• Institute of Pathology, University Hospital Cologne, Cologne, Germany; Medical Faculty, University of Cologne, Cologne, Germany.
• Institute of Pathology, University Hospital Cologne, Cologne, Germany.
• Clinic I of Internal Medicine, Center for Integrated Oncology, Germany.
• Department of Cardiothoracic Surgery, University Hospital Cologne, Germany.
• Institute of Pathology, Charité University Hospital, Berlin, Germany.
• Institute of Pathology, University Hospital Cologne, Cologne, Germany. Electronic address: yuri.tolkach@gmail.com.

BACKGROUND AND AIM: Lung cancer is the leading cause of cancer-related mortality worldwide, highlighting the importance of refining diagnostic modalities. This study's main focus is the development of a digital pathology, prognostic algorithm for fully automatized quantification of stroma-tumor ratio (STR) in patients with resectable non-small cell lung cancer (NSCLC).

MATERIALS AND METHODS: The developed STR algorithm is built upon a powerful multi-class tissue segmentation algorithm that generates precise maps of the full tumor region. One retrospective exploration cohort of NSCLC patients (n = 902) and three validation cohorts (n = 784) of patients with lung adenocarcinoma (LUAD) and squamous cell carcinoma (LUSC) were included to identify and validate optimal prognostic cut-offs and different risk stratification methods with regard to different clinical endpoints: overall survival (OS), cancer-specific survival (CSS) and progression-free survival (PFS).

RESULTS: For LUAD, we show that the minimal STR value for the whole case is decisive for prognostic evaluation. Different approaches (single STR cut-off, multiple STR cut-offs, using STR as a continuous parameter) allow for robust stratification of patients into prognostic risk groups, independent of the classical clinicopathological variables and conventional histological grading. For LUSC, STR may assist in identifying a small subset of patients with unfavorable prognosis (based on the maximum STR for the whole case), however, its prognostic impact varies between cohorts.

CONCLUSION: STR quantification in LUAD NSCLC subtype shows a promising role as a prognostic biomarker. It can be easily implemented in routine diagnostics and could be considered as a component of advanced prognostic systems in LUAD. Our results in LUSC cohorts suggest that STR quantification in its current implementation is of limited value in this subtype.