TITLE:
Applying Item Response Theory Methods to Improve the Measurement of Fatigue in a Clinical Trial of Rheumatoid Arthritis Patients Treated with Secukinumab*
AUTHORS:
Mark Kosinski, Jakob B. Bjorner, Ari Gnanasakthy, Usha Mallya, Shephard Mpofu
KEYWORDS:
Patient-Reported Outcomes; Fatigue; Rheumatoid Arthritis; Item Response Theory; Clinical Trial
JOURNAL NAME:
Open Journal of Rheumatology and Autoimmune Diseases,
Vol.3 No.4,
November
18,
2013
ABSTRACT: Background: Many clinical trials include multiple patient-reported outcomes (PROs) to measure fatigue as secondary or exploratory endpoints of treatment effectiveness. Often, these instruments have overlapping content. The objective of this study was to compare the combined measurement properties of two fatigue scales, the Functional Assessment of Chronic Illness Therapy-Fatigue (FACIT-Fatigue) and SF-36 vitality (VT) scale using item response theory (IRT). Methods: The FACIT-Fatigue and SF-36v2 were administered at baseline and weeks 2, 4, 7, 12, and 16 to rheumatoid arthritis (RA) patients (n = 237) enrolled in a 52-week multicenter, randomized, double-blind, placebo-controlled, parallel-group, dose finding study to evaluate the efficacy and safety of subcutaneous secukinumab administered to pa- tients with active RA. Confirmatory factor analysis (CFA) was used to investigate unidimensionality among FACIT- Fatigue and VT items. A generalized partial credit IRT model was used to cross-calibrate the FACIT-Fatigue and VT items and weighted maximum-likelihood estimation was used to score a composite fatigue index. Analysis of variance was used to compare the composite fatigue index with the original scales in responding to ACR improvement and treatment effects. Results: CFA found less than adequate fit to a unidimensional model. However, specifications of alternative multidimensional models were insufficient in explaining the common variance among items. An IRT model was successfully fitted and the composite fatigue index score was found to be more responsive than the original scales to ACR improvement and treatment effects. Effect sizes and significance tests for changes in scores on the composite index were generally larger than those observed with the original scales. Conclusion: IRT methods offer a promising approach to combining items from different scales measuring the same concept that could improve the detection of treatment effects in clinical studies of RA.