A disease potential-driven graph attention model for comorbidity risk prediction of hypertension

Area of Science:

• Medical Informatics
• Computational Biology
• Cardiovascular Disease Research

Background:

• Hypertension significantly increases the risk of severe complications and comorbidities.
• Current data-driven comorbidity prediction models lack clinical plausibility and interpretability.
• Effective methods are needed to integrate patient features and identify individual differences for accurate risk prediction.

Purpose of the Study:

• To develop an advanced model for predicting hypertension comorbidities.
• To address limitations of existing data-driven prediction methods.
• To enhance the interpretability and early detection of hypertension-related comorbidities.

Main Methods:

• Proposed a novel Disease Potential-Driven Graph Attention (DP-GA) model.
• Integrated patient disease features and structural information using a fusion mechanism.
• Implemented a similarity-difference balance mechanism to differentiate patient relationships.
• Designed a disease potential-driven attention mechanism for risk assessment and mask construction.

Main Results:

• The DP-GA model demonstrated significant improvements in comorbidity risk prediction for hypertension patients.
• Outperformed baseline and state-of-the-art methods across three comorbidity datasets.
• Analysis of the comorbidity network enhanced prediction interpretability and early detection capabilities.

Conclusions:

• The DP-GA model offers a robust and interpretable approach for hypertension comorbidity risk prediction.
• This method effectively captures fused patient features and identifies critical relationships.
• The findings support improved clinical decision-making and patient management for hypertension.