Towards Patient-Driven Phenotyping and Similarity for Precision Medicine

Clinical phenotyping provides important insight into the manifestation and outcome of rare and complex diseases. Traditional phenotyping techniques often require multiple iterations of refinement with a domain expert, lack interoperability, and have limited reproducibility. In comparison, patient similarity-based techniques derive personalized patient risk models that are highly accurate, even when applied to sparse data or poorly characterized diseases/outcomes. We introduce a novel, semi-supervised data-driven method for applying clinical similarity to pediatric phenotyping.