Project Summary:

Respiratory support methods as described by current ontologies are often too granular to contain distinctions that are unimportant for clinical prediction models. We proposed a classification scheme and heuristic for implementing the scheme in real EHR data.

Relevant Skills:

• Clinical Medicine
• Ontologies
• Machine Learning (scikit-learn, XGBoost)
• Data visualizations (Matplotlib)
• Feature Importance and Explainability (SHAP)

Validation:

The classification scheme was validated and feature were shown to be useful by performing an ablation analysis and predicting 30 day mortality for all COVID-19 tested patients seen in the ED both with and without the respiratory information.

Citation:

Sean C Yu, Mackenzie R Hofford, Albert M Lai, Marin H Kollef, Philip R O Payne, Andrew P Michelson, Respiratory support status from EHR data for adult population: classification, heuristics, and usage in predictive modeling, Journal of the American Medical Informatics Association, 2022;, ocac005, https://doi.org/10.1093/jamia/ocac005