Multilevel Generalized Linear Mixed Model: Importance of Adjusting for Correlated Outcomes

• Activity Number: 195
• Type: Contributed
• Date/Time: Monday, July 30, 2012 : 10:30 AM to 12:20 PM
• Sponsor: Section on Statistics in Sports
• Author(s): Kaitlyn Fernandez*+
• Companies: Rho, Inc.
• Keywords: Correlated outcomes ; Sports Injury

Abstract

Clustered binary responses arise frequently in clinical practice; however, many analyses fail to utilize techniques appropriate for the analysis of correlated outcomes. Failure to accommodate for non-independent responses results can lead to incorrect conclusions because of biased standard errors.

This study examines how different levels of clustering can affect model estimates. The study population consisted of over 5,000 men and women from the United States Naval, Marines, or Air Force Academies. Clustering levels under investigation included subject within commanding officer within academy, subject within academy (ignoring commanding officer), and subject within commanding officer (ignoring academy) while exploring many biomechanical risk factors for ACL tears using a multilevel generalized linear mixed model approach. All three of the models, accounting for differing levels of clusters, were then compared to a naïve logistic regression which ignored clustering and non-independence assumptions. The resulting model comparisons highlight the importance of correctly constructing analyses that account for correlated outcomes.