Objective: To assess the four approaches to applying Rasch analysis to ordinal, repeated measures: (1) to perform separate Rasch analyses; (2) to perform a single analysis by treating the subjects measured on different occasions as distinct ones; (3) to perform a single analysis by treating items used on different occasions as distinct ones; and (4) to construct a 3-facet model by adding occasions as the third facet.
Design: Various Rasch and 3-facet models applied to retrospective, cohort data.
Setting: A stroke program at a rehabilitation hospital.
Patients: 99 of the 107 consecutive patients admitted between March and October, 1990, average age 69 years, 40% of whom were women, with an average stay of 54 days.
Intervention: Medical inpatient rehabilitation.
Main outcome measure: The Functional Independence Measure (FIM), administered within 1 week of admission and discharge, and approximately 6 weeks after discharge.
Results: The four approaches resulted in various occasion-specific and generalized measures of item difficulty, patient ability, and occasion severity. Although these measures turned out to be congruent with each other as they should, they generated different sources of information needed for formative and summative evaluation of health care outcomes. All four approaches were found to be useful, although each had its own limitations. A four-step strategy based on these approaches was therefore designed and recommended.
Conclusion: Rasch analysis of ordinal, repeated measures poses a special challenge to those interested in health care outcomes. A multi-method strategy incorporating various approaches is therefore needed to optimize its use in health care research.