Data-driven upper limb sensorimotor recovery prediction in patients with stroke: A prospective cohort study

Stroke is a major cause of long-term disability globally, with upper limb sensorimotor deficits being among the most debilitating consequences. Accurate prediction of recovery trajectories can help clinicians individualise rehabilitation planning and optimize resource allocation, particularly in low- and middle-income countries (LMICs), where access to high-cost neurophysiological or imaging tools is limited. This protocol describes a prospective cohort study designed to develop and validate both conventional statistical and machine learning-based models for predicting upper limb sensorimotor recovery and quality of life at three and six months post-stroke. One hundred first-ever stroke survivors will be recruited and assessed using a standardised measurement tool consisting of demographic details and motor, sensory, and quality-of-life domains and complemented by 3D kinematic and video-based movement analysis. Statistical and machine learning algorithms (multivariable regression, logistic regression, random forest, support vector machine, and deep neural networks) will be trained and validated to forecast outcomes. This protocol aims to establish an accessible and scalable prediction framework to guide post-stroke rehabilitation strategies.•Development and validation of a multimodal, data-driven model for upper limb sensorimotor recovery prediction.•Multidomain and 3D kinematic variables are combined for comprehensive prognostication.•This provides a scalable, low-cost approach suitable for diverse rehabilitation settings.