167P - Tumor-immune spatial interactions on NSCLC H&E slide images predicts immunotherapy response: Preliminary external validation

Background

Immune Checkpoint Inhibitors (ICIs) are pivotal in the treatment of late stage NSCLC. Currently, PDL1-high (TPS>=50%) patients generally receive ICI alone, while PDL1 low and negative patients typically receive chemoimmunotherapy (chemo-ICI). However, many patients in both groups fail to respond, suggesting opportunities for more effective intervention - particularly PDL1 high patients who may benefit from escalation to chemo-ICI. We developed an AI-based digital pathology immunotherapy response (PIRe) score based upon tumor infiltrating lymphocyte density (denTIL) and their spatial interactions with tumor cells (spaTIL).

Methods

Pre-treatment hematoxylin and eosin (H&E) stained whole slide images (WSIs) of patients with mostly late stage NSCLC (94.3% stage III/IV, 5.7% stage I/II) treated with ICI were collected from 3 institutions (D1-D3). Nuclei of tumor cells and TIL were segmented and classified automatically via the Picture Health Px platform for spaTIL and denTIL feature calculation. A neural network classifier was trained with these features on a training set (D1 + D2, N = 77) for primary adenocarcinoma and squamous cell carcinoma to predict objective response (OR). PIRe, PDL1 TPS, and a combined PIRe+PDL1 model were then tested on an external test set (D3, N=29) of ICI and chemo-ICI recipients.

Results

PIRe included 5 features, such as morphology of TIL clusters and abundance of TILs surrounding tumor cells. Overall test AUC was 0.71 with 0.82 for ICI and 0.66 for chemo-ICI. PIRe was significantly elevated among responders to ICI (p=0.04, n=11), but not to chemo-ICI (p=0.14, n=16), as assessed by one-sided Wilcoxon rank-sum test. When combined, PDL1 and PIRe consistently outperformed PDL1 TPS alone (Table). Table: 167P

AUC for ICI response prediction with different models for different treatment subgroups

Conclusions

PIRe predicted ICI response in NSCLC patients, showing added value to PDL1. Further validation in larger cohorts to confirm these promising early findings is essential.

Clinical trial identification

Editorial acknowledgement

Legal entity responsible for the study

The authors.

Funding

Picture Health.

Disclosure

A. Madabhushi: Financial Interests, Personal, Advisory Board, Serve on SAB and consult.: SimbioSys; Financial Interests, Personal, Advisory Board: Aiforia, Picture Health; Financial Interests, Personal, Full or part-time Employment: Picture Health; Financial Interests, Personal, Ownership Interest: Picture Health, Elucid Bioimaging, Inspirata Inc; Financial Interests, Personal, Royalties: Picture Health, Elucid Bioimaging; Financial Interests, Institutional, Funding: AstraZeneca, Bristol Myers Squibb, Boehringer Ingelheim, Eli Lilly. N. Braman: Financial Interests, Personal, Stocks/Shares: Picture Health, Tempus Labs. All other authors have declared no conflicts of interest.