Integrating Machine Learning into Statistical Methods in Disease Risk Prediction Modeling: A Systematic Review

Area of Science:

• Medical Informatics
• Computational Biology
• Biostatistics

Background:

• Disease prediction models often rely solely on statistical methods or machine learning, increasing error risk.
• Integrating machine learning into statistical methods may enhance prediction model robustness.
• Current global development of integrated disease prediction models requires comprehensive assessment.

Purpose of the Study:

• To systematically review and assess the development of global disease prediction models that integrate machine learning with statistical methods.
• To identify common integration strategies, application scenarios, and performance metrics.

Main Methods:

• A systematic literature search was conducted across multiple databases (PubMed, EMbase, Web of Science, CNKI, VIP, WanFang, SinoMed) up to May 2023.
• Studies focusing on prediction models integrating machine learning and statistical methods were included.
• Data extracted included study characteristics, integration approaches, application areas, modeling details, and performance.

Main Results:

• Twenty-one studies (20 English, 1 Chinese) were included, focusing on diagnostic (5) and prognostic/occurrence prediction (16) models.
• Integration strategies included voting, stacking, and model selection for classification, and statistical combinations for regression.
• Integrated models demonstrated superior performance (AUROC > 0.75) compared to single methods in most cases, with stacking suitable for high-predictor scenarios.

Conclusions:

• Research on integrating machine learning with statistical methods for disease prediction is emerging but shows significant potential.
• Integrated models can outperform individual statistical or machine learning approaches.
• This review offers guidance on selecting integration methods and highlights the need for further research on strategy improvement and validation.