|
Activity Number:
|
621
- Beyond Linear Regression: Nonlinear Association, Quantile Regression and Generalized Linear Models
|
|
Type:
|
Contributed
|
|
Date/Time:
|
Thursday, August 1, 2019 : 8:30 AM to 10:20 AM
|
|
Sponsor:
|
Section on Statistics in Epidemiology
|
|
Abstract #302933
|
Presentation
|
|
Title:
|
Methods to Study Thresholds of Hematocrit That Impact Blood Transfusion in Cardiac Surgery
|
|
Author(s):
|
Xiaoting Wu* and Chang He and Donald Likosky
|
|
Companies:
|
University of Michigan and The Michigan Society of Thoracic and Cardiovascular Surgeons Quality Collaborative and Univeristy of Michigan
|
|
Keywords:
|
Splines;
Regression Models;
Non-linearity ;
Breakpoints
|
|
Abstract:
|
The risk of blood transfusion during cardiac surgery increases with low preoperative hematocrit- a continuous measurement of volume proportion of red blood cells in whole blood. The thresholds of hematocrit that impact blood transfusion can provide guidance for blood reservation strategies, however these breakpoints remain obscure. Previous studies either used a linear form of hematocrit to model its relationship with blood transfusion or chose break points to categorize hematocrit without validated data evidence. Many other options, such as transformation and restricted cubic splines, could be considered to model a non-linear relationship; these model estimates however are hard to be interpreted clinically. Linear splines can be used to assess potential breakpoints of the data and provide meaningful results to guide clinical practice. This paper compares methods to model non-linearity in a multivariable regression model. We demonstrate the use of linear splines to identify hematocrit thresholds for increasing risk of blood transfusion. We further derive inference of hematocrit in linear spline terms and its corresponding blood transfusion risk.
|
Authors who are presenting talks have a * after their name.
