Artificial intelligence-powered assessment of tumor microenvironment in pre-treatment and on-treatment biopsies informs treatment outcomes to pembrolizumab in patients with rare tumors
Mohamed H. Derbala, Kathryn E. McGonagle, Harsh Batra,3 Bettzy Stephen, Serdar A. Gurses, Seungeun Lee, Woochan Hwang, Soohyun Hwang, Chan-Young Ock, Mohamed Gouda, Lilibeth Castillo, Abdulrazzak Zarifa, Enedelia Rodriguez, David S. Hong, Sarina Anne Piha-Paul, Jordi Rodon Ahnert, Daniel David Karp, Funda Meric-Bernstam, Maria Gabriela Raso, Aung Naing
SITC, 2024
Background Pathologic tumor response and changes in the tumor microenvironment (TME) following immunotherapy serve as surrogate markers to predict clinical outcomes in patients treated with immune checkpoint inhibitors.1-3 However, these associations have rarely been systematically studied in rare tumors. This study investigated whether analysis of pre-treatment and on-treatment biopsies can predict response to pembrolizumab in patients with rare tumors.
Methods We evaluated 259 baseline and 248 on-treatment (cycle 1, days 15-21) biopsies from 84 patients with rare tumors (10 histologic subtype groups) treated with pembrolizumab. Tumor infiltrating lymphocyte density (iTIL; defined as count per tumor area) and tumor content (TC; defined as the ratio of cancer area to the total of cancer area, stroma, and background tissue) were assessed on H&E-stained slides using an artificial intelligence (AI)-powered whole-slide image analyzer (Lunit SCOPE IO) developed on 18,935 whole-slide images including rare tumor types. Baseline iTIL, and changes in iTIL and TC from baseline to on-treatment, were correlated with histologies and treatment outcomes per immune-related Response Evaluation Criteria in Solid Tumors (irRECIST). iTIL levels were categorized as high or low using a cutoff value of 60/mm².
Results Table 1 outlines patients’ demographics. In baseline biopsies, iTIL varied across tumor histologies, ranging from 5.8–57.46/mm². A higher baseline iTIL was associated with favorable progression-free survival (PFS), (HR, 0.49 [95% CI, 0.25–0.99]; P=.042) in only histologic subtypes with higher iTIL (figure 1), though this association was trending towards significance in the overall cohort (HR, 0.62 [95% CI, 0.37–1.10]; P=.077). In patients with matched baseline and on-treatment biopsies (n = 83), there was a trend suggesting that both an increase in iTIL and a decrease in TC were associated with prolonged PFS and overall survival (OS) (iTIL: HR 0.65 [95% CI, 0.40–1.10]; P=.081 for PFS and HR, 0.59 [95% CI, 0.35–1.00]; P=.052 for OS; TC: HR, 0.51 [95% CI, 0.29–0.90]; P=.015 for PFS and HR, 0.54 [95% CI, 0.29–0.98]; P=.039 for OS) (figure 2). The combination of changes in iTIL and TC was significantly associated with both PFS and OS (PFS: HR, 0.32 [95% CI, 0.14–0.71]; P=.003; and OS: HR, 0.28 [95% CI, 0.10–0.78]; P=.009) (figure 3).
Conclusions AI-powered assessment of TME using pre-treatment and on-treatment biopsies may help predict treatment response to pembrolizumab in patients with rare tumors, particularly in histologies with higher iTIL levels. This knowledge may facilitate the development of personalized treatment strategies and provide guidance for future immunotherapy trials in these patients.