22B. Machine learning approaches to improve prediction of neuropsychological outcome in pediatric cancer groups

Up to 50% of pediatric cancer survivors experience neuropsychological impairments during or after treatment. In clinical practice, personal, medical, and environmental factors are considered during diagnostic formulation, but these are not analyzed with a statistical approach that can identify patients who will have stable versus declining neurocognitive profiles over time. Methods such as latent class modeling or artificial intelligence can provide this subgrouping within neuropsychological assessment. The inclusion of risk and protective factors along with assessment results could provide a personalized predictive tool that identifies which children are at highest risk for future neuropsychological problems and lower quality of life.

The aim of this project is to develop a model that can predict neuropsychological outcomes for an individual child over time, which includes a consideration of medical, individual, and psychosocial factors and advanced statistical methods (e.g., machine learning). These approaches will provide detail on which factors and models can predict poorer neurocognitive outcomes in children with cancer, and can also be used to guide assessment and intervention methods in the future.