A Rasch Measurement Theory Approach to Improve the Interpretation of Patient-reported Outcomes

Abstract

Introduction: Interpretability is a key challenge for researchers, clinicians, and patients interested in using the output of patient-reported outcome (PRO) instruments. When rich and detailed meaning is required to assist individual patients to make treatment choices we propose that the item content of psychometric rating scales should be better leveraged to improve interpretability.

Methods: Modern psychometric methods such as Rasch measurement theory allow PRO users to map patient progress up or down a scale over time to item benchmarks. These benchmarks represent the relative difficulty associated with each item contained in a scale. The most difficult items benchmark the best overall outcome on a scale and the least difficult items represent the worst overall outcome. The group-level effects of different treatment options can also be mapped to item benchmarks and compared with those of individual patients.

Discussion: The proposed method leverages the content already available in PRO instruments to improve interpretation. This approach locates both individuals (in this instance breast cancer patients undergoing surgery) and treatments (in this instance breast reconstruction techniques), on a hierarchy of health variables where each variable represents a progressively more advanced step in the recovery process. The approach can specify the aspects of a health concept where patients are currently competent, and the aspects which they might gain from a new treatment. It can also assist attempts by industry to communicate specific treatment benefits to their target audience.

Key points: The method is best applied when patients need guidance about the likely benefits of different treatment options, when a PRO instrument has been developed using a modern psychometric method such as Rasch measurement theory, when there is good evidence from well-conducted studies of the group-level benefits of different treatment options, and these benefits have been measured using appropriate PROs. The method depends on good "fit" of individual patient responses to an underlying model. However, even when fit is poor it may be useful for patients to understand the "ladder" of health achievements for their condition (from the perspective of the average patient) and where different treatment options sit on this ladder.