Abstract 231P
Background
Lung cancer is a leading cause of cancer-related deaths. For unresectable stage III NSCLC, standard treatment includes concurrent or sequential CRT with optional adjuvant immunotherapy. Prognostic outcomes vary widely, making accurate prediction of progression-free survival (PFS) and overall survival (OS) essential for personalized treatment. This study aims to develop a prognostic model combining clinical factors with radiomic features from lung and tumor regions to improve prediction accuracy.
Methods
This study analyzed 213 stage III NSCLC patients treated at Maastro Clinic (Netherlands) from 2015 to 2022, divided into training (70%) and validation (30%) sets. Clinical predictors of PFS and OS were identified using univariate and multivariate Cox models, forming a clinical prediction model (Model-C). Performance was assessed using the concordance index (C-index). Kaplan-Meier curves evaluated the impact of adjuvant durvalumab.
Results
For PFS, univariate Cox analysis identified advanced age at initial treatment, PS score (2–3), hypertension, and chronic kidney disease as risk factors, while adjuvant durvalumab and concurrent CRT were protective. Multivariate analysis confirmed PS score (HR=2.00, 95% CI: 1.19–3.37, p=0.01) and adjuvant durvalumab (HR=0.47, 95% CI: 0.25–0.89, p=0.02) as significant. Model-C achieved a C-index of 0.7118 (training) and 0.6950 (validation). For OS, univariate analysis identified similar factors. Multivariate analysis revealed PS score (HR=2.50, 95% CI: 1.36–4.59) and adjuvant durvalumab (HR=0.32, 95% CI: 0.13–0.78) as significant, with a C-index of 0.7286 (training) and 0.7369 (validation). Kaplan-Meier curves showed significant differences in OS and PFS based on adjuvant durvalumab (Log-Rank Test p < 0.05).
Conclusions
Adjuvant durvalumab significantly improves PFS and OS. Concurrent CRT offers better outcomes than sequential CRT. Patients with higher PS scores are at greater risk, emphasizing the need for tailored treatment strategies. Future work will integrate radiomic features selected using SHAP values to develop tumor (Model-TC), lung (Model-LC), and combined radiomics models (Model-LTC) to enhance predictive accuracy.
Legal entity responsible for the study
The authors.
Funding
Has not received any funding.
Disclosure
All authors have declared no conflicts of interest.