OBJECTIVES: Ordinal scales with discrete response options are commonly used in assessment of patient’s experience with a given health condition or treatment. Close conceptual spacing within response sets can interfere with translation during linguistic validation (LV), reduce comprehension for patients, and impact data validity. Traditional LV makes qualitative assessments of translations via cognitive debriefing to generate quality translations that are faithful to source content, but these qualitative assessments can’t diagnose or remedy poorly designed response sets. The present research leverages patient ratings of commonly used response options to quantitatively assess and visualize their distribution in conceptual space, and to provide data to aid in the selection of conceptually distinct response options.

METHODS: 75 English-speaking participants from 12 locales (see Table 1) rated between 1 and 3 response options from a set of 15 (see Table 2) on a VAS scale, with 0 representing a minimum severity rating and 100 representing a maximum. Participants responded to prompts such as those in Figure 1.

RESULTS: Figure 2 shows rating distributions for the 15 response options. Distributions for low severity response options ‘barely’, ‘a little’ and ‘mildly’ were clustered and highly variable. Distributions for high severity response options were also clustered; these results suggest conceptual overlap of response options and patient confusion. ‘Moderately’ and ‘somewhat’, typical scale midpoints, showed different rating distributions, suggesting poor fit for ‘somewhat’ as a midpoint.

CONCLUSIONS: These results indicate conceptual clustering of response options that are intended to be distinct, as well as high-variability rating distributions possibly indicating poor target concept comprehension. Furthermore, when combined with planned research on additional response options and concepts (such as ‘frequency’ or ‘interference’), they can guide selection of maximally distinct and comprehensible response sets for scales or provide insight for post-hoc analyses of data collected via large and possibly problematic response sets.