Main Session
Sep
29
PQA 07 - Head and Neck Cancer, Lung Cancer/Thoracic Malignancies, and Nursing and Supportive Care
3624 - Survival Benefits of Radiation Dose Escalation in Locally Advanced NSCLC: A Causal Forest Analysis of the National Cancer Database
Presenter(s)
Igor Shuryak, MD, PhD - Columbia University, New Milford, Bergen
I. Shuryak1, D. DeStephano2, and S. K. Cheng3,4; 1Center for Radiological Research, Department of Radiation Oncology, Columbia University Irving Medical Center, New York, NY, 2Columbia University, New York, NY, 3Department of Radiation Oncology, Columbia University Irving Medical Center, New York, NY, 4Herbert Irving Comprehensive Cancer Center, Columbia University Irving Medical Center, New York, NY
Purpose/Objective(s):
The optimal radiation dose for locally advanced non-small cell lung cancer (LA-NSCLC) remains debated. RTOG 0617 showed no benefit from 74 vs 60 Gy, yet observational data suggest moderate escalation (<74 Gy) may benefit select patients. We applied Cox regression and causal survival forests (CSF) to estimate the effects of dose escalation on survival using a large national cohort.Materials/Methods:
We identified 30,678 stage IIIA/IIIB NSCLC patients treated with definitive chemoradiotherapy from the National Cancer Database (2004-2022). Patients were classified into standard dose (ST; 59.4-62.9 Gy; n=17,907), moderate escalation (ME; 63-66 Gy; n=8,812), or higher escalation (HE; 66.1-73.9 Gy; n=3,959). Cox proportional hazards models assessed the association between dose group and overall survival, controlling for 18 confounders including demographics, comorbidities, tumor characteristics, treatment modality, and year of diagnosis. CSF, a state-of-the-art causal machine learning method, estimated heterogeneous treatment effects on restricted mean survival time (RMST) at 3-, 5-, 10-, and 15-year horizons. Multiple imputation (m=5) with Rubin’s rules addressed missing data.Results:
On multivariate Cox analysis, both ME (HR 0.94, p<0.01) and HE (HR 0.92, p<0.01) showed improved survival vs ST. CSF analysis demonstrated RMST gains for ME vs ST of 0.80 months (95% CI: 0.42-1.18; p<0.001) at 3 years, 1.72 (0.80-2.63; p<0.001) at 5 years, 3.55 (0.28-6.82; p=0.034) at 10 years, and 3.22 (-2.20 to 8.65; p=0.24) at 15 years. HE vs ST yielded larger benefits: 1.16 (0.67-1.65; p<0.001) at 3 years, 2.69 (1.50-3.88; p<0.001) at 5 years, 6.03 (2.61-9.45; p=0.001) at 10 years, and 8.06 (4.71-11.41; p<0.001) at 15 years. Tumor size and age were the primary treatment effect modifiers, with smaller tumors associated with greater benefit, and the age effect being more complex.Conclusion:
Both Cox regression (classical statistical modeling) and CSF (causal machine learning) analyses demonstrate dose-dependent survival benefits from radiation dose escalation beyond 63 Gy up to 74 Gy in LA-NSCLC, with treatment effect heterogeneity driven mainly by tumor size and age. If validated prospectively, these findings could support individualized dose-escalation strategies in the context of modern techniques and immunotherapy consolidation.