A Bayesian Ordinal Logistic Model for Evaluating Error in Color Chart Based Diagnosis of Airway Bleeding in Racehorses

• Activity Number: 218 - Contributed Poster Presentations: Section on Statistical Consulting
• Type: Contributed
• Date/Time: Tuesday, August 4, 2020 : 10:00 AM to 2:00 PM
• Sponsor: Section on Statistical Consulting
• Author(s): Clark Kogan* and Warwick Bayly and Renaud Léguillette and Linnea Warlick
• Companies: Washington State University and Washington State University and University of Calgary and Washington State University
• Keywords: bayesian; ordinal; logistic; diagnosis

Abstract

Exercise-induced pulmonary hemorrhage (EIPH), or bleeding in the lower airway, is a common condition that occurs in racehorses following vigorous exercise. Severe EIPH can negatively impact performance if not properly diagnosed and treated. EIPH diagnosis is typically made by either tracheobronchoscopic examination (TBE) or bronchoalveolar lavage (BAL). The BAL procedure involves counting red blood cells (RBC), with a microscope, from fluid infused into and then aspirated from the lower airway; however, estimates can be procured at the racetrack by comparison of BAL fluid to a pre-calibrated color chart. Design for discrete-style color charts involve choosing the total number and shade of colors; the latter involves balancing simplicity (fewer colors) with accuracy (more colors). In designing a colorimetric chart, it is advantageous to have a model for accuracy as a function of the design parameters. We use in-vivo data from post-race BAL tests where RBC estimates were made both via microscope counts and via color chart from multiple individuals to parameterize a Bayesian ordinal logistic regression model. The model is used to help inform diagnosis and evaluate color chart design.