Huaxiu Yao

Supplementary Figure S4. Associations between symptom progression parameters and progression-free survival. (A-C) Progression-free survival stratified by maximum improved PRO effects (PMAX), PRO progression rate (SLP), or baseline LCSS score (L0). The “High” level indicates parameters greater than the median, while the “Low” level indicates parameters less than or equal to the median. The shaded area represents the 95% CI, and p-values are calculated using the log-rank test. (D) Cox proportional hazards model incorporating model parameters and LASSO-selected survival-related clinical characteristics. The black boxes with horizontal error bars represent hazard ratio estimates with 95% CI. P-values for each covariate are labeled on the right. LCSS, Lung Cancer Symptom Scale; LASSO, least absolute shrinkage and selection operator; CI, confidence interval; NEUT, neutrophils; AST, aspartate aminotransferase; SMOKE, smoking status.

Supplementary Data S1. Supplementary Data S1. Population model and machine learning codes.

Supplementary Figure S7. Impact of disease stage on PRO progression. The effect of disease stage on PRO progression was assessed using the SLP parameter. No statistically significant differences were observed across stages (ANOVA, p = 0.39) in the NSCLC Phase III clinical trial (NCT00532155). Subgroup sample sizes were as follows: Stage I, N = 24; Stage II, N = 10; Stage III, N = 111; and Stage IV or unknown stage, N = 223.

Supplementary Table S5. PRO model parameter estimates on external validation cohort. RSE, relative standard error; SHR, shrinkage.

Supplementary Table S2. Final model parameter estimates. RSE, relative standard error; CI, confidence interval; SHR, shrinkage.

Supplementary Figure S2. Normality of η distributions checks. The blue bars represent the distribution of model estimated random effects ηs for three parameters L0, PMAX, and SLP. The black solid lines represent the theoretical normal distributions. The η distributions aligned well with normal distribution indicating the appropriate selection of random effect models.

Supplementary Figure S8. AUC-ROC curve of machine learning predictive model on external validation cohort. AUC-ROC, Area Under the Receiver Operating Characteristic curve.

Supplementary Figure S1. Normality of IWRES distributions checks. (A) The IWRES density (blue bars) overlay with normal distribution (black dashed line). (B) Probability of IWRES (blue solid line) overlay with theoretical distribution (black dashed line). The IWRES distribution overlay well with theoretical distribution indicating a good normality of residual error model. IWRES, individual weight residuals.