Development, temporal stability, and prognostic discrimination of a dynamic functional-nutritional score and its trajectories in advanced non-small cell lung cancer

Abstract

Background Conventional functional assessment in advanced non-small cell lung cancer (NSCLC) relies largely on baseline performance status and may not fully characterize longitudinal changes in functional and nutritional measures. We constructed a dynamic functional-nutritional score (DFNS) integrating serum albumin, handgrip strength, 6-m gait speed, and reverse-coded Eastern Cooperative Oncology Group performance status (ECOG PS), and examined its temporal stability, longitudinal trajectories, and association with subsequent overall survival (OS). Methods In this single-center retrospective cohort study, 461 patients with stage III–IV NSCLC receiving systemic therapy were included. DFNS coefficients were derived using baseline principal component analysis (PCA) and applied to measurements obtained at baseline and Weeks 6, 12, 18, and 24. Temporal stability was assessed by cosine similarity between baseline and follow-up PCA coefficient vectors. Latent class growth analysis (LCGA) characterized longitudinal DFNS trajectories. A Week 24 landmark design was used to examine post-landmark OS. Cox proportional hazards models, Harrell’s C-index, and 52-week time-dependent area under the curve (AUC) were used to assess prognostic discrimination relative to baseline ECOG PS and individual DFNS components. Results The first principal component explained 37.6%–40.2% of total variance across assessment points, with high similarity between follow-up and baseline coefficient vectors (cosine similarity, 0.952–0.982). LCGA identified three trajectories: High and Stable (Class 1, 60.1%), Decline–Recovery (Class 2, 32.1%), and Persistently Low (Class 3, 7.8%). In the Week 24 landmark analysis, Class 3 was associated with a higher estimated hazard of death than Class 1 (adjusted hazard ratio [aHR], 2.84; 95% CI, 1.69 – 4.78; P<.001). The DFNS trajectory model showed higher observed discrimination than the baseline ECOG PS and 4-component models, with C-indices of 0.742, 0.612, and 0.685, respectively, and 52-week AUCs of 0.768, 0.625, and 0.702, respectively. Conclusions In this retrospective cohort, the DFNS component structure showed temporal stability over 24 weeks, and longitudinal DFNS trajectories were associated with subsequent OS after the Week 24 landmark. The trajectory-based model showed greater observed prognostic discrimination than models based on baseline ECOG PS or individual components. Prospective multicenter studies are needed to evaluate the reproducibility and clinical utility of longitudinal DFNS assessment.