Background: Older adults living with Alzheimer’s Disease and Related Dementias (ADRD) have an increased risk of mortality compared to those without an ADRD diagnosis. Various statistical methods have been used to predict mortality in the ADRD population. However, they face several limitations when applied to high-dimensional data. We compared the predictive performance of three machine learning models using the South Carolina Alzheimer’s Disease Registry (SCADR).

Method: Data from the South Carolina Alzheimer’s Disease Registry (2007–2018) were analyzed. Three machine learning models, LASSO regression, Random Forest, and Support Vector Machines, were developed in R using the glmnet, ranger, and kernlab packages. Data were split into training (70%) and testing (30%) sets, with normalization and stratification by mortality status. Class imbalance was addressed using the Synthetic Minority Over-sampling Technique (SMOTE). Five-fold cross-validation was used to tune model hyperparameters. Optimal classification thresholds were determined using Youden’s Index. Model performance was compared using the area under the receiver operating characteristic curve (ROC-AUC).

Results: A total of 13,620 participants from the SCADR were included (mean age = 77 years), with an overall mortality rate of 71%. Twenty-five predictors were evaluated across models. The LASSO model demonstrated strong discrimination for mortality prediction (ROC-AUC = 0.81). The Support Vector Machine model achieved the highest overall accuracy (74%), outperforming both LASSO and Random Forest models. Nine predictors, including year of diagnosis, race, sex, and disease-specific conditions, were consistently identified as influential across all three machine learning approaches.

Conclusion: Machine learning models demonstrated moderate to very good discrimination and good overall accuracy in predicting mortality within the Registry. These approaches may be extended to predict other clinically relevant outcomes and support risk stratification, informing tailored interventions and resource allocation in the Registry.

Keywords: Alzheimer’s Disease and Related Dementias, Machine Learning, Mortality, Prediction, Outcomes