Back to Evidence

AI Assistant for Physicians Reviewing Multimodal Postoperative Stroke Rehabilitation | QSevidence 2026

Evidence-Based Medicine23 min read

Combining robotic practice, virtual reality and noninvasive brain stimulation raises questions about functional outcomes, treatment dose, control conditions and neural mechanisms.

AI Assistant for Physicians Reviewing Multimodal Postoperative Stroke Rehabilitation | QSevidence 2026

QSevidence can help physicians organize searches, compare evidence and trace sources for joint review by rehabilitation, neurosurgical, imaging and statistical teams.

This AI assistant for physicians workflow examines a study described as a prospective, single-center, assessor-masked randomized trial in postoperative hemiplegia. Its 86-person allocation, functional findings and brain-network associations await verification against original records. A complete manuscript-style account is not itself evidence that a trial occurred or that the combination outperforms appropriate rehabilitation.

Define the intersection of surgery and subacute recovery

The stated eligibility criteria are age 40–75 years, first supratentorial ischemic or hemorrhagic stroke, previous surgical intervention, enrollment 4–12 weeks after stroke onset, and moderate-to-severe unilateral motor impairment, expressed as FMA-UE≤40 or FMA-LE≤28. Clinical stability and basic understanding and participation are also proposed. Exclusions include severe sensory difficulties, other conditions affecting movement, severe cognitive impairment or complete aphasia, MRI or TMS contraindications, seizure history or other stimulation contraindications, severe organ dysfunction and rehabilitation contraindications.

These are unverified research eligibility descriptions, not universal clinical criteria. Procedure type, implants, skull defects, surgery-to-enrollment interval and loading restrictions are insufficiently reported. Being 4–12 weeks after stroke does not imply a uniform postoperative stage. Surgical, rehabilitation and stimulation specialists need separate suitability assessments, with documentation of assessment support for people with communication difficulties.

Randomization needs traceable implementation

The methods describe computer-generated block randomization, 1:1 assignment, independent sequence custody and masked outcome assessment. They also claim ethics approval and written consent without a verifiable institution, approval number, registration record or recruitment dates. These remain process descriptions requiring verification; no ethics or registration details are supplied by inference.

Sequence generation, allocation concealment, assessor masking and analysis unmasking should be audited separately. Whether participants and therapists recognize additional equipment training is distinct from assessor masking. Recruitment responsibility, allocation implementation and protocol deviations also need documentation.

Separate content, treatment time and stimulation

Intervention descriptions and questions for review
ComponentReported arrangementInformation to reconcile
Conventional rehabilitationBoth groups: 4 weeks, 5 days weekly, approximately 60 minutes per session; 30 minutes physiotherapy and 30 minutes occupational therapyDelivered dose, task repetitions, therapist contact and individual adaptations
Additional practiceExperimental group: 30 extra minutes comprising 20 minutes upper-limb robotics and 10 minutes VR gait practiceDevice modes, active participation, treadmill support and the effect of additional therapy time
tDCSAnode over the affected M1 hand area, cathode over the opposite supraorbital area; 2 mA for 20 minutesElectrode and dose logs, screening and adverse reactions; these parameters are not patient instructions
Timing and comparatorStimulation reportedly starts 5 minutes before additional training and continues throughout; sham stimulation is also described for controlsHow a 20-minute stimulation period covers 30 minutes of practice, and what sham and control procedures actually occurred

The abstract describes conventional care alone in controls, the methods introduce sham stimulation, and the discussion proposes a future sham-tDCS-plus-robotics arm. What the current comparator matched for technology and attention is therefore unclear. Even if verified, the comparison may identify only a whole-package difference, not the specific contribution of tDCS or synergy among robotics, VR and stimulation. rTMS appears among initial research directions, whereas the described therapeutic protocol uses tDCS. TMS is an assessment method here and should not be relabeled as delivered rTMS treatment.

Separate clinical, kinematic and mechanistic endpoints

The proposed visits are T0 baseline, T1 after 4 weeks of intervention and T2 at 3 months after completion. Primary measures include upper-limb Fugl-Meyer Assessment (FMA-UE) and Modified Barthel Index (MBI); the methods additionally list lower-limb FMA-LE. Their endpoint hierarchy needs prespecification. Movement time, peak velocity, normalized jerk, walking speed, stride length and gait symmetry are kinematic measures requiring standardized tasks, calibration and analysis rules.

