AI-Driven Identification of Success Factors in Children with Cerebral Palsy

Sammanfattning

Cerebral palsy is the most common cause of physical disability in childhood, yet the relationship between rehabilitation exposure and motor development remains in- completely understood. This thesis investigates factors associated with functional improvement in children with cerebral palsy using longitudinal data collected through a life-mapping questionnaire. The questionnaire captured retrospective information on therapy participation, medical interventions, and motor development across four age intervals in children aged 1–16, reported by parents and guardians. At the time of analysis, the dataset consisted of 34 participants. A multi-stage analytical pipeline was developed, including statistical group compar- isons, linear regression, dimensionality reduction and clustering, and milestone timing analysis. Structured motor outcome scores were constructed to quantify milestone at- tainment and impairment burden. Additionally a locally deployed large language model (LLM) was applied to free text responses to generate narrative-adjusted motor scores, enabling direct comparison with rule-based structured scores. Across most analytical approaches, results indicated a positive directional relationship between rehabilitation exposure and motor development outcome. Children accumu- lating higher training volumes tended to demonstrate more favourable developmental trajectories and earlier achievement of independent walking. However, effect sizes were small and no result reached statistical significance, reflecting the limited sample size and observational study design. LLM-derived scores recovered additional clini- cally relevant information from parent narratives. The developed pipeline provides a scalable foundation for future analyses as addi- tional data is collected.