[The comparison of logistic regression model selection methods for the prediction of coronary artery disease]

Area of Science:

• Cardiology
• Biostatistics
• Medical Informatics

Background:

• Coronary artery disease (CAD) poses a significant global health challenge.
• Accurate prediction models are crucial for early diagnosis and intervention.
• Logistic regression is a widely used statistical method for binary outcomes like disease presence.

Purpose of the Study:

• To compare the effectiveness of different logistic regression model selection methods for predicting coronary artery disease (CAD).
• To evaluate the performance of various methods in identifying significant predictors of CAD.

Main Methods:

• Utilized coronary artery disease data from 237 patients at Inönü University Faculty of Medicine.
• Applied logistic regression model selection methods to data with continuous and discrete independent variables.
• Assessed model goodness-of-fit using the Hosmer-Lemeshow statistic and compared models with the likelihood-ratio statistic.

Main Results:

• All tested logistic regression model selection methods demonstrated high performance, with sensitivity, specificity, and accuracy rates exceeding 91.9%.
• The Hosmer-Lemeshow statistic confirmed the success of these methods in describing CAD data.
• Key factors associated with CAD were identified and evaluated.

Conclusions:

• Logistic regression model selection methods are highly effective for predicting coronary artery disease (CAD).
• Stepwise model selection methods proved superior to the Enter method when assessed by the likelihood-ratio statistic.
• Predictors such as age, diabetes mellitus, hypertension, family history, smoking, LDL, triglycerides, stress, and obesity are valuable for CAD prediction.