A Framework for Analysis, Ontological Evaluation, and Visualization in Preparation to Predictive Analytics in Pediatric Brain Tumor Research

We provide a generalizable framework for the systematic analysis of complicated, longitudinal clinical features in pediatric cancer. We use a threefold pipeline of exploratory data analysis, ontological categorization through a multi-modal data transformation process towards predictive analytics. We derive a data-driven phenotype from a subset of a sample of over 1900 brain tumor cases focused specifically on High-Grade Gliomas. We implement an analyst-friendly process to make machine learning-ready data sets based on domain ontologies ready for enumeration and vectorization. The results are clinical domain expert readable data points from 4.3 million observational events across 16,000 patient days. In this research, we address the gap in phenotypic data features by utilizing extensive harmonized observational clinical data and identify resources and specific processes for their use in rare tumor research.