11P - Tumor-infiltrating lymphocytes on routine H&E staining with automated quantification predict outcomes in resectable non-small cell lung cancer

Background

Adjuvant treatment has become more personalized in resectable NSCLC. New biomarkers are needed to predict treatment outcomes. In this study, we looked for a broadly applicable biomarker using deep learning on hematoxylin and eosin (H&E) histology. Based on recommendations [Salgado et al., Ann Oncol 2015], we focused on analyzing tumor infiltrating lymphocytes in the stromal (sTILs) and the intratumoral (iTILs) compartment. We assessed TILs fully automated at slide and tumor bulk level, guided by manual annotations of the tumor bulk. We hypothesized high iTIL and sTIL density being associated with improved disease-free survival (DFS) in the total study population.

Methods

In this retrospective cohort study, 106 chemo-naive patients with resectable NSCLC were identified in two cohorts, based on recurrence (n=45, cohort A) versus no recurrence (n=61, cohort B). Whole mount tumor slides were stained with H&E and digitalized. Deep learning models were used to compute tissue masks [J. Spronck et al., PLMR 2023] and for TIL detection [HoverNet, Med Imag Anal. 2019]. Tumor bulk was manually annotated to exclude distant tumor lets. Distribution of iTIL and sTIL densities in cohort A and B were compared using Mann-Whitney U based on absence of normal distribution patterns. DFS was measured by log-rank and cox regression to correct for covariates after a binary distinction based on median TIL densities.

Results

The slide and annotation iTIL density were significantly higher in cohort B versus cohort A (slide: 300 vs 232 TILs/mm2, p=0.007 and annotation: 318 vs 230 TILs/mm2, p=0.013). The slide and annotation sTIL density were also higher in cohort B versus A (slide: 1489 vs 1153 TILs/mm2, p=0.035 and annotation: 1753 vs 1469, p=0.030). DFS was significantly higher in patients with a high slide iTIL density compared to patients with a low slide iTIL density (HR 0.44; 95% CI 0.24 to 0.81; P<0.009). DFS remained significant after correction for age, gender, smoking history co-morbidity and clinical stage, but not for ECOG performance.

Conclusions

The iTIL density detected on slide level could serve as favorable prognostic factor in patients with resectable NSCLC, potentially being a biomarker for adjuvant treatment.

Legal entity responsible for the study

The authors.

Funding

NWO/VIDI.

Disclosure

All authors have declared no conflicts of interest.