Presentation (CTI)

OBJECTIVES: The U.S. Food and Drug Administration (FDA) recommends using meaningful score differences (MSDs) to interpret treatment effects. The anchor-based methods (ABM), estimating MSDs by linking changes in patient-reported outcome (PRO) scores to anchor items, has limitations, such as not accounting for measurement error. A new approach using longitudinal item response theory (LIRT) aims to address these issues. This study compares the MSD estimates from both the anchor-based and LIRT methods.
METHODS: Data (N = 153) for the EORTC QLQ-C30 and four anchor items were generated to match the mean vector and correlation matrix from a Phase 3 advanced solid cancer trial. MSDs for both improvement and decline of 1 and 2 points on anchor items were estimated for the physical functioning (PF) subscale using four anchor items. MSDs were estimated using both the mean and median with the commonly used anchor-based method and the LIRT method.
RESULTS: The LIRT graded response models converged without issue, which is notable given the relatively small sample size in this study. MSD estimates given by the LIRT method were generally larger in absolute magnitude than the ABM estimates for both improvement and decline in PF scores. However, most MSD estimates for improvement between the two estimation methods were within one standard error of measurement. Additionally, the 95% confidence intervals for all but two of the MSD estimates overlapped between the anchor-based and LIRT methods. The LIRT and ABM estimates were, however, notably further apart for decline than for improvement.
CONCLUSIONS: The study indicates that LIRT estimates are typically larger than ABM estimates, aligning with prior research. However, LIRT estimates are not overly conservative, as shown by overlapping confidence intervals and closeness to ABM values within one standard error. This supports the use of LIRT for estimating measurement sensitivities in clinical trial data, even with small sample sizes.