Artificial intelligence-assisted evaluation of tumor infiltrating CD3+and CD8+T cells for prognostication and prediction of benefit from adjuvant chemotherapy in early stage colorectal cancer (CRC): A retrospective analysis of the QUASAR trial.

204 Background: High CD3 + (all) and CD8 + (cytotoxic) T cell densities in the core (CT) and invasive margin (IM) of primary CRCs have been shown to be associated with superior prognosis at all stages of disease. Their predictive effect on benefit from adjuvant chemotherapy in early stage CRC has not been tested. Methods: FFPE samples from participants (pts) in the QUASAR trial (adjuvant fluorouracil/folinic acid vs observation in stage 2/3 CRC) were analysed for CD3 and CD8 immunohistochemistry (IHC). Pathologists annotated the core and peritumor areas on digital slide images. Artificial intelligence (AI) algorithms delineated the CT and IM, and calculated the densities (cells/mm 2 ) of each marker in each region (CD3-CT, CD3-IM, CD8-CT, CD8-IM). Pts were randomly partitioned into test and validation sets (1:1). In the test set, each measure’s prognostic effect on recurrence-free interval (RFI) (primary endpoint), colorectal cancer mortality (CCM) and overall survival (OS) in each trial arm was assessed. Maximum likelihoods methods were used to develop optimal cut-points. Analyses were repeated in the validation set. Analysis of 425 pts in each set would give > 95% power (α = 0.05, 2-sided) to detect a twofold difference in recurrence risk. In predictive analyses, 2-year recurrence rate was the primary outcome; biomarker-treatment interactions were assessed. Results: Tumor tissue from 868 pts (797 [92%] stage 2; 531 [61%] colon) was analysed, with evaluable results for CD3-CT in 851 (98.0%), CD3-IM in 833 (96.0%), CD8-CT in 849 (97.0%) and CD8-IM in 820 (94.5%) pts. In the test set, optimal cut-points of 318, 798, 81 and 186 cells/mm 2 were defined for CD3-CT, CD3-IM, CD8-CT and CD8-IM respectively. The recurrence rate in the high-risk group was twice that in the low-risk group for all measures (CD3-CT: rate ratio [RR] 2.00, [95%CI 1.33-2.94], p = 0.0008; CD3-IM: 2.38, [1.59-3.57], p < 0.00001; CD8-CT: 2.17, [1.59-3.57], p = 0.0001; CD8-IM: 2.13 [1.43-3.23], p = 0.0001), which was closely replicated in the validation set (CD3-CT: RR 1.96, [1.30-2.94], p = 0.002; CD3-IM: 1.79, [1.18-2.70], p = 0.005; CD8-CT: 1.72, [1.18-2.56], p = 0.005; CD8-IM: 1.72 [1.15-2.56], p = 0.008). In multivariate analyses, prognostic effects were similar in colon and rectal cancers, and in stage 2 and 3 disease. CD3/8 counts were not predictive of benefit from adjuvant chemotherapy, with similar efficacy in the high and low risk groups. Conclusions: AI-assisted CD3 and CD8 counts were strongly associated with tumor recurrence rates. With no biomarker-treatment interactions, proportional reductions in recurrence with chemotherapy were similar in high and low-risk disease. Hence, numbers of high-risk patients needed to treat to prevent one recurrence were about half the number for low-risk patients.