Conscientious use of patient-reported outcome measures in supportive care

Patient-reported outcome measures (PRO) are critical tools to developing an understanding of cancer patients' experience. This paper presents some of the lesser-understood implications of using patient-reported outcome measures in clinical research.

This study uses a combination of literature sources, real-world examples from supportive care studies, and statistical simulations to demonstrate the operating characteristics of patient-reported measures.

It is demonstrated that care must be taken in the analysis of PROs as the assumptions of the most common mean-based approaches are often violated including linearity, normally distributed errors, interference with asymptotic convergence via boundary values, and more. Further, the implications of subjective discretization are shown to reduce the apparent statistical power of PRO-based studies.

PRO-based studies must be designed conscientiously as each PRO item will demonstrate a varying degree of subjectivity in a given population. Sample sizes of randomized studies using PROs must be inflated to account for this. Analyses should consider using ordinal statistical models until such time as the assumptions of mean-based models can be verified.