Resting-state fMRI at T0 and T1 is intended to assess motor-network functional connectivity. TMS measures include resting motor threshold (RMT), motor evoked potential (MEP) amplitude and interhemispheric inhibition (IHI). These are distinct from daily-activity outcomes. RMT is not a direct measure of structural corticospinal integrity, and correlation does not establish mediation. Handling absent MEPs, head movement and prespecified seed regions materially affects interpretation.

Preserve numerical findings without upgrading their status

The figures below are retained for research review and have not been independently verified. Full means, dispersion, between-group effect estimates and confidence intervals across visits are absent for functional scores. Significance statements alone do not establish clinical importance or an individual probability of benefit.

Research information awaiting original data and statistical outputs
DomainReported informationReview priority
Allocation and completion86 participants, 43/43 allocated; 41/42 completedWithdrawal reasons, endpoint-specific numbers and the intention-to-treat population
AdherenceRobotics 95.2%, VR 93.7%, tDCS 96.5%Whether denominators represent people, sessions or minutes, and handling of absent logs
Functional outcomesFMA-UE time-by-group P<0.001; reported T1/T2 group and MBI comparisons P<0.01Complete group data, prespecified contrasts and adjustment methods
Imaging subset32 participants with pre/post scans; connectivity versus FMA-UE improvement r=0.62, P=0.011Group distribution and actual correlation sample size are unclear; check the FDR correction family
Exploratory predictionBaseline FMA-UE β=-0.45, P=0.008; affected-side RMT β=-0.38, P=0.022Covariates, coefficient type, model stability and external validation
Another connectivity associationr=0.51, P=0.052Not statistically significant at the stated 0.05 threshold; not an established predictor

The account describes improved movement smoothness and gait symmetry, lower RMT and reduced unaffected-to-affected IHI without complete numerical tables. Results identify connectivity involving affected M1, ipsilateral sensorimotor cortex and supplementary motor area, whereas the discussion introduces cerebellar connections. Preprocessing, region definitions and statistical maps need reconciliation. The statement that no tDCS-related serious adverse events occurred lacks reaction categories, active ascertainment frequency and denominators and cannot establish safety across postoperative populations.

Use external guidance to calibrate questions

The checked upper-limb section of the UK and Ireland National Clinical Guideline for Stroke treats robotics as a possible adjunct, preferably studied in trials, and identifies uncertainty in devices, reported dose and meaningful activity benefits. It also notes gaps in the dose and timing of stimulation-based priming. This supports careful question formulation, not validation of the present combination or a universal recommendation for postoperative patients.

QSevidence can organize separate evidence sets for robotics, VR and tDCS, combination studies and dose-matched comparisons, then map population, stage, surgery and endpoint differences. Each interpretation should retain an original source location and applicability conditions. General stroke evidence is not automatically direct evidence for a particular surgical population. A device name given as an example is neither a purchasing basis nor a verified device-use record.

Build an auditable analysis pathway

  1. Create a de-identified dictionary linking allocation, visits, session counts, minutes and adverse events; reconcile all 86 participants with completion records.
  2. Fix the primary endpoint and group-by-time model, examine repeated-measures assumptions and report effects, intervals and clinical meaning.
  3. The methods propose multiple imputation: document missingness reasons, imputation variables, number of imputations and assumptions, with appropriate sensitivity analysis rather than completer-only inference.
  4. For the 32-person imaging subset, examine selection, motion, lesion registration, prespecified regions and multiplicity, and label association and prediction analyses as exploratory.
  5. Bring clinical, imaging, electrophysiological, statistical and ethics reviewers together before producing a traceable report and subsequent protocol.

Further validation could use matched contact time and a factorial design with explicit sham procedures to separate stimulation and training components, while recording equipment and staffing costs. Longer follow-up can examine home activity, participation and maintenance. Closed-loop modulation and predictive biomarkers are future research directions, not demonstrated clinical capabilities or completed achievements of this trial.

Frequently asked questions

How can QSevidence support comparison of robotics and stimulation studies?

Build an evidence matrix covering populations, procedures, training dose, sham conditions and endpoints, retaining guideline text and comparator details. Professionals interpret package effects using bias assessment, feasibility and verified data.

How can QSevidence organize connectivity and functional-score relationships?

Map regions, sample sizes, visits and statistical corrections to functional endpoints for imaging and statistical review. Mechanistic interpretation then draws on study design and additional validation evidence.

How can QSevidence support a subsequent trial?

Organize matched-dose comparators, primary endpoints, missing-data plans and cost measures into review questions. Clinical, statistical and ethics teams confirm the formal protocol.

Sources and research notice

This page supports evidence-based research discussion. Qualified teams assess eligibility for brain stimulation, equipment training, imaging and electrophysiology; it is not individual treatment or home-use guidance.