Disease progression subtype discovery from longitudinal EMR data with a majority of missing values and unknown...

Area of Science:

• Biomedical Informatics
• Machine Learning
• Data Science

Background:

• Electronic medical records (EMR) offer valuable longitudinal patient data for research.
• Utilizing EMR for disease progression analysis is challenging due to missing data and undefined time series start points.
• Identifying disease progression subtypes is crucial for medical research and patient care.

Purpose of the Study:

• To present a novel Bayesian machine learning model for analyzing complex EMR data.
• To overcome limitations of missing data and variable patient disease stages in time-series analysis.
• To identify consistent disease progression subtypes from longitudinal EMR data.

Main Methods:

• Developed a Bayesian machine learning model capable of handling highly incomplete time-series measurements.
• The model accommodates varying lengths of patient data and aligns similar disease trajectories.
• The approach identifies consistent disease progression subtypes within patient cohorts.

Main Results:

• Successfully applied the model to identify chronic kidney disease progression subtypes.
• Demonstrated the model's ability to manage missing data and diverse patient disease stages.
• Validated the model's effectiveness in discovering meaningful disease progression patterns.

Conclusions:

• The proposed Bayesian machine learning model effectively addresses challenges in analyzing EMR time-series data.
• This method enables the discovery of disease progression subtypes, advancing medical research.
• The approach has significant potential for improving patient care through subtype-specific insights.