Quantile Regression in Improving the Visualization and Estimation of the Relationship Between Risk Factors and Outcome Measures

• Activity Number: 79
• Type: Contributed
• Date/Time: Sunday, August 4, 2013 : 4:00 PM to 5:50 PM
• Sponsor: Section on Risk Analysis
• Author(s): Wanying Li*+ and Michael D. Lock and Douglas Haney
• Companies: and Crescendo Bioscience, Inc. and Crescendo Bioscience, Inc.
• Keywords: quantile regression ; outcome measure prediction ; nonparametric method ; risk analysis ; clinical research

Abstract

In biomedical studies, the risk of adverse outcome is often estimated by logistic regression methods or other generalized linear models. Modified empirical cumulative probability plots are sometimes overlaid to investigate differences between the distributions of the outcome measures at different levels of the risk factors. There is a recognized gap between the assumptions required by the modeling and the features that are made apparent through data visualization. Additional challenges arise in a research setting when the underlying relationship is the main objective of the study, and the assumptions of logistic regression model may be violated. This paper presents an application of quantile regression techniques, from nonparametric to nonlinear quantile models, in a clinical research study to explore the relationship between a multi-biomarker disease activity (Vectra DA algorithm) score (range 1-100) and structural joint damage in patients with rheumatoid arthritis. We successfully identified a score range for which the risk of joint damage is elevated, something that could help define thresholds to guide clinicians when considering various treatment interventions.