Purchase this article with an account.
Trefford Simpson, Marc M. Schulze, Paul Stolee; Rasch Analysis of Clinical Grading of Corneal Staining. Invest. Ophthalmol. Vis. Sci. 2011;52(14):1967.
Download citation file:
© ARVO (1962-2015); The Authors (2016-present)
To use Rasch Analysis to examine clinical grading of corneal staining extent and type (and their product) determining whether grading is a measurement (is it in accord with Rasch theory?) and if so, from the category curves, is there an optimum number of categories for the staining grading scale.
10 subjects experienced in clinical grading but naive about the intent of the experiment graded 50 corneal staining images preselected to range from none to moderate corneal staining, using a 0-100 point scale. They rated the extent and type separately and Rasch analysis was performed on ratings divided into 20, 10 and 5 bins, using Winsteps software on extent and type and on their product.
Generally raters behaved similarly and the images too were rated as though they were a similar unidimensional latent variable (based on infit and outfit statistics and principal component analysis of residuals). For the 20 bin staining type scale, however, 3 of the 10 raters performed as though they were NOT rating along a single continuum (all outfit Z > +/-2.0). For extent and type, clear separation of measures occurred using scales with 10 categories. Better separation of categories occurred with 5 steps when the combination scale was analyzed (fig.; each line represents a category).
Rasch analysis shows that the clinical grading of staining is a "measure" inasmuch as it meets requirements based on Rasch theory. When 10 categories of extent and type are used, the measurements are well behaved and when type and extent (2 ‘different’ latent variables) are combined, the combination still appears to be unidimensional but fewer categories appear to be optimal.
This PDF is available to Subscribers Only