Online Program Home
My Program

Abstract Details

Activity Number: 526 - Contributed Poster Presentations: Section on Statistical Consulting
Type: Contributed
Date/Time: Wednesday, July 31, 2019 : 10:30 AM to 12:20 PM
Sponsor: Section on Statistical Consulting
Abstract #304766
Title: Using Linear Discriminant Analysis to Classify Patients with Variant Von Willebrand Disease (VWD)
Author(s): Ke Yan* and Jonathan C Roberts and Robert Montgomery and Pamela Christopherson and Pippa Simpson
Companies: Medical College of Wisconsin and Bleeding & Clotting Disorders Institute and Blood Center of Wisconsin and Blood Center of Wisconsin and Medical College of Wisconsin
Keywords: Multiclass classification ; Linear discriminant analysis

Introduction: 134 patients with VWD, a common inherited bleeding disorder, were evaluated. There are different types of VWD depending on the VWF activity defect, and a novel assay to discriminate the various activities of VWF was developed. Of interest was whether this assay was useful in correctly classifying different variants of VWD. Methods: Classification tree analyses can be very tedious and difficult to interpret using one-vs-the rest or all-vs-all others algorithm. Multinomial logistic regression analyses results were unstable because of the quasi-complete separation problem. Linear discriminant analysis (LDA) was ultimately used with centered quadratic terms to account for non-linearity and multicollinearity in the data. The classification results using the study assay were compared to the actual clinical diagnoses. And the final model was cross-validated using the leave-one-out algorithm. Results: We were able to classify patients with an overall accuracy of 93.3% (125 out of 134). When using the leave-one-out algorithm, 118 out of 134 (88.1%) patients were correctly classified. Conclusion: LDA can be very useful in dealing with multiclass classifications.

Authors who are presenting talks have a * after their name.

Back to the full JSM 2019 program