Presentation (CTI)

OBJECTIVES: Change from baseline (CfB) in PROs is a critical endpoint in oncology clinical trials. Although mixed-model for repeated measures (MMRM) is a recommended approach for evaluating the magnitude of change, it excludes patients lacking a valid baseline measurement. As an alternative, we aim to assess the performance of the cLDA model, which allows the inclusion of patients with missing baseline in treatment effect estimation.
METHODS: PRO data from a phase 3 trial in acute myeloid leukaemia patients (NCT02993523, N=431) were used to estimate the treatment effect on longitudinal magnitude of change in EORTC QLQ-C30 physical functioning (PF), comparing estimates (Least Squares mean [LSMean]) and precision (width of confidence interval [CI]) using CfB MMRM vs cLDA models. We simulated datasets with increasing levels of missingness (LoM) at baseline (10%, 20% and 30%) and three scenarios of missingness (SoM): 1) missing at random, 2) missing more likely in the control arm, and 3) missing more likely for lower baseline PF scores.
RESULTS: With no baseline missingness, results from the cLDA and CfB MMRM models were equivalent. Across all LoM and SoM, the estimated treatment effects (LSmean) were similar. However, the cLDA model resulted in narrower CIs. The improvements in the CI width for the cLDA model was greater as the LoM increased, and in scenarios with higher missingness in the control arm (SoM2: 95% CI width 11.9 [cLDA] vs 13.3 [MMRM] with 20% missing, 12.0 vs. 14.4 with 30% missing).
CONCLUSIONS: At higher magnitudes of baseline LoM and across SoMs, the cLDA model improves precision of the estimate of treatment effect compared to using a CfB MMRM. Using cLDA and including all baseline assessments allows for all patients who provided PRO at any time in the study to be included in the analysis, aligning more closely with the intent-to-treat population.