Upper GI
Daniel Skubleny, MD PhD
Postgraduate trainee
University of Alberta
Edmonton, Alberta, Canada
Daniel Skubleny, MD PhD
Postgraduate trainee
University of Alberta
Edmonton, Alberta, Canada
Dan Schiller, MD, MSc
Associate Professor
University of Alberta, United States
Gina Rayat, PhD
Professor
University of Alberta, United States
Molecular classification in gastric cancer has identified relevant disease heterogeneity with prognostic implications. However, limited comparative analysis of molecular classification systems has occurred. We assessed the effect of the Tumour Microenvironment Score (TME), the Cancer Genome Atlas (TCGA) and Asian Cancer Research Group (ACRG) classification systems on disease-free survival (DFS) in non-metastatic gastric cancer patients.
Methods:
Previously characterized machine learning models were created for TCGA, ACRG and TME classes using 10-fold nested cross validation. Each patient in the ACRG and TCGA datasets were assigned molecular classes using these models. DFS in stage I-III patients was assessed using univariable and multivariable Cox Proportional Hazards models. Sensitivity analyses were performed to assess the strength of effects to confounders.
Results:
Mean model classification accuracy was 89.5% (TCGA), 84.7% (ACRG) and 89.3% (TME). Similar proportions of TCGA, ACRG and TME molecular classes were present between TCGA and ACRG cohorts (Chi-squared test, p > 0.05). The proportion of TME high tumours was greatest for TCGA microsatellite instability high (MSI-H) (70%), ACRG MSI-H (67%) and TCGA Epstein-Barr virus type (64%). In mesenchymal type tumours such as ACRG epithelial-mesenchymal-transition and TCGA genomically stable, TME high was present in 3.1% and 0% of tumours, respectively. Using available data for 454 gastric cancer patients and 135 DFS events, only TCGA MSI (Hazard Ratio (HR) 0.54 [95% CI 0.30, 0.97]; p < 0.05) and TME High (HR 0.29 [95% CI 0.16, 0.52]; p < 0.001) were statistically significant factors in univariable models. In a multivariable Cox model including TCGA, ACRG and TME subtypes, only a high TME score was associated with improved DFS (HR 0.22 [95% CI 0.11, 0.45]; p < 0.001). The significant effect of TME high score was maintained after sensitivity analysis that adjusted for stage, age, sex, chemotherapy, radiation, tumour location, and study (TME High HR 0.36 [95% CI 0.18, 0.74)]; p < 0.01).
Conclusions:
In an integrated analysis comparing TCGA, ACRG and TME scores, a high TME score is the only independent molecular prognostic factor for DFS in non-metastatic gastric cancer. Additional investigation into implications of the heterogeneity of the TME score relative to the TCGA and ACRG classifications may yield additional insight into gastric cancer biology and treatment.