<b>Predictive Use of Wearable Sensors for Detecting Gait Deterioration in Children with Cerebral Palsy: A Narrative Review</b>

Background: Wearable sensor technologies are increasingly used to quantify gait in children with cerebral palsy (CP) outside laboratory settings, with potential to support earlier identification of clinically meaningful gait decline. Objective: To synthesize recent evidence on wearable sensor modalities, gait parameters, and analytical approaches—particularly machine learning—for monitoring and predicting gait deterioration in pediatric CP. Methods: This narrative review used a structured literature search of PubMed/MEDLINE, Scopus, and IEEE Xplore for English-language, peer-reviewed studies published from 1 January 2019 to 31 December 2025, supplemented by reference-list screening. Studies were eligible if they included children/adolescents with CP and used wearable sensors to quantify gait parameters, validate wearable metrics against clinical or laboratory references, or apply analytical models to classify or predict gait-related outcomes. Results: Twenty-five studies (approximately 1,050 participants) were included. Inertial measurement units were used in 19/25 studies (76%), and 15/25 studies (60%) reported validation against clinical or laboratory reference measures. Wearable-derived gait speed and cadence showed consistent clinical associations, with correlations between IMU-derived gait speed and clinical walking tests ranging from r = 0.72 to 0.91 and test–retest reliability for key parameters ranging from ICC = 0.82 to 0.94. Machine learning was applied in 11/25 studies (44%), typically for gait phase or pattern classification with reported accuracies of 88–96% using internal validation. Only 3/25 studies (12%) evaluated longitudinal prediction of gait deterioration (6–12 months), reporting AUC values of 0.74–0.83 without external validation, limiting certainty. Conclusion: Wearable sensors provide feasible and valid tools for real-world gait monitoring in pediatric CP, particularly for spatiotemporal parameters; however, evidence for predicting gait deterioration is limited and methodologically heterogeneous, with low certainty due to small samples and lack of external validation.