Epidemiological predictive modeling: lessons learned from the Kuopio ischemic heart disease risk factor study

Area of Science:

• Epidemiology
• Biostatistics
• Machine Learning

Background:

• Traditional statistical models like Linear, Logistic, and Cox regressions are commonly used in epidemiological studies.
• The application of predictive models in epidemiology remains limited, with a focus on conventional statistical approaches.

Purpose of the Study:

• To investigate the predictive capabilities of machine learning models with embedded variable selection in a high-dimensional epidemiological cohort.
• To compare the performance of various predictive models, including Logistic Lasso Regression, Random Forest, and Multilayer Perceptron, against traditional Logistic Regression for cardiovascular death prediction.

Main Methods:

• Utilized the Kuopio Ischemic Heart Disease Risk Factor Study cohort (1984-1989) with 746 predictor variables for 2682 men.
• Compared Simple Logistic Regression with k-Nearest Neighbors, Logistic Lasso Regression, Decision Tree, Random Forest, and Multilayer Perceptron.
• Evaluated two scenarios for handling competing risks in cardiovascular death prediction over a 30-year follow-up.

Main Results:

• Logistic Lasso Regression achieved the highest average Area Under the Curve (AUC) of 0.8075 (scenario 1) and 0.7155 (scenario 2).
• These AUC values were significantly higher (6.04% and 5.50%) than the baseline AUC from traditional Logistic Regression.
• Models including Logistic Lasso Regression, Random Forest, and Multilayer Perceptron demonstrated superior performance over Simple Logistic Regression.

Conclusions:

• Logistic Lasso Regression, Random Forest, and Multilayer Perceptron models outperformed Simple Logistic Regression in predicting cardiovascular death.
• The study suggests that advanced machine learning models with embedded variable selection offer improved predictive accuracy in high-dimensional epidemiological data.
• These findings advocate for the broader adoption of sophisticated predictive modeling techniques in epidemiological research for enhanced risk assessment.