A Machine Learning Computational Framework Develops a Multiple Programmed Cell Death Index for Improving Clinical Outcomes in Bladder Cancer.

ABSTRACT

Comprehensive action patterns of programmed cell death (PCD) in bladder cancer (BLCA) have not yet been thoroughly investigated. Here, we collected 19 different PCD patterns, including 1911 PCD-related genes, and developed a multiple programmed cell death index (MPCDI) based on a machine learning computational framework. We found that in the TCGA-BLCA training cohort and the independently validated GSE13507 cohort, the patients with high-MPCDI had a worse prognosis, whereas patients with low-MPCDI had a better prognosis. By combining clinical characteristics with the MPCDI, we constructed a nomogram. The C-index demonstrated that the nomogram was significantly more accurate compared to other variables, including MPCDI, age, gender, and clinical stage. The results of the decision curve analysis demonstrated that the nomogram had a better net clinical benefit compared to other clinical variables. Subsequently, we revealed the heterogeneity of BLCA patients, with significant differences in terms of overall immune infiltration abundance, immunotherapeutic response, and drug sensitivity in the two MPCDI groups. Encouragingly, the high-MPCDI patients showed better efficacy for commonly used chemotherapeutic drugs than the low-MPCDI patients, which suggests that MPCDI scores have a guiding role in chemotherapy for BLCA patients. In conclusion, the MPCDI developed and verified in this study is not only an emerging clinical classifier for BLCA patients, but it also serves as a reliable forecaster for both chemotherapy and immunotherapy, which can guide clinical management and clinical decision-making for BLCA patients.