Presentation (CTI)

OBJECTIVES: Patient-reported outcome (PRO) scores are commonly analysed as if they measure quantities, despite being based on ordered response categories. Consequently, PRO estimators may rely on assumptions not justified by ordinal data alone. This study identified quantitative PRO estimators used in oncology, examined assumptions needed for interpretation, and proposed interpretable alternatives compatible with ordinal information.
METHODS: Estimators were identified through a review of PubMed-indexed oncology clinical trials published from 2021-2026. For each estimator, implied outcome variables and scale properties were identified, and underlying clinical questions were reformulated using ordinal outcomes. Replacement estimators were selected to preserve intended treatment comparison while avoiding interpretation of numerical score differences as quantities.
RESULTS: Across 236 estimators from 74 clinical trials, the most frequent were mean change from baseline (25.0%), between-group least-squares mean difference (18.6%), time to fixed-point worsening (16.1%), mean score difference (12.1%), and fixed-point change (10.7%). Only 11.1% of estimators were suitable for ordinal measures.
Mean change and least-squares mean differences treat score differences as quantities, requiring interpretable score intervals. Fixed-point change and time-to-worsening estimators assume that a numerical change has consistent meaning across severity levels and the score range. These estimators can be replaced without changing the underlying clinical question. Mean change can be replaced by probabilities of improvement, worsening, or no change in ordered categories. Between-group mean differences can be replaced by marginal cumulative probabilities, cumulative odds estimators, or the probability that an outcome in a treatment group is better than one in a comparator. Fixed-point response can be replaced by category-based improvement, no worsening, or attainment of a defined state. Time to fixed-point worsening can be replaced by time to first worsening in ordered severity category.
CONCLUSIONS: Common oncology PRO estimators often require assumptions unsupported by ordinal data alone. Ordinally compatible estimators can address the same questions without treating arbitrary score distances as quantities.