Comparison Between Regression with Latent Variables and Regression with Composite Scores for Subjective Outcomes from Instruments in Clinical Trials and Medical Research

• Activity Number: 75
• Type: Contributed
• Date/Time: Sunday, July 31, 2011 : 4:00 PM to 5:50 PM
• Sponsor: Biometrics Section
• Author(s): Chengwu Yang*+ and Barbara C. Tilley and Anbesaw Selassie and Ruth Greene
• Companies: Penn State University and The University of Texas Health Science Center at Houston and Medical University of South Carolina and Johnson C. Smith University
• Keywords: Latent Variable Analysis ; Multiple-indicator Multiple-causes (MIMIC) Models ; Differential Item Functioning (DIF) ; Patient Reported Outcomes (PRO) ; Structural Equation Modeling (SEM) ; Measurement Bias

Abstract

A conventional way to analyze composite scores from instruments such as the SF-36 is to directly regress them on covariates such as treatment arm. However, this method cannot assess if the instrument's factor structure is sustained and/or if the covariates' effects are contaminated by measurement bias, especially, differential item functioning (DIF). Therefore, results from this method can be invalid. On contrast, the multiple-indicator multiple-causes (MIMIC) models that regress the instruments' latent domain scores on the same covariates can assess both of the instrument's factor structure and possible impacts of DIF. Three existing datasets were analyzed using these two methods, in order to illustrate three different situations when analyzing composite scores: 1) the instrument's factor structure sustained, and there is no DIF; 2) the instrument's factor structure sustained, but there is DIF; 3) the instrument's factor structure didn't sustain. Advantages and disadvantages of the two methods were compared, and strategies to analyze composite scores under different situations were offered.