Rasch analysis of the evaluation in Ayres sensory integration (EASI)

Abstract

Sensory Integration (SI) refers to the neurological process by which a person takes in sensory information, interprets this information, and uses it to inform movement and goal-directed action (Ayres, 1989). For children with a variety of diagnoses, as well as some children who are otherwise typically developing, SI may be impaired or delayed (Bundy & Lane, 2020a). Occupational therapists and other clinicians who treat children with sensory integration (SI) dysfunction face a dearth of appropriate instruments to evaluate SI function. The Evaluation in Ayres Sensory Integration (EASI; Mailloux et al., 2018) is a novel assessment tool that may allow therapists to evaluate SI in a way that is aligned with SI theory. EASI consists of 21 individual tests that measure constructs of SI. The EASI authors have collected normative data for children across the globe. However, this data must be assessed for validity and reliability before it can be used as the basis for normative scoring on the EASI. In this dissertation, I used Rasch analysis to evaluate data with 19 of the 21 tests. The Rasch model is a latent-trait psychometric model that (1) transforms ordinal-level data to interval-level data and (2) allows users to assess evidence for construct validity (unidimensionality and invariance) and internal reliability (Bond, Yan & Heene, 2020). For each of the 19 tests, I evaluated item fit statistics, rating scale fit statistics, person fit statistics, principal components analysis (PCA) of standardized residuals, differential item functioning (DIF) based on sex, person reliability index and strata. The dataset for this study comprised 2653 children from 51 countries; all data were collected by trained EASI examiners. Overall, results revealed promising evidence for construct validity and internal reliability of data collected using 16 of the19 EASI tests evaluated in this dissertation. However, across many tests, I observed lower-than-desired person fit statistics and reliability. Notably, these results were not far from the desired values. I hypothesized that these findings are the result of the overall high ability level of the normative population. EASI is designed to evaluate children with lower-than-average or poor SI function; therefore, these findings are not unexpected nor are they particularly concerning. Three EASI tests (Proprioception: Force, Proprioception: Joint Position, and Balance) required substantial revision as a result of these analyses; each had threats to construct validity that exceeded my expectations. In this dissertation, I provided potential solutions for these three tests; future studies will evaluate the extent to which these solutions resolve concerns. In conclusion, the normative data form an acceptable basis for creating norm-based scores for clinical interpretation. However, larger studies must be conducted with clinical populations to ensure that the tests can differentiate between children with and without SI dysfunction. Further, future studies should investigate the role of culture, language and other factors on the validity of EASI test scores.