Study on influencing factors of age-adjusted Charlson comorbidity index in patients with Alzheimer's disease based on machine learning model.

Background: Alzheimer's disease (AD) is a widespread neurodegenerative disease, often accompanied by multiple comorbidities, significantly increasing the risk of death for patients. The age adjusted Charlson Comorbidity Index (aCCI) is an important clinical tool for measuring the burden of comorbidities in patients, closely related to mortality and prognosis. This study aims to use the MIMIC-V database and various regression and machine learning models to screen and validate features closely related to aCCI, providing a theoretical basis for personalized management of AD patients. Methods: The research data is sourced from the MIMIC-V database, which contains detailed clinical information of AD patients. Multiple logistic regression, LASSO regression, random forest, Support Vector Machine (SVM), and Extreme Gradient Boosting (XGBoost) models were used to screen for feature factors significantly correlated with aCCI. By comparing model performance, evaluating the classification ability and prediction accuracy of each method, and ultimately selecting the best model to construct a regression model and a nomogram. The model performance is evaluated through classification accuracy, net benefit, and robustness. The feature selection results were validated by regression analysis. Results: Multiple models have performed well in classifying aCCI patients, among which the model constructed using LASSO regression screening feature factors has the best performance, with the highest classification accuracy and net benefit. LASSO regression identified the following 11 features closely related to aCCI: age, respiratory rate, base excess, glucose, red blood cell distribution width (RDW), alkaline phosphatase (ALP), whole blood potassium, hematocrit (HCT), phosphate, creatinine, and mean corpuscular hemoglobin (MCH). The column chart constructed based on these feature factors enables intuitive prediction of patients with high aCCI probability, providing a convenient clinical tool. Conclusion: The results of this study indicate that the features screened by LASSO regression have the best predictive performance and can significantly improve the predictive ability of aCCI related comorbidities in AD patients. The column chart constructed based on this feature factor provides theoretical guidance for personalized management and precise treatment of AD patients.