ASCO 2025 | AI model predicts claudin 18.2 and immune phenotype in gastric cancer

This study actually aimed to develop an AI-based model to predict Claudin-18.2 expression in gastric cancer. So with the introduction of various potent first-line regimens like ICI plus chemotherapy, recently two phase three trials have demonstrated a survival benefit of addition of zolbetuximab to chemotherapy in Claudin-18.2 positive tumors. So the treatment landscapes have become increasingly complex. So in this regard, there needs to be some guidance as to which regimen to go. So this AI-based approach employed two AI models. The first one is to develop an AI model to create Claudin expression so that the patient could screen for the zolbetuximab-based chemotherapy. And the second model, which was presented in part in ESMO last year, this AI model predicts immune phenotypes. So, inflamed immune phenotypes is associated with the benefit of adding ICI to chemotherapy in the first-line setting. So, in this study, we developed a model which showed a quite promising predictive value in terms of predicting actual Claudin expression. So the AUC was around 0.8, which is quite acceptable, and sensitivity for detecting Claudin-18.2 expression was above 80%. So, and combining these two AI models, we were able to find a subgroup of patients who had non-inflamed phenotype and who were predicted to be positive for Claudin-18.2 expression. So in the validation of the clinical outcomes, we found that for these patients, ICI plus chemotherapy was not effective for prolonging progression-free survival, and overall survival. So meaning that these patients may not be candidates for ICI plus chemotherapy, requiring alternative treatment options like zolbetuximab plus chemotherapy. So that’s the main message of this study. And regarding the accuracy of the biomarkers, currently available ones, about one-third of patients who are classified to be immune non-inflamed and Claudin-18.2 positive actually had high PD-L1 expressions, meaning that currently available biomarker, which hugely rely on PD-L1 expression, may not be accurate. So this study can provide some aspects to compensate the currently available biomarker. Yeah, so we need to validate whether this AI model could predict patients who are likely to see, who are likely to have favorable outcomes with zolbetuximab plus chemotherapy. So after that validation, there could be more clear guidance on the deciding treatment regimens.

This transcript is AI-generated. While we strive for accuracy, please verify this copy with the video.