Predication of clinical outcomes of advanced cutaneous squamous cell carcinoma to PD1 inhibition directly from histopathology slides using inferred transcriptomics.

Abstract

2630 Background: Metastatic or locally advanced cutaneous squamous cell carcinoma (CSCC) not amenable to local therapy is treated with programmed death (PD)-1 inhibitors, namely Cemiplimab . While response rates are relatively high at approximately 45%, no predictive biomarkers for PD-1 inhibition have been validated in advanced CSCC subjecting some patients, especially older, to unnecessary immune-related adverse events (irAE). We present a retrospective analysis of ENLIGHT-DP, a novel biomarker for response to PD-1 inhibition in advanced CSCC, calculated directly from histopathological slides. Methods: We retrospectively examined high resolution hematoxylin and eosin (H&E) slide scans from archived tumor-tissue samples of advanced CSCC patients treated with Cemiplimab to generate an individual prediction score to PD-1 inhibitors using ENLIGHT-DP. This is composed of two main steps: (I) prediction of individual mRNA expression directly from H&E slides using the digital pathology-based DeepPT algorithm (II) use these values as input to ENLIGHT, a transcriptomics-based precision oncology platform for prediction of response to cancer therapies. We then unblinded clinical outcomes and assessed the predictive values of ENLIGHT-DP. Results: We evaluated 39 cases of advanced CSCC (tumors from various origins) at a median age of 81 years old (range 57-100). Of them, 32 cases (82%) were initially treated with surgery or radiotherapy with curative intent, but ultimately suffered disease progression. The objective response rate (ORR) was 69%, median progression free survival (PFS) was 11 months (CI 95% 10.4-17.6) and 6 patients (15%) suffered from severe irAE necessitating treatment cessation. ENLIGHT-DP was predictive of response with ROC AUC = 0.67. Using a binary threshold for classification, calibrated on previous lung and head and neck cohorts, ENLIGHT-DP displays promising biomarker characteristics: 81.8% PPV, 66.6% sensitivity and 4.0 OR (p = 0.04) for matched vs. unmatched patients. ENLIGHT-DP successfully stratified PFS with HR 0.02 (CI 95% 0.0005-0.96, p = 0.05). Importantly, omitting cases in which ENLIGHT-DP was calculated on samples which were taken more than 6 months prior to Cemiplimab therapy (i.e., during diagnostic excisions) did not influence the results. No other patient characteristics (e.g., age, stage, co-morbidities, previous treatments) were associated with outcomes. Conclusions: ENLIGHT-DP demonstrates high predictive values for clinical outcomes of PD-1 inhibition in advanced CSCC, relying solely on easily accessible archived H&E slides.