Leveraging Longitudinal Patient-Reported Outcomes Trajectories to Predict Survival in Non-Small-Cell Lung Cancer

Purpose: Despite their potential, patient-reported outcomes (PRO) are often underutilized in clinical decision-making, especially when improvements in PRO do not align with clinical outcomes. This misalignment may result from insufficient analytical methods that overlook the temporal dynamics and substantial variability of PRO data. To address these gaps, we developed a novel approach to investigate the prognostic value of longitudinal PRO dynamics in non-small-cell lung cancer (NSCLC) using Lung Cancer Symptom Scale (LCSS) data. Methods: Longitudinal patient-reported LCSS data from 481 NSCLC participants in the placebo arm of a Phase III trial were analyzed. A population modeling approach was applied to describe PRO progression trajectories while accounting for substantial variability in the data. Associations between PRO model parameters and survival outcomes were assessed using Cox proportional hazards models. Model-informed PRO parameters were used to predict survival via machine learning. Results: A PRO progression model described LCSS dynamics and predicted a median time to symptom progression of 229 days (95% confidence interval [CI]: 15-583). Faster PRO progression rates were significantly associated with poorer survival (Hazard ratio [HR] 1.13, 95% CI: 1.076-1.18), while greater improved PRO effects by placebo/prior treatment correlated with improved survival (HR 0.93, 95% CI: 0.883-0.99). A machine learning model using PRO parameters achieved an AUC-ROC of 0.78, demonstrating their potential to predict overall survival. Conclusions: This study demonstrates that longitudinal PRO data can provide prognostic insights into survival in NSCLC. The findings support the use of PRO dynamics to improve clinical decision-making and optimize patient-centered treatment strategies.