Innovative Measures of Patient and Disease Phenotyping: Optimizing Linguistic and Machine Learning Techniques in the Investigation of Electronic Health Record (EHR) Data

Abstract

The complexities of understanding symptomatology, disease progression, patient susceptibility to disease, and the general intricacies of diagnostic medicine challenge the everyday work of medical professionals, negatively impacting patient care and leading to practice variation, clinical error, and unnecessary expense. Phenotyping diseases and defining patient populations is time-consuming and is often impractical with existing tools resulting in slow progress to ascertaining understanding of the diseases and the risk of further disintegration of condition or additional system dysfunction. Although there have been many attempts to leverage electronic health record (EHR) data in the development of various machine learning (ML) models, they have been unable to deliver success in all parameters necessary for effective implementation in a clinical context: performance, transparency, trustworthiness, and interpretability. We introduce a novel ML approach driven by transdisciplinarity and intensive human feedback that offers a resolution to this problem, in which linguistic feature engineering, competitive modeling, and continual human validation delivers the success needed to improve clinical pathways.

Tasks

Reproductions