Liquid biopsy-based prediction of clinical benefit from immune checkpoint inhibitors in advanced biliary tract cancer.

e16179 Background: Immune checkpoint inhibitors (ICIs) show promising efficacy as first-line therapy for advanced biliary tract cancer (aBTC), but validated predictive biomarkers are lacking. Here, we identified liquid biopsy-based biomarkers and developed a prediction model. Methods: aBTC patients who received ICIs-based combination therapy as the first-line treatment were prospectively enrolled in this study (NCT06074029). Plasma samples were collected before the first and second immunotherapy infusion to perform cell-free DNA (cfDNA) sequencing of 1021 genes. Circulating tumor DNA (ctDNA) concentration was calculated by multiplying the mean variant allele frequency by the cfDNA concentration and ctDNA fold change was calculated by dividing on-treatment ctDNA concentration by the pre-treatment ctDNA concentration. Tumor mutation burden (TMB) was normalized using the cfDNA concentration. In addition, RNA was extracted from peripheral blood mononuclear cells to profile the leukocyte transcriptome using whole transcriptome sequencing, and the relative proportions of immune cells in leukocytes was deconvoluted using CIBERSORT algorithm. Log-rank and univariate Cox regression were used to evaluate the relationship between features and progression-free survival (PFS). Results: 37 patients were enrolled, with 62.2% having intrahepatic cholangiocarcinoma. The median age was 61 years and 54.1% of patients were females. The median PFS was 4.4 months. Variables related to PFS with the log-rank p value smaller than 0.05 (Table) were entered into a multivariate Cox regression using a stepwise approach for final variable selection. As a result, ctDNA fold change, KRAS mutation, and resting memory CD4 T cell level were used to construct a COX regression model to predict clinical benefit from ICIs. The model predicting 6-month PFS demonstrated an AUC of 0.909 (95% CI, 0.795-1.024). Internal validation, performed through a bootstrap analysis with 100 iterations, yielded an optimism-adjusted AUC estimate of 0.859. Conclusions: Our preliminary results support the excellent performance of this liquid biopsy-based multivariable model to early predict durable benefit from ICIs in aBTC. A large sample size and verifying this model with internal and external data are warranted. Clinical trial information: NCT06074029 . Table: see text