Comparing exercise with virtual reality gaming on gait and cognition in relapsing-remitting multiple sclerosis: a randomized controlled trial

Background: Exercise and virtual reality gaming may mitigate gait and
cognitive deficits in relapsing-remitting multiple sclerosis (RRMS). The
main aim was to compare the efficacy of both interventions on gait and
cognition and gait in RRMS. Secondary aims were to explore the efficacy of
both interventions on serum biomarkers and to explore the predictors of
treatment response. Methods: Forty-eight participants with RRMS were
randomized to exercise (n=19), VR (n=19), or wait-list control (n=10) for
eight weeks. Primary outcomes were the 10-meter walk test (10MWT) and the
Symbol Digit Modalities Test (SDMT). Secondary outcomes included serum
levels of neurofilament light chain (NfL), brain-derived neurotrophic
factor (BDNF), and insulin-like growth factor-1 (IGF-1). Extreme Gradient
Boosting (XGBoost), Random Forest, and logistic regression models were
trained to predict treatment response. Results: The exercise group
improved 10MWT performance by 2.41 seconds and increased IGF-1 levels by
100.25 ng/ml, significantly more than the VR and control groups (both
p<0.001). The VR group improved on the SDMT by 1.95 points (p=0.001
vs. control; p=0.05 vs. exercise). Both interventions reduced NfL
concentrations compared to control (exercise: –2.07 pg/ml; VR: –0.60
pg/ml), with exercise showing a greater reduction than VR (p=0.02).
XGBoost demonstrated highest predictive accuracy (10MWT: 87%; SDMT: 86%).
SHapley Additive exPlanations (SHAP) analysis identified baseline IGF-1
and BDNF as top predictors of 10MWT, and baseline CognICA, BDNF, and age
as predictors of SDMT performance. Conclusion: Exercise preferentially
improves gait and IGF-1, whereas VR gaming yields modest cognitive gains.
Serum biomarkers enhance machine learning prediction of treatment
response, supporting a precision rehabilitation approach in RRMS.
Keywords: Multiple Sclerosis, Virtual Reality, Exercise, Gait, Cognition,
Machine Learning