What question did this study set out to answer?

The aim is to assess if AI-derived H&E features predict differential survival benefits for immunotherapy compared to chemotherapy in metastatic NSCLC.

April 5, 2026

Abstract 6725: AI-derived H&E biomarkers predict differential survival under immunotherapy vs chemotherapy in phase-3 metastatic NSCLC study NEPTUNE.

Key Points

The aim is to assess if AI-derived H&E features predict differential survival benefits for immunotherapy compared to chemotherapy in metastatic NSCLC.
Conducted a retrospective analysis of H&E slides from a phase 3 NEPTUNE trial.
Processed slides from 426 patients using AI to extract 15 tumor microenvironment features.
Analyzed overall survival using univariate Cox proportional hazards models with FDR control.
Evaluated treatment × marker interaction terms to assess predictiveness.
Higher stromal fibroblast density was linked to worse overall survival for both treatments.
Higher stromal immune density improved outcomes for immunotherapy with a median OS increase of 5.8 months.
AI-derived features provided better stratification of outcomes than conventional biomarkers like PD-L1 and bTMB.

Abstract

Abstract Background: Immunotherapy outcomes in metastatic NSCLC are variable, and existing biomarkers (PD-L1, TMB) incompletely predict benefit. Routine H CTX n=272) were processed using models trained independently on external, non-NEPTUNE datasets (6, 000 pathologist-annotated H bTMB ≤20 mut/Mb) as described in the NEPTUNE study (de Castro Jr G et al. , J Thorac Oncol, 2023;18: 106-119). Results: Higher stromal fibroblast density was associated with inferior OS in both arms (D+T HR=1. 60 1. 14-2. 25; CTX HR=1. 36 1. 05-1. 76) with no treatment interaction (nominal pᵢnteraction=0. 82), indicating a prognostic effect. Higher stromal immune density was associated with improved benefit for D+T (HR=0. 64 0. 45-0. 90; median OS difference +5. 8 months; FDR p=0. 04) but not CTX (HR=0. 84 0. 65-1. 08, FDR p=0. 38), with a significant treatment interaction (nominal pᵢnteraction=0. 03), supporting a predictive effect. Overall, AI-H Part 1 (Regular Abstracts) ; 2026 Apr 17-22; San Diego, CA. Philadelphia (PA): AACR; Cancer Res 2026;86 (7 Suppl): Abstract nr 6725.

Perguntar à IA

Bookmark