Abstract
Functional Independence Measure (FIM) is arguably the main outcome measure in rehabilitation medicine and an important casemix tool. It was devised in 1984 in the USA to be used as universal assessment tool in the Uniform Data System for medical rehabilitation. Its use has since been reported in over one thousand published articles in patients with various pathologies and for various purposes. At the University Rehabilitation Institute in Ljubljana, which is the only rehabilitation hospital in the country and thus provides comprehensive rehabilitation for the whole territory of Slovenia (admitting over 1300 cases per year), compulsory FIM assessment at admission and discharge has been performed (and integrated into the hospital information system) since 2004. FIM consists of the motor and the cognitive subscale with 13 and 5 items, respectively, all rated on a 7-point scale. After numerous publications proved its acceptable reliability and validity within the classical test theory perspective, it has been more recently almost equally widely criticised from the item response theory perspective, especially from the Rasch model perspective. Based on the data from the 2004-2006 period, we previously reported on comprehensive FIM analyses at our Institute (including patient demographics, duration of rehabilitation, other classifications, admission and discharge diagnosis, complications and treatment cessations, and type of admission, treatment and discharge) with emphasis on the general aspects of data visualisation for decision support and healthcare quality monitoring, as well as on modelling FIM patient progress. This paper incorporates data from the 2007-2009 period and focuses more deeply on psychometric issues. Internal validity is tested (in terms of unidimensionality, using Parallel Analysis and the Minimum Average Partial Test, and internal consistency) and the relationship between independence level of single items and the corresponding total subscale scores is assessed using ordinal logistic regression modelling.
Author(s): Gaj Vidmar, Helena Burger