Basic Science and Pathogenesis.

Area of Science:

• Medical Informatics
• Machine Learning in Healthcare
• Predictive Analytics

Background:

• Alzheimer's disease and related dementias (ADRD) affect millions, with projected increases and substantial financial costs.
• Early detection and prediction are crucial for managing ADRD and identifying at-risk individuals.
• Leveraging electronic health records (EHR) and claims data offers a scalable solution for ADRD prediction.

Purpose of the Study:

• To develop and validate machine learning models for predicting ADRD up to 10 years in advance.
• To assess the efficacy of predictive models using both EHR and claims data.
• To ensure model explainability and fairness across demographic groups.

Main Methods:

• Utilized the Mimic-IV dataset for EHR data analysis.
• Employed machine learning algorithms on a 5% Medicare claims data sample.
• Explored methods for model interpretability and demographic fairness.

Main Results:

• Achieved an AUC of 0.92 for predicting ADRD up to 10 years in advance using Mimic-IV EHR data.
• Attained AUCs of 0.86, 0.85, and 0.83 for 1, 3, and 5-year ADRD prediction using Medicare claims data.
• Demonstrated the effectiveness of machine learning in predicting ADRD from administrative healthcare data.

Conclusions:

• Machine learning models can effectively predict Alzheimer's and Related Dementias using readily available administrative data.
• EHR and claims data are valuable resources for developing robust ADRD prediction tools.
• The findings support the integration of predictive analytics into healthcare systems for proactive ADRD management.