Medical Profile Model: Scientific and Practical Applications in Healthcare.

Area of Science:

• Medical Informatics
• Machine Learning
• Computational Biology

Background:

• Electronic health records (EHRs) contain valuable longitudinal patient data.
• Effective representation learning is crucial for extracting insights from complex EHR data.
• Existing methods may not fully capture the temporal nature of disease progression.

Purpose of the Study:

• To develop a novel unsupervised representation learning method for EHRs.
• To create generalized patient profiles incorporating demographic and disease data.
• To demonstrate the utility of learned patient embeddings in downstream tasks.

Main Methods:

• Patient histories represented as temporal sequences of diseases.
• Unsupervised learning using a transformer-based neural network.
• Integration of demographic parameters into the embedding space.
• Training on a large-scale dataset of over one million patients.

Main Results:

• The proposed model significantly outperforms state-of-the-art methods in diagnosis prediction.
• A new Harbinger Disease Discovery method was developed, aiding epidemiological study design.
• Patient embeddings improved performance metrics in insurance scoring tasks.

Conclusions:

• The developed patient profile model offers a powerful tool for EHR representation learning.
• The model facilitates knowledge transfer across medical domains.
• Applications demonstrate significant advancements in disease prediction, discovery, and risk assessment.