Machine Learning Decision Support for Elderly Healthcare Using MIMIC Dataset

Elderly patients often face high risks of chronic diseases such as heart conditions, and improving clinical decision support for their care is a priority. This paper presents a comprehensive review and empirical study of machine learning-driven decision support systems for elderly healthcare, leveraging the MIMIC-III and MIMIC-IV critical care datasets. We examine data preprocessing strategies such as handling missing values via imputation and normalization, feature engineering techniques including domain-specific feature creation and dimensionality reduction, and Machine Learning (ML) models for heart disease risk prediction with an emphasis on ensemble methods such as Random Forests and XGBoost (Extreme Gradient Boosting). We also explore medication recommendation systems powered by ML and discuss approaches to integrating predictive models into user-friendly healthcare dashboards. Our experiments focus on predicting heart disease-related outcomes in patients aged 60 and above using Random Forest and XGBoost, showing that both models achieve strong predictive performance, with XGBoost outperforming Random Forest in terms of AUC. We integrate these findings with the existing literature, highlighting challenges such as data quality, interpretability, and ethical considerations. The results demonstrate the potential of ML, especially tree-based ensembles, to enhance decision support for elderly care, while emphasizing the significance of robust data preprocessing and the need for interpretable and clinically integrated solutions. We conclude with discussions on future research directions, including addressing bias in models, ensuring generalizability, and improving clinician adoption of AI-driven tools.