An overview of modern machine learning methods for effect measure modification analyses in high-dimensional settings

Area of Science:

• Public Health
• Epidemiology
• Biostatistics

Background:

• Identifying heterogeneous exposure effects across population subgroups is crucial for public health policy and assessing research validity.
• Traditional methods for effect measure modification are often impractical in high-dimensional data settings.
• Machine learning offers data-driven approaches to estimate heterogeