Patient Subtyping via Learning Hidden Markov Models from Pairwise Co-occurrences in EHR Data

Area of Science:

• Computational biology
• Medical informatics
• Data science

Background:

• Patient subtyping aids in understanding disease progression and clinical characteristics.
• Electronic Health Records (EHR) contain rich data for patient analysis.
• Identifying distinct patient subtypes is crucial for personalized medicine.

Purpose of the Study:

• To develop and validate a novel method for patient subtyping using EHR data.
• To leverage hidden Markov models (HMM) for uncovering latent structures in patient data.
• To derive clinically meaningful patient subtypes for improved medical service categorization.

Main Methods:

• Utilized a hidden Markov model (HMM) approach.
• Applied the model to real-world electronic health record (EHR) data.
• Evaluated the model's ability to identify latent Markovian structures and patient subtypes.

Main Results:

• The HMM-based model successfully identified underlying Markovian structures within EHR data.
• Clinically plausible patient subtypes were derived from the analysis.
• The identified subtypes demonstrated utility in categorizing patients based on their conditions.

Conclusions:

• Hidden Markov models provide an effective framework for patient subtyping from EHR.
• This approach facilitates a deeper understanding of patient heterogeneity and disease trajectories.
• The derived subtypes can inform clinical decision-making and patient management strategies.