Date of Award
2018
Embargo Period
1-1-2018
Document Type
Dissertation
Degree Name
Doctor of Philosophy (PhD) in Health & Rehabilitation Science
College
College of Health Professions
Abstract
This study utilizes electronic health record data from the Medical University of South Carolina’s intensive care units as the basis for this Monte Carlo simulation study— which compares four methods for handling missing SOFA scores, both at the composite and component levels. The four methods examined herein include: complete case analysis, median imputation, zero imputation (the method recommended by the creators of the SOFA score), and multiple imputation. This study found that zero imputation introduced the most bias across all three outcomes studied, and therefore is not recommended. Complete case analysis, or ignoring missing data, caused varying amounts of bias—as did median imputation. Multiple imputation, on the other hand, performed well for all three outcomes studied, both at the composite and component levels, demonstrating this method’s superior value in the presence of missing SOFA scores.
Recommended Citation
Brinton, Daniel Lee, "Missing Data Methods for ICU SOFA Scores in Electronic Health Records Studies: Results from a Monte Carlo Simulation Study" (2018). MUSC Theses and Dissertations. 311.
https://medica-musc.researchcommons.org/theses/311
Rights
All rights reserved. Copyright is held by the author.