2518 - Acute-Phase Semiquantitative Immune Cell Kinetics as Multi-Endpoint Prognostic Biomarkers for Survival Metrics in LA-NSCLC treated with Chemoradiotherapy

Purpose/Objective(s): To evaluate whether acute-phase peripheral blood immune-cell kinetics improve multi-endpoint prognostic modeling in locally advanced non-small cell lung cancer (LA-NSCLC) treated with concurrent chemoradiotherapy (cCRT), beyond clinical and radiotherapy (RT) planning factors.

Materials/Methods: We analyzed 432 LA-NSCLC patients treated with cCRT (proton, n=153; photon, n=279); carboplatin/paclitaxel was most common (n=302), and 115 received consolidation immunotherapy post-RT. We extracted 16 clinical (C), 14 RT planning (R), 41 semiquantitative kinetic (S), and 7 kinetic model-fit (K) parameters. S and K were computed from serial absolute lymphocyte/neutrophil counts (ALC/ANC) from baseline through 30 days post-RT. Endpoints were progression-free survival (PFS), overall survival (OS), and distant metastasis-free survival (DMFS). The cohort was split into training (n=302) and test (n=130) sets. Within training, stable features were selected using 200× nested cross-validation (CV; 5×3 folds) with a multitask group-lasso Cox model, with stability-threshold optimization and correlation pruning. Random survival forests (RSF) were trained per endpoint with 5-fold CV joint tuning across endpoints; test risk scores were averaged across 20 independently seeded full-training refits. Comparative analyses were performed across single, pairwise, and three-group C/R/S/K input sets using the same feature-selection and RSF pipeline.

Results:

R+S ranked best overall, with C+R as the best non-S set. Across the top three sets, feature selection consistently retained ALC recovery slope, ANC recovery ratio, and ANC end value (defined as the last measurement = 30 days post-RT); the latter ranked 1st in S and 2nd in R+S and C+R+S by RSF importance. Gross tumor volume (GTV) and effective dose to immune cells (EDIC) were consistently selected in R+S and C+R+S, with GTV ranked 1st in both. In the best RSF model (R+S), partial dependence for the top five features identified higher-risk thresholds of GTV > 270 cc, ANC end value > 3.3 K/µL, ANC max/ALC nadir ratio > 40, baseline ANC/ALC ratio > 5.8, and EDIC > 4 Gy. This model significantly outperformed the R-only OS C-index (0.68 vs.0.62, p=0.02) and stratified 5-year OS between test-set RSF-predicted low- vs. high-risk groups (Kaplan-Meier: 37.2% vs. 22.4%; log-rank p=0.004).

Conclusion: Semiquantitative immune-cell kinetics improved OS, PFS and DMFS prediction in LA-NSCLC treated with cCRT. ANC-to-ALC ratio and ALC/ANC recovery features added prognostic value to tumor-burden and dosimetric factors, supporting immune-cell biomarkers for optimizing cCRT in LA-NSCLC.

Table 1. Top 10 input sets by mean test RSF C-index.

Rank	Input set	PFS	OS	DMFS	Mean	Note
1	R+S	0.63	0.68	0.66	0.65	Best overall
2	C+R+S	0.60	0.65	0.65	0.63
3	S	0.61	0.65	0.62	0.63	Best single-group
4	C+R	0.60	0.64	0.61	0.62	Best non-S set
5	S+K	0.59	0.63	0.63	0.62
6	C+R+K	0.60	0.64	0.61	0.62
7	R+S+K	0.59	0.64	0.62	0.62
8	C+S	0.59	0.62	0.64	0.62
9	R	0.60	0.62	0.62	0.61
10	C	0.58	0.61	0.59	0.59