A Comorbidity Knowledge-Aware Model for Disease Prognostic Prediction.

Area of Science:

• Medical Informatics
• Machine Learning in Healthcare
• Clinical Decision Support

Background:

• Prognostic prediction models are crucial for estimating patient disease risk, aiding healthcare providers and patients by reducing harm and costs.
• Existing models often prioritize either predictive performance or interpretability, creating a gap in comprehensive clinical utility.
• There is a need for models that simultaneously achieve high accuracy and provide transparent, understandable decision-making logic.

Purpose of the Study:

• To develop a novel knowledge-aware Bayesian model for prognostic prediction.
• To simultaneously address model accuracy and transparency (explainability) in risk estimation.
• To evaluate the model's performance against state-of-the-art methods using real-world electronic health records.

Main Methods:

• Development of a knowledge-aware Bayesian framework integrating accuracy and transparency.
• Utilizing territory-wide electronic health records spanning four years for case studies.
• Comparative analysis against existing prognostic prediction models on key performance metrics.

Main Results:

• The proposed Bayesian model demonstrated superior performance compared to state-of-the-art models in terms of accuracy and c-statistic.
• The model successfully generated intelligible and explainable rules, enhancing its interpretability.
• Validation through real-world case studies confirmed the model's practical applicability and effectiveness.

Conclusions:

• The novel knowledge-aware Bayesian model effectively balances predictive accuracy with rule-based transparency in prognostic prediction.
• This approach offers a significant advancement for clinical decision support systems, providing reliable and understandable risk assessments.
• The model's ability to generate explainable rules facilitates trust and adoption by healthcare practitioners.