Artificial intelligence model for predicting the efficacy of sintilimab plus chemotherapy in treating patients with esophageal squamous cell carcinoma

e16030 Background: Blockade of programmed cell death 1 (PD-1) could induce tumor regression in patients with advanced esophageal squamous cell carcinoma (ESCC), although little is known about the biomarker to guide immunochemotherapy. In this study, we sought to evaluate the efficacy and discover biomarkers and establish an artificial intelligence (AI) model to predict the efficacy of the combination of sintilimab and chemotherapy in the treatment of patients with ESCC. Methods: Fifty-nine (59) patients enrolled in our study received chemotherapy with sintilimab and follow-up data were available. Treatment outcomes were assessed using the Response Evaluation Criteria in Solid Tumors (v1.1). Eleven (11) clinicopathologic characteristics including age, albumin, HLA-1 evolutionary divergence (HED), TMB, body mass index (BMI), neutrophil-lymphocyte ratio (NLR), platelets, hemoglobin, stage, sex, leukocyte, lymphocyte and therapy were collected. Results: Thirty-four (34) patients with ESCC who received sintilimab as first-line chemotherapy had a median follow-up of 13.7 (range, 0.53-35.57) months. At the time of analysis, six (6, 17.6%) patients were still receiving treatment, 24 patients achieved PR, and the confirmed ORR was 67.65%. Four (4) patients had SD and the DCR was 79.41%. Median PFS (mPFS) was 10.06 months. Median overall survival (mOS) was 23.27 months. The mPFS of the PD-L1 positive population was significantly better than that of the PD-L1 negative population (19.23 months vs. 6.57 months, p = 0.046). There was no significant difference in mPFS and mOS between the high and low TMB groups. Patients were divided into response (CR or PR) and non-response (SD or PD) groups according to their response to combined immunotherapy. The response rate of patients with wild-type MET was higher than that of patients with MET mutation (75% vs. 25%, p=0.0889). Compared to patients with wild-type BRCA2, patients with mutant BRCA2 had a significantly shorter mPFS (4.2 months vs. 13.3 months, p=0.027). Based on the efficacy data and 11 additional clinicopathologic characteristics of the 59 patients who received sintilimab in combination with chemotherapy, a random forest AI model was constructed to predict response to immunochemotherapy. Patients were randomly divided into a training set and a test set in a 6:4 ratio. In the test set, the top three predictive biomarkers were HED, age, and BMI. The area under the curve of the predictive model in the test set was 0.82. Conclusions: Sintilimab combined with chemotherapy could be an effective treatment for advanced ESCC, and an AI model based on multiple clinical features at baseline could predict the efficacy of immunochemotherapy. Further studies with larger sample size are needed to validate our findings.