Comparing Response Scaling Formats Used in Patient-Reported Outcome (PRO) Instruments

Abstract

Background: Commonly used response scales in patient-reported outcome (PRO) measures include the visual analogue scale, 11-point numeric rating scale, 5-point numeric rating scale, 5-point verbal rating scale, and 5-point verbal-numeric rating scale. Although prior studies have explored the interpretation of response scale labels and compared scores resulting from the response scale, many questions remain. Purpose: To identify sets of verbal descriptors interpreted with the least variation and to explore whether the response scales provide interval level data and whether the scales are interchangeable. Methods: Subject recruitment and screening was through an online drug-drug interaction service (MediGuard.org). Via an online survey platform, subjects used a scale (0 = lowest possible and 10=highest possible) to assign interpretation scores to verbal descriptors. Repeated measures analysis of variance informed the test interval data between scores. Subjects also completed repeated administrations of four symptom-specific item stems with different response scales. Ordinal regression informed the analysis of scores assigned to verbal descriptors, comparison of probabilities of responding in given categories across scales, and prediction of response category on one scale conditional on observed response on another scale. Cut-points informed tests for interval level data. Results: The sample (n=350) comprised 223 females and 127 males with a mean (SD) age of 56.9 (12.1) years. Number of health conditions per subject ranged from 1 to 12 (median = 5). Age, sex, level of education, and number of health conditions were associated with the interpretation of verbal descriptors. Scores assigned to "poor," "fair," "good," "very good," "excellent," "somewhat," "sometimes," and "quite a bit," had the largest variation. The probability of responding in the same categories on the different response scales was significantly different across scales before and after collapsing categories. No scale yielded interval level data. The 11-NRS data tended more towards interval level than the data from other scales. Conclusions: Using different response scales with verbal descriptors in non-randomized studies may introduce bias. Differential item functioning and subgroup analyses should be investigated in the development and use of these response scales. The scales are not interchangeable. Compared with other scales the 11-NRS produced data approaching interval level. Collapsing categories entails significant probabilities of misclassification.