Introduction
The efficacy of immune checkpoint inhibitors (ICIs) for advanced esophageal squamous cell carcinoma (ESCC) remains suboptimal. This study aims to construct and validate a clinically accessible model to better identify populations that may potentially benefit from ICIs.
Methods
This study enrolled advanced ESCC patients treated with ICIs at Peking University Cancer Hospital from January 14, 2016, to January 26, 2024, forming the training cohort. Combined positive score (CPS) was recorded to evaluate the predictive value of programmed cell death ligand-1 (PD-L1). Baseline clinical characteristics and laboratory test results were identified as predictors through a 2-phase selection based on Cox proportional hazard regression and minimization of Akaike information criterion (AIC). The prediction model was internally validated using bootstrapping and externally validated in patients from Harbin Medical University Cancer Hospital between January 10, 2019, and July 6, 2022.
Results
A total of 430 patients from Peking University Cancer Hospital and 184 patients from Harbin Medical University Cancer Hospital were ultimately enrolled. PD-L1 expression failed to discriminate survival outcomes (HR=0.94, 95% CI: 0.74-1.19, P = .6). The final model incorporates 10 variables: stage, bone metastasis, line of therapy, treatment, lactate dehydrogenase, carcinoembryonic antigen, carbohydrate antigen 199, lymphocyte count, prognostic nutritional index, and systemic immune-inflammation index. The C-index was 0.725 (95%CI: 0.694-0.756) in the training cohort, 0.722 (95%CI: 0.688-0.751) after bootstrapping, and 0.691 (95%CI: 0.650-0.733) in the external validation cohort, outperforming PD-L1 in prognostic prediction and risk stratification. An interactive online prediction tool (https://escc-survival.shinyapps.io/shiny_app/) was subsequently developed.
Conclusions
This is the first model for individualized survival prediction in advanced ESCC patients treated with ICIs based on large-scale, high-quality real-world data, potentially guiding clinical decision-making and optimize treatment strategies.
