Development and psychometric evaluation of the Flexilevel Scale of Shoulder Function

BACKGROUND: Existing measures of self-reported shoulder function fail to measure effectively the full range of shoulder functioning. The classic approach for improving the reliability of a scale is adding items, but a scale with a substantial number of items imposes a large response burden on participants. A more efficient approach is to use modern psychometric methods to construct an adaptive scale in which patients respond only to items that are targeted at their level of shoulder function. METHODS: We developed a Flexilevel Scale of Shoulder Function (FLEX-SF). This scale includes three testlets that target low, medium, and high shoulder function. Scores on the testlets were equated to a common mathematical metric. DESIGN AND SUBJECTS: We developed an initial pool of 68 items. This pool was administered to 400 patients, and responses were calibrated using a rating scale model. Subsets of items were identified for an easy, medium difficulty, and hard testlet. Properties of the scale were evaluated in a 3-month longitudinal study of 200 shoulder patients. RESULTS: The FLEX-SF exhibited high reliability at both the scale level (intraclass correlation coefficient [3,1] = 0.90) and specific trait levels. The validity of the FLEX-SF was supported by its internal and external responsiveness (Guyatt responsiveness index = 1.12) and the pattern of its associations with other health status measures. CONCLUSIONS: The FLEX-SF can be used as a primary endpoint in clinical trials even when there are relatively few people in each treatment group. The scale also has excellent properties for use in clinical settings tracking individual changes over time.