Interpretation of verbal descriptors for response options commonly used in verbal rating scales in patient-reported outcome instruments.

purposeTo assess the variation in the interpretation of common verbal descriptors (VDs) used in response scales and examine factors associated with those interpretations.

methodsSubjects were recruited through MediGuard and they assigned interpretation scores (11-point scale; 0 = lowest possible, 10 = highest possible) to five common sets of VDs: set one (none, mild, moderate, severe, very severe); set two (never, rarely, sometimes, often, always); set three (poor, fair, good, very good, excellent); set four (not at all, a little bit, moderately, quite a bit, extremely); and set five (not at all, a little bit, somewhat, quite a bit, very much). One-sample test for proportions and T-tests examined equality of proportions (anchors) and means scores (non-anchors) with the fixed intervals (0.0, 2.5, 5.0, 7.5, and 10.0). Ordinal regression examined adjusted associations between demographic/clinical factors and VD scores.

resultsOf the 350 subjects, 68 % were females and mean (SD) age was 56.9 (12.1). Two sets had two VDs with mean (95 % CI) scores not different than the fixed intervals. Set one had mild = 2.50 (2.33; 2.66) and moderate = 5.01 (4.89; 5.13) with 98.8 % (97.3 %; 100 %) assigning none = 0. Set five had a little bit = 2.35 (2.17; 2.53) and quite a bit = 7.65 (7.43; 7.87) with 95.0 % (95 % CI 91.7; 98.2) assigning not at all = 0. Significant associations (p ≤ 0.05) included age and education with somewhat and income and comorbidities with very severe. Age, sex, and education showed associations with other VDs albeit in nonsignificant models.

conclusionsSets one and five yielded data closest to the fixed intervals. Demographic and clinical factors are associated with the interpretation of some VDs and should be adjusted for in analyses of non-randomized data.