Par10: Implementing Rasch Analysis in Psychometric Evaluation of Patient-Physician Interaction Scale

Abstract

OBJECTIVES: The study involved validating the scaling properties of patient‐physician interaction scale in a pulmonary specialty clinic using a clinic‐specific scale through the implementation of Rasch analysis. DATA: Cross‐sectional data from 65 adult asthma patients at the University of Illinois Asthma clinic was used. Physician interaction was measured using eight Likert‐type items. Patients responded from “strongly disagree” (1) to “strongly agree” (4). METHODS: Scaling properties were assessed by investigating its fit to a Partial Credit Rasch Rating scale model that enabled item‐by‐item analysis. Winsteps® was used for analysis. Model determined scale robustness in terms of unidimensionality, additivity and functioning of the rating scale. RESULTS: Analyses found person separation index of 1.86 with reliability of 0.77. The mean patient measure (0.69 logit) was greater than mean difficulty of items (0.00 logit) implying patient ability was greater than item difficulty. On average, ordering of items found item ‘physician instructed patients on home steroid treatment’ (STETX) had the highest logit measure of 0.69, however it misfitted the model. Item ‘physician asked about smoking habits’ (ASKSMOK) had lowest measure of −0.70 logits. Fit statistics revealed high infit and outfit mean square (MNSQ) values (>1.4) for 16 patients. Seven items had MNSQ values within desired range of 0.6–1.4. CONCLUSIONS: Items exhibited adequate reliability in separating persons, but they displayed ceiling and floor effects in measurement. Some evidence of construct validity was established since only one item misfitted the model. Item STETX was the most difficult to endorse (higher on construct), since it might not be applicable to all, but severe asthmatics. Item ‘ASKSMOK’ was easiest to endorse, probably because it is a standard asthma care question. Nevertheless, misfitting persons implied inappropriate measurement of some patient attitudes. Thus, some plausibility in the unidimensionality and validity of the scale existed, and it exhibited moderate scaling properties.