Incorporating Patient Similarity and Clinical Temporality in Disease Prognostic Modeling

Area of Science:

• Biomedical Informatics
• Artificial Intelligence in Healthcare
• Clinical Decision Support

Background:

• Existing health recommender systems (HRSs) often overlook patient phenotype correlations, impacting prognostic accuracy.
• Incorporating patient similarity into prognostic modeling is crucial for improving diagnostic capabilities.

Purpose of the Study:

• To explore correlations within diagnosis, procedure, and medication data.
• To improve diagnostic accuracy by integrating patient similarity and clinical temporality.

Main Methods:

• Developed SIM-PR, a static model using patient similarity and PageRank on personalized patient graphs.
• Created dynamic temporal prediction models utilizing multilayer perceptron and long short-term memory networks.
• Compared performance against standard supervised machine learning baselines.

Main Results:

• Proposed static and temporal models significantly reduced false positives and negatives.
• Achieved superior predictive accuracy compared to existing health recommender systems.
• Demonstrated effectiveness on the MIMIC-III database.

Conclusions:

• Patient similarity and clinical temporality are key factors for enhancing prognostic models.
• The developed static and dynamic models offer improved diagnostic accuracy in health recommender systems.
• This approach advances personalized medicine and clinical decision support.