Other research personnel will not have access to the random allocation sequence. Interventions Participants will complete 36 supervised sessions (three times a week for 12 weeks) at one of three sites: JRH (Laval, Québec, Canada), McGill University (Montreal, Quebec, Canada) or McMaster University (Hamilton, Ontario, Canada). Each 60-minute session will include a warm-up (~5 min), resistance followed by aerobic exercise and a cool-down period (~5 min). To avoid contamination bias, training sessions will be scheduled at separate times. Sessions will be supervised by cardiopulmonary resuscitation (CPR)-certified and trained interventionists. Training workshops will be delivered bi-annually by KM and LR to interventionists to ensure intervention/comparator groups are being administered, progressed and monitored accordingly as per protocols. Interventionists will participate in routine bi-weekly meetings with KM, LR, AT and MR to monitor treatment fidelity via tracking attendance, exercise training logs (eg, sets, reps, training volume, ratings of perceived exertion) and heart rate responses via heart rate monitors to promote adherence to protocols. Training materials will be accessible on https://clinicaltrials.gov/study/NCT07445841 after database lock. Multimodal training will include individualised moderate to high intensity resistance and aerobic exercise and the comparator group will leverage low-intensity exercises developed through the Fitness and Mobility Exercise Programme.26 Due to the nature of the intervention, participants and interventionists delivering the exercise programmes cannot be blinded to group allocation. Exercise prescriptions (frequency, intensity, time, type and progression) for both groups are presented in figure 2. Exercise prescription for the multimodal training and comparator groups. HIIT, high intensity interval training. Outcome assessment and measures Blinded outcome assessment will occur at baseline, post-intervention (primary endpoint) and at 6- and 12-month follow-up (figure 3). Assessor workshops will be delivered bi-annually by KM and LR to blinded assessors to ensure all outcomes are being administered, collected and entered accordingly as per-protocols. To maintain assessor blinding, participants will be instructed not to discuss their assigned intervention with outcome assessors and assessment sessions will be conducted in spaces separate from the intervention areas. To promote participant retention and complete outcome assessments throughout the follow-up, research coordinators at each site will contact participants via telephone and/or email to confirm and provide reminders of upcoming assessment appointments. Participants who discontinue from the assigned intervention protocol will be encouraged to complete all scheduled outcome assessments where possible. For participants who withdraw entirely, the reason for withdrawal and consent to retain data collected up to the point of withdrawal will be documented. Any missed assessments and the reasons for non-completion will be recorded. Outcome assessment schedule by time point. ADAS-Cog-13, 13-item Alzheimer’s Disease Assessment Scale-Cognitive assessment; RM, repetition maximum. Primary outcome The primary outcome will be change in cognition immediately post-intervention (T1, primary endpoint) as measured by the ADAS-Cog-13,27 a 13-item cognitive assessment with scores ranging from 0 to 85, where lower scores indicate better cognition. The minimal clinically important difference (MCID) on this scale is 2.28 Secondary and tertiary outcomes The secondary outcomes include the ADAS-Cog-Plus, a composite score that includes additional standardised cognitive assessments such as the Trail Making Test Parts A and B, Digit Span Forward and Backward, Digit Symbol Substitution and Category Fluency (animals and vegetables) to increase the responsiveness of the original ADAS-Cog.29 30 The ADAS-Cog-Plus is scored via a multidimensional item response theory model that provides a global score that is calibrated with the Alzheimer’s Disease Neuroimaging Initiative one with scores ranging from ≈−1.0 to 1.0, indicating cognitively healthy (−1.0), mild cognitive impairment (≈0.0) and dementia (≈1.0).31 There are no published MCIDs for the ADAS-Cog-Plus, but it has been shown to be responsive to change following exercise in chronic stroke.18 Neuroimaging outcomes associated with neurodegeneration32 and increased risk of cognitive decline and dementia will be included.33 Whole brain and regional cerebral blood flow,33 blood-brain barrier permeability,32 34 brain volume and white matter structure and integrity35 will be collected using multimodal MRI approaches including pseudo-continuous arterial spin labelling, T1 and T2-weighted imaging, diffusion and tensor imaging.36–38 The middle cerebral artery flow velocity will be collected via transcranial Doppler at rest, during cognitive tasks and orthostatic challenges and aerobic exercise to capture changes in cerebral blood velocity regulation.39 Blood biomarkers of inflammation (eg, interleukin (IL)-6, IL-6, IL-10, tumour necrosis factor-alpha, IL-1beta, IL-18, CRP, growth/differentiation factor-15 (GDF-15), myeloperoxidase (MPO), monocyte chemoattractant protein-1 (MCP-1), neurodegeneration (eg, amyloid-β, tau, klotho) and axonal injury (neurofilament light) that are associated with cerebrovascular risk, cognitive decline and dementia will be collected using serum or plasma and will be analysed with validated assay kits.1 2 40–45 Many of these blood biomarkers have been associated with cognition,41 44 46 ADAS-Cog scores47 and are responsive to exercise training.48–51 Tertiary outcomes will include measures of cardiorespiratory fitness,52 physical function (eg, short physical performance battery),53 functional mobility (eg, usual and fast gait speed, timed-up and go)54 and estimated muscle strength (10-repetition maximum during chest press, leg press, seated row, knee extension and hand grip dynamometry, etc), body anthropometrics and composition (dual-energy X-ray absorptiometry), neuropsychological battery55 patient-reported cognition,56 quality of life,57 fatigue58 and healthcare utilisation will also be assessed. Data safety monitoring board Three independent members, including a physician, exercise researcher and a statistician, will compose the Data Safety Monitoring Board for this trial. Exercise trainers and assessors will monitor and report any adverse events (serious/not serious) that are related and unrelated to the intervention that occur throughout the study to the board. Unblinding of individual participant allocation may be permitted in the event of a serious adverse event where knowledge of group assignment is deemed clinically necessary by the site investigators. The rationale for unblinding will be documented and reported to the board. Sample size estimation To estimate the required sample size for the primary outcome at T1, we used effect estimates from a previous exercise RCT that found a between-group ADAS-Cog treatment effect of 1.83 (pooled SD=2.91, Cohen’s d=0.63).59 The complete Bayesian adaptive group sequential design was simulated across 20 000 Monte Carlo trials, targeting ≥80% power at a one-sided type I error rate of 2.5% under a weakly informative prior centred at zero with a variance of 100. The minimum required was 45 completers per arm (n=90); accounting for 25% anticipated attrition, up to 120 participants (60 per arm) will be enrolled. Timing of interim analyses Two blinded interim analyses will be conducted by a blinded statistician at 50% and 75% of the number of participants who have provided primary outcome data at T1. Interim 1 will be triggered at 22 completers per arm (44 total) and Interim 2 at 34 completers per arm (68 total). At each interim analysis, the posterior probability that the treatment effect exceeds zero and the predictive probability of success (PPoS) will be computed to inform efficacy and futility stopping rules. Sample size re-estimation At the first blinded interim analysis (50% of completers), a blinded sample size re-estimation will be conducted by a blinded statistician based on the PPoS. If PPoS falls within a prespecified promising zone (eg, 0.10–0.80), enrolment will be increased to achieve a target PPoS of 0.90 (NMax=140, 70/arm). Efficacy and futility stopping rules The trial will stop for early efficacy if the posterior probability exceeds 0.999 at Interim 1, 0.995 at Interim 2 and 0.975 at the final analysis.60 The trial will stop for futility if PPoS falls below 0.10, indicating less than a 10% chance of achieving a positive result at full enrolment. Statistical analyses Statistical analyses will follow a prespecified analysis plan using an intention-to-treat and complete case approach and will be finalised before database lock. Bayesian linear mixed models will be estimated using Markov Chain Monte Carlo sampling. A normal conjugate model will be used for inference on the treatment effect. Models will include fixed effects for group, time point and a group×time interaction to assess whether between-group differences vary over time. Unadjusted and adjusted models for age and sex, and random intercepts and slopes will be evaluated. Weakly informative priors will be specified for all parameters. Prior sensitivity analyses will be conducted under alternative prior specifications. The primary treatment endpoint will be group×time interaction at T1. Treatment effect estimates will be reported as posterior means and 95% credible intervals. For each outcome and time point, we will report the posterior probability that the treatment effect exceeds zero and, where relevant, a published MCID. Exploratory subgroup analyses will assess whether treatment effects on the primary outcome vary by the baseline outcome, stroke severity, lesion location, sex, gender and age. Data management Data will be collected from assessments and recorded on paper case report forms. Data from these forms will be entered into a secure, password-protected electronic database (eg, Research Electronic Data Capture), with access restricted to authorised study personnel. All information collected from participants will be recorded under a study-specific subject identifier unique to each site to ensure confidentiality. All personal information collected during the study will be de-identified and coded to ensure its confidentiality. Each participant will be assigned a unique code that is linked to their data, with their name kept separately. Group allocation and exercise training data will be stored separately from outcome data and will not be accessible to outcome assessors or the blinded statistician to maintain blinding throughout the data management process. KM and LR will monitor data entry and perform quarterly audits, including range checks for data values and verification of a random subset of entries against source documents, to ensure data quality and integrity. Trial monitoring Trial monitoring will be conducted on a monthly-basis by members of the steering committee. Site-level monitoring will include quarterly reviews of recruitment progress, protocol adherence, data management and adverse event reporting. Cross-site audits will be conducted by KM and LR to verify compliance with data management procedures. Patient and public involvement Clinician partners and persons with lived experience of stroke are engaged as partners on the PROTECT trial. Partners were involved during the early proposal development stage and funding acquisition stage, where they advised on the suitability and anticipated impact of the trial and the framing of cognitive health as a recovery priority for stroke survivors. Their input has also informed the design of the multimodal exercise intervention and the comparator arm, including the feasibility of session structure, intensity and the overall assessment burden from the perspective of stroke survivors. Partners have also contributed to reviewing recruitment strategies, screening pathways and have advised on transportation and accessibility barriers to participation. Furthermore, partners will sit on the PROTECT steering committee and act as liaisons with stroke community groups and clinical sites to support recruitment and end-of-grant dissemination. Partners will be invited to participate in dissemination of peer-reviewed publications, presentations and knowledge translation activities that target patients, clinicians and the broader public. Ethics and dissemination Ethical approval to conduct this study has been granted by the Centre de recherche interdisciplinaire en réadaptation du Montréal métropolitain (CRIR MP-50-2025-2294) and Hamilton Integrated Research Board (HIREB 19222). Any protocol amendments will be submitted to the appropriate ethics boards. Written informed consent to participate in this study will be obtained from all participants by study coordinators or research assistants (refer to online supplemental file 1 for a sample consent form). Study results will be published and reported in peer-reviewed journal following CONSORT guidance61 and Adaptive Designs CONSORT extension guidelines.21 Study organisation and funding This RCT is funded by operating grants from the Canadian Institutes of Health Research (CIHR #525414) and a Heart and Stroke Foundation of Canada Grant-in-Aid (HSFC GIA #G-24-0037903). The CIHR nor the HSFC had no role in the design of this study and will not have any role during its execution, analyses, interpretation of the data or decision to submit results. Discussion Canadian Stroke Best Practices recommend the use of aerobic and resistance training for managing lower and upper limb deficits.62 However, multimodal training is not considered and specific recommendations regarding what type of exercise is most effective to improve cognition post-stroke are lacking. The results of this RCT may improve current best practices by providing clinicians with a new strategy to implement in early stages of recovery. Summary and conclusions This RCT will provide evidence for multimodal training for promoting cognitive recovery in the early stages of stroke recovery. Supplementary material