Nonlinear Mixed Effect Modeling of Patient-Reported Outcomes to Improve Treatment Effect Estimation in an Oncology Clinical Trial
Listed in
This article is not in any list yet, why not save it to one of your lists.Abstract
Background
Patient-reported outcomes (PROs) capture how treatments affect patients’ symptoms, functioning, and quality of life, but substantial variability can obscure treatment effects. We evaluated whether NLME modeling could improve treatment-effect estimation for PRO endpoints by accounting for data variability.
Methods
We retrospectively analyzed participant-level data from the phase III coBRIM trial ( NCT01689519 ), comparing cobimetinib plus vemurafenib with placebo plus vemurafenib in melanoma. EORTC QLQ-C30 global health status/quality of life (GHS/QoL) scores were evaluated using observed-score comparisons, mixed models for repeated measures (MMRM), and NLME modeling. The NLME framework, informed by pharmacology knowledge and context, characterized longitudinal GHS/QoL trajectories while quantifying between- and within-patient variability. Model-estimated scores and trajectory parameters were compared between treatments.
Results
A total of 202 participants with 1,443 GHS/QoL assessments were analyzed. Observed GHS/QoL scores were highly variable, and neither observed-score comparisons nor MMRM identified statistically significant treatment differences at Cycle 8 Day 1. NLME modeling identified significantly less deterioration in longitudinal GHS/QoL with cobimetinib plus vemurafenib (P = 0.036). Model-estimated GHS/QoL scores were also significantly higher with combination treatment at Cycle 8 Day 1 (P = 0.049). Model simulations estimated a 10.4-point between-group difference (95% interval, −6.2 to 26.9), exceeding a commonly used threshold for clinical importance, although uncertainty remained.
Conclusions
NLME modeling can improve PRO data analysis by accounting for substantial between- and within-patient variability and incorporating pharmacological principles to characterize disease- and treatment-related PRO changes over time. This approach may improve treatment-effect estimation and reveal clinically meaningful differences overlooked by conventional analyses.
