Application of an Item Response Theory model to describe Amyotrophic Lateral Sclerosis Functional Rating Scale data
Åsa M. Johansson (1), Inez de Greef (2), Bernard Muller (2), and Piet H. van der Graaf (1)
(1) Leiden Academic Centre for Drug Research, Leiden University, Leiden, The Netherlands, (2) Treeway B.V., Rotterdam, The Netherlands.
Objectives: To develop an Item Response Theory (IRT) model which describes Amyotrophic Lateral Sclerosis Functional Rating Scale (ALSFRS) data. The IRT model will, in a later stage, be used in a disease progression model for Amyotrophic Lateral Sclerosis (ALS).
Methods: The severity of ALS is assessed through the ALSFRS, consisting of 10-12 tasks with 5 levels of fulfilment (0-4). IRT can be applied to ALSFRS data to link the probabilities to complete the different tasks (to certain extents), to the patient’s disability level (disease stage) [1]. The disease stage is a latent trait, which is only assessable through the evaluation of the data. The latent trait is assumed to be normally distributed in the population. The IRT model can be combined with a pharmacometric model to describe how the disease progresses over time [2].
ALSFRS data from 4,838 ALS patients were accessed through the Pooled Resource Open-Access ALS Clinical Trials (PRO-ACT) database. A total of 42,775 evaluations of ALSFRS were available, and each evaluation was regarded as a separate individual in the fitting of the IRT model. The individuals were divided into three groups, based on the location of their first symptoms: limb onset, bulbar onset, and combined limb and bulbar onset. Separate sets of parameters (probability curves) were estimated for the three onset groups. A homogeneous graded response model [3], and different versions of a heterogeneous graded response model [4], were fitted to the data. The models were evaluated based on objective function values (OFV), distribution of the latent trait variable (should be approximately normally distributed), and visual comparisons between predicted and observed probability curves.
Results: The homogeneous graded response model resulted in a skewed distribution of the latent trait variable, and the visual comparison of predicted and observed probability curves revealed a model misspecification at later stages of the disease. The final heterogeneous graded response model resulted in a significantly lower OFV, and a normally distributed latent trait variable.
Conclusions: An IRT model describing ALSFRS data was developed. The IRT model will be used in a disease progression model for ALS, to evaluate the effect of new therapies.
References:
[1] Nering M.L., Ostini R. Handbook of polytomous Item Response Theory models. Taylor & Francis Group, 2010.
[2] Ueckert S., et al. Pharmaceutical Research. 2014; 31(8): 2152-2165.
[3] Samejima F. Psychometrica. 1969; Psychometric Monograph 17.
[4] Samejima F. Psychometrica. 1995; 60: 549-572.
