Items, Instruments, Crosswalks, and PROMIS

In the English language, there are 165 published questionnaire instruments intended to assess “disability” or “physical function” health outcomes, containing 1860 items1. If “health related quality of life” (HRQOL) questionnaires are included, the numbers are yet larger. With integration of the new sciences of Item Response Theory (IRT)2 and Computerized Adaptive Testing (CAT)3 into instrument development, the numbers of items and instruments could again grow larger. This unbounded proliferation of health status instruments is problematic and raises both serious issues and intriguing opportunities. For understanding these issues, some knowledge of IRT and CAT is required. IRT2,4 works at the level of the specific item, which has measurable characteristics such as information content, degree of difficulty, reliability, clarity, ease of translation, performance in different populations, importance to the subject, and others. IRT is sometimes termed “latent trait theory” as a major application of IRT is to estimate the value of a trait (or domain) such as “disability” or “quality of life,” where the trait itself cannot be directly observed. Two IRT requirements are that the items aggregated to estimate a trait are unidimensional in that they measure a single concept, and that they are not redundant (“locally dependent”) with other items in the group2. Given sufficient information about each item, one can predict the performance of one outcome assessment instrument compared with another. For example, one can quite readily, by selecting the better items from an item bank, create instruments that make more precise outcome assessments than the Health Assessment Questionnaire Disability Index (HAQ-DI)5 or the 6-item Short Form Health Survey (SF-6D) HRQOL6 instruments. In turn, this permits major increases in study statistical power or allows use …