Walton – Can You Validly Use a Total ALSFRS-R Score? A Rasch Analysis to Provide Interval Scale Estimates

Author(s): Young C.A.; Paris-Davies J.; Mills R.J.; Tennant A.

Source: Amyotrophic Lateral Sclerosis and Frontotemporal Degeneration; 2018; vol. 19 ; p. 267

Publication Date: 2018

Publication Type(s): Conference Abstract

Abstract:Background: In Amyotrophic Lateral Sclerosis or Motor Neurone Disease (ALS/MND), the most common scale for measuring functional limitation is the Amyotrophic Lateral Sclerosis Functional Rating Scale, now in its revised format (ALSFRS-R). While the total score from the scale is widely used, recent analyses from both modern and classical perspectives have indicated that a total score is invalid, instead a profile should be reported of the Bulbar, Motor and Respiratory domains (1), or as the four domains of Bulbar, Fine Motor, Gross Motor and Respiratory (2,3). Objective(s): To explore whether a valid total score, ideally with interval scale estimates, can be generated from the ALSFRS-R, using modern test theory. Method(s): Data were collected through a questionnaire pack for self-completion, in the ongoing longitudinal Trajectories of Outcomes in Neurological Conditions (TONiC) study in the UK. Data from the ALSFRS-R were fit to the Partial Credit parametrisation of the Rasch model. Where a breach of the local independence assumption was observed (items that are partly replications of each other), ‘testlets’ were used to absorb the dependencies within the scale. A conditional Chi-Square test of fit (CTF) was reported together with the variance retained in the latent estimate, reported as the Expected Common (nonerror) Variance (ECV). In the literature, an ECV value of 0.90 is regarded as the minimum level consistent with a unidimensional general first factor. Result(s): 636 patients with a mean age of 65.1 years had a median duration since diagnosis of 11.2 months (IQR 4.6-29.9). Classical item-based Rasch analysis of the ALSFRSR total score showed substantial misfit with 16 pairs of items displaying residual correlations above 0.1 (average residual correlation 0.1), indicating local dependency. Only the Bulbar domain showed consistency with Rasch model expectations. Implementing a testlet design for the 12 items, good fit to the model was observed. The CTF was 0.279 while the ECV was 0.91, indicating that just 9% of the variance was discarded to make a unidimensional latent estimate. An alternative item testlet design to the Motor subscale also showed good fit. Discussion and conclusions: The ALSFRS-R is multidimensional to classical and modern test theory. Rasch transformation with testlets provides an adequate solution for both the Total score and Motor subscale; a transformation table can be made available for their resulting rawscores to interval scale estimates. Use of the Respiratory subscale is discouraged as both classical and modern test theory analysis have failed to support its validity.

Database: EMBASE

Leave a Reply

Your email address will not be published. Required fields are marked *