A Computational Method for Learning Disease Trajectories From Partially Observable EHR Data.

Area of Science:

• Computational biology
• Health informatics
• Disease progression modeling

Background:

• Disease progression varies even with shared risk factors.
• Disease trajectories, or the order of disease manifestation, can predict progression.
• Electronic Health Records (EHRs) contain valuable longitudinal patient data.

Purpose of the Study:

• To develop a novel computational method for learning disease trajectories from EHR data.
• To assess the predictive power of these trajectories for disease progression.
• To improve the interpretability and accuracy of disease progression models.

Main Methods:

• Developed an algorithm for extracting disease trajectories from EHR data.
• Implemented three criteria for filtering and selecting relevant trajectories.
• Utilized a likelihood function to assess the risk of outcomes given trajectory sets.
• Validated the method on EHR data from Mayo Clinic and M Health Fairview.

Main Results:

• The proposed algorithm extracted comprehensive disease trajectories.
• The method explained observed disease progressions significantly better than competing approaches.
• Filtering criteria yielded a small, highly interpretable subset of trajectories.
• A minimal (5%) loss in explanatory power was observed after filtering.

Conclusions:

• The novel computational method effectively learns disease trajectories from EHR data.
• These trajectories offer superior predictive capabilities for disease progression.
• The filtering approach enhances interpretability without significant loss of predictive accuracy.