339 Background: Immunotherapy, including anti-PD-1 inhibitor, represents a promising breakthrough treatment in poor prognostic cancers such as biliary tract cancer (BTC). However, subset of pts suffer from unexpected acceleration of tumor growth following the initiation of immunotherapy, termed hyperprogressive disease (HPD). We assessed HPD in BTC pts treated with anti-PD-1 inhibitor in association with clinicopathologic features. Methods: This retrospective study included pts with BTC who were treated with pembrolizumab 200mg IV as palliative second or third line treatment in 4 tertiary hospitals in South Korea. Previous proposed tumor dynamic parameters including time to treatment failure (TTF), tumor growth kinetics (TGK) and tumor growth rate (TGR) were calculated per RECIST v1.1. Neural network modeling technique was adopted to find clinicopathologic features that can predict HPD. Results: Total of 223 pts with BTC were treated with palliative second or third line pembrolizumab between December 2015 to August 2020. ORR was 11.2% (n=1 for CR, n=24 for PR, n=70 for SD, n=118 for PD, and 10 pts not evaluable). Among patients with best response as PD (n=118), 41 patients (18.4% from total pts) met the criteria for HPD definition (>2-fold increase in both TGK and TGR in the experimental period compared to the reference period). HPD pts had worse prognosis compared to non-HPD pts in terms of TTF (median 1.4 vs 2.7 months, P =<0.0001) and OS (median 3.57 vs 5.27 months, P =0.03). Clinicopathologic features including age, ECOG performance status, primary tumor site, pathology, metastatic organ, previous treatment history (surgery or radiotherapy), pretreatment PD-L1 score, tumor markers, and baseline laboratory results (LDH, albumin, or neutrophil-to-lymphocyte ratio) did not predict HPD (compared to non-HPD pts or PD without HPD pts). By neural network modeling and plotting with hidden vectors, we could define HPD-high-risk cluster (n=37) and HPD-lower-risk cluster (n=80). Higher CA 19-9-to-lymphocyte ratio ( P =0.0004) or higher platelet-to-PD-L1 CPS (combined positivity score) ratio ( P =0.002) predicted HPD-high-risk-cluster. HPD pts (n=41) tend to have progression due to liver (73.2% vs 45.6%, P =0.0026) or peritoneum (53.7% vs 21.4%, P=<0.0001) metastases, compared to non-HPD pts (n=182). Conclusions: This study for the first time describe the clinicopathologic features of HPD among pembrolizumab treated BTC pts. Baseline pathology and laboratory tests may help optimal patient selection for immunotherapy to avoid HPD in BTC pts. Further validation of BTC-specific HPD predictive features and definition (TGK or TGR ratio cut-off) will be presented after comparison with clinicopathologic features among second line 5FU-based chemotherapy treated BTC pts.
