An Evidence-Based Machine Learning System for Ankle-Foot Orthosis Prescription and Patient Compliance Monitoring: A Dual-Dashboard Clinical Platform

Background: Ankle-foot orthosis (AFO) prescription for post-stroke foot drop lacks a standardised objective decision-support tool. Up to 41% of prescribed AFOs are abandoned within 12 months. The first machine learning study predicting AFO need was published in 2021 [1]. Methods: A six-class ML classifier was trained on 4,000 synthetic records using 17 clinical features derived from the Choo et al. (2021) feature schema [1]. A stiffness scoring algorithm was implemented based on published evidence. A QUEST-aligned patient satisfaction assessment and rule-based compliance risk profiling were integrated. The dual-dashboard system serves physiotherapists (prescription, stiffness, risk) and patients (recommendation, education, wearing log). Results: LightGBM achieved macro-average ROC-AUC 0.987 across six AFO classes (PLS, Solid Ankle, Hinged, Carbon Fibre, GRAFO, FES). The stiffness algorithm demonstrated face validity across 20 synthetic case profiles. Compliance risk profiling achieved 100% sensitivity on 50 synthetic high-risk profiles. Conclusion: A dual-dashboard AFO platform achieves technically robust six-class classification performance. The system addresses five documented gaps in AFO prescription and monitoring practice. Validation against real orthotist prescription decisions is required.