Logistic regression was as good as machine learning for predicting major chronic diseases

Area of Science:

• Biostatistics
• Epidemiology
• Machine Learning in Healthcare

Background:

• Predicting major chronic diseases like cardiovascular diseases (CVDs), chronic kidney disease (CKD), diabetes (DM), and hypertension (HTN) is crucial for public health.
• Machine learning (ML) models are increasingly explored for their potential in disease risk prediction.
• The comparative performance of various ML algorithms against traditional logistic regression using simple clinical predictors requires evaluation.

Purpose of the Study:

• To assess the predictive performance of five different ML algorithms.
• To compare the efficacy of ML models against logistic regression in predicting the risk of CVD, CKD, DM, and HTN.
• To determine the optimal model for risk prediction using simple clinical data in Asian adults.

Main Methods:

• A prospective cohort study involving 6,762 Asian adults was conducted.
• Five ML models were evaluated: neural network, support vector machine, random forest, gradient boosting machine, and k-nearest neighbor.
• Performance was compared against standard logistic regression using area under the receiver operating characteristic curve (AUC).

Main Results:

• Incidences at 6 years were 4.0% for CVD, 7.0% for CKD, 9.2% for DM, and 34.6% for HTN.
• Logistic regression achieved the highest AUC for CKD (0.905) and DM (0.768).
• Neural network (0.753) and support vector machine (0.780) performed best for CVD and HTN, respectively, with minimal differences compared to logistic regression.

Conclusions:

• Logistic regression demonstrated comparable performance to ML models in predicting the risk of major chronic diseases.
• The predictive accuracy differences between ML models and logistic regression were small and not statistically significant.
• For conditions with low incidence and using simple clinical predictors, logistic regression remains a highly effective tool for risk prediction.