Longitudinal and dynamic measurement invariance of the FACIT-Fatigue scale: an application of the measurement model of derivatives to ECOG-ACRIN study E2805.

Author(s): Estabrook R,  Cella D,  Zhao F,  Manola J,  DiPaola RS,  Wagner LI,  Haas NB

Journal: Qual Life Res

Date: 2018 Jun

Major Program(s) or Research Group(s): NCORP

PubMed ID: 29508208

PMC ID: PMC6004788

Abstract: PURPOSE: While quality of life measures may be used to assess meaningful change and group differences, their scaling and validation often rely on a single occasion of measurement. Using the 13-item FACIT-Fatigue questionnaire at three timepoints, this study tests whether individual items change together in ways consistent with a general fatigue factor. METHODS: The measurement model of derivatives (MMOD) is a novel method for measurement evaluation that directly assesses whether a given factor structure accurately describes how individual test items change over time. MMOD transforms item-level longitudinal data into a set of orthogonal change scores, each one representing either a within-person longitudinal mean or a different type of longitudinal change. These change scores are then factor analyzed and tested for invariance. This approach is applied to the FACIT-Fatigue scale in a sample of patients with renal cell carcinoma treated on 'ECOG-ACRIN Cancer Research Group (ECOG-ACRIN) study 2805. RESULTS: Analyses revealed strong evidence of unidimensionality, and apparent factorial invariance using traditional techniques. MMOD revealed a small but statistically significant difference in factor structure ([Formula: see text], [Formula: see text]), where factor loadings were weaker and more variable for measuring longitudinal change. CONCLUSIONS: The differences in factor structure were not large enough to substantially affect scale usage in this application, but they do reveal some variability across items in the FACIT-Fatigue in their ability to detect change. Future applications should consider differential sensitivity of individual items in multi-item scales, and perhaps even capitalize upon these differences by selecting items that are more sensitive to change.