Improving the Interpretability of Ordinal Endpoints in NMAs
EVERSANA, West Porters Lake, NS, Canada
OBJECTIVES: Ordinal endpoints are common in NMAs for health technology assessment submissions. PASI is a continuous endpoint used in psoriasis trials that is typically reported as number of patients reaching 50/75/90/100% improvement. In NMAs, PASI is usually analyzed as an ordinal outcome leveraging the parallel lines or proportional odds assumption. While clinicians are familiar with % improvement thresholds, they may be more interested in summaries that not typically reported in trials: The full distribution of PASI % change, mean PASI % change, or median PASI % change. We provide proof-of-concept for a unified framework that allows estimation of clinically relevant estimates using any PASI NMA.
METHODS: We use simulated data to create a realistic distribution of PASI % change that is modeled with a flexible semi-parametric proportional odds model. We combine this baseline model with odds ratios output from a proportional odds NMA of PASI 50/75/90/100 to allow for the generation of the full PASI % change distribution in the population described by the individual participant data. Potential deviation from the proportional odds assumption is explored with linear and spline terms, which can be incorporated in subsequent NMA.
RESULTS: The proposed method can be used to calculate differences in any quantity of interest with uncertainty, including visualizations of the difference in entire distributions of PASI % change. The extension to partial proportional odds models creates the possibility to explore how different therapies may have greater effects on some areas of the distribution than others.
CONCLUSIONS: Proportional odds models can allow for the use of any ordinal NMA output to generate clinically relevant contextualization of results for summaries not commonly reported in trials.
Conference/Value in Health Info
Value in Health, Volume 26, Issue 6, S2 (June 2023)
Economic Evaluation, Methodological & Statistical Research, Organizational Practices, Study Approaches
Artificial Intelligence, Machine Learning, Predictive Analytics, Best Research Practices, Meta-Analysis & Indirect Comparisons, Value of Information
No Additional Disease & Conditions/Specialized Treatment Areas