Novel growth pattern‐specific digital marker of TILs improves stratification of lung adenocarcinoma patients

TL;DR

A new AI-based marker called GPS-TILs improves the prediction of lung adenocarcinoma patient outcomes by combining tumor growth patterns and immune cell data.

Contribution

The novel GPS-TILs marker integrates tumor growth patterns and TILs data to provide more accurate patient stratification in lung adenocarcinoma.

Findings

GPS-TILs showed strong prognostic value for overall survival with a C-index of 0.59.

GPS-TILs outperformed conventional TIL-based measures and morphology-based approaches in patient stratification.

Abstract

Lung adenocarcinoma (LUAD) is one of the most prevalent forms of cancer and continues to be associated with high mortality rates, despite recent advances in cancer therapy. Effective risk stratification is critical for guiding treatment decisions and improving our understanding of disease mechanisms. However, current prognostic approaches face considerable limitations. Growth pattern‐based grading serves as a prognostic indicator of tumour aggressiveness, but is inherently subjective and prone to a high degree of variability among observers. Other well‐established prognostic indicators, such as tumour infiltrating lymphocytes (TILs) and stromal TILs (sTILs) scores, provide valuable prognostic information but require labour‐intensive assessment. The pronounced heterogeneity of LUAD further complicates prognosis and underscores the need for robust, integrative biomarkers that capture both the morphological and immunological characteristics of the tumour. To address this need, we propose an AI‐based growth‐pattern‐specific TILs (GPS‐TILs) marker that quantifies TILs and sTILs within each growth pattern separately. By integrating morphological information from the tumour growth patterns and immune microenvironment data from TILs, we demonstrate that the proposed GPS‐TILs marker improves patient stratification. We evaluated the prognostic utility of GPS‐TILs using survival analysis with Cox proportional hazards models in a cross‐validation setting using The Cancer Genome Atlas LUAD (TCGA‐LUAD) cohort. Our findings revealed that GPS‐TILs offers strong prognostic value for overall survival (p < 0.0001, C‐index = 0.59), outperforming conventional TIL‐based measures and morphology‐based stratification approaches. These results highlight the potential of GPS‐TILs as a more objective and effective tool for improving patient risk stratification in LUAD. © 2025 The Author(s). The Journal of Pathology published by John Wiley & Sons Ltd on behalf of The Pathological Society of Great Britain and Ireland.

Figures