Development and Validation of Comorbidity Severity Adjustment Methods in Mortality Models for Acute Cerebrovascular Disease Using Survival and Machine Learning Analyses

<b>Background/Objectives</b>: This study aimed to develop and validate comorbidity-based severity adjustment methods for acute cerebrovascular disease by recalibrating the Charlson Comorbidity Index (CCI) and constructing a CCS-based comorbidity index to improve mortality risk prediction. <b>Methods</b>: Using the Korea Disease Control Agency's Discharge Injury In-depth Survey Dataset (2013-2022), we applied Cox proportional hazards regression and machine learning techniques, including LASSO, CART, Random Forests, GBM, and ANN, to recalibrate the CCI and develop a CCS-based comorbidity index. <b>Results</b>: The recalibrated Charlson Comorbidity Index (m-CCI) and the newly developed CCS-based comorbidity index (m-CCS) demonstrated improved predictive performance for in-hospital mortality. Among the machine learning models, GBM (AUC = 0.835) and ANN (AUC = 0.830) demonstrated the highest predictive accuracy, with m-CCS consistently outperforming other indices. <b>Conclusions</b>: The recalibrated m-CCI and newly developed m-CCS comorbidity indices enhance mortality risk adjustment for acute cerebrovascular disease patients in Korea. The superior performance of machine learning models underscores their potential for enhancing severity adjustment in hospital benchmarking and quality assessment.