The study protocol was approved by institutional review boards, with informed consent from all patients. A real-world cohort of 3,278 patients treated with
ICI
○
Search Drug
from 2014-2019 was analyzed, excluding certain patients, resulting in 2,035 patients across 17 cancer types. Clinical and demographic data were collected, with missing values imputed. Machine learning models were constructed to predict clinical benefit and overall survival, using feature selection and ensemble models. The study also compared the prognostic performance of machine learning models with TMB and PD-L1, and analyzed bulk RNA-seq data for immune cell composition.