Using Bayesian networks with Tabu-search algorithm to explore risk factors for hyperhomocysteinemia

Area of Science:

• Cardiovascular Science
• Biostatistics
• Medical Informatics

Background:

• Hyperhomocysteinemia (HHcy) is strongly associated with cardiovascular and cerebrovascular diseases.
• Identifying and intervening in HHcy risk factors is crucial for disease prevention.
• Traditional Logistic Regression models have limitations in analyzing complex risk factor relationships.

Purpose of the Study:

• To construct a Tabu Search-based Bayesian Network (BNs) for identifying hyperhomocysteinemia risk factors.
• To compare the performance of Tabu Search-based BNs with Hill Climbing-based BNs and Logistic Regression.
• To explore the clinical application and predictive capabilities of BNs in managing HHcy.

Main Methods:

• A Tabu Search algorithm was employed to build a Bayesian Network model for HHcy.
• Maximum Likelihood Estimation was used to calculate conditional probabilities between network nodes.
• Performance comparison was conducted against Hill Climbing-based BNs and Logistic Regression models.

Main Results:

• Direct risk factors for HHcy identified: Age, sex, α1-microglobulin to creatinine ratio, fasting plasma glucose, diet, and systolic blood pressure.
• Indirect risk factors for HHcy identified: Smoking, glycosylated hemoglobin, and BMI.
• Tabu Search-based BNs demonstrated superior performance over Hill Climbing-based BNs in risk factor detection.

Conclusions:

• Bayesian Networks with the Tabu Search algorithm offer a valuable supplement to Logistic Regression for analyzing HHcy.
• BNs effectively reveal complex network relationships and overall linkages between HHcy and its risk factors.
• Bayesian reasoning provides a more robust approach for clinical risk prediction of HHcy, warranting wider adoption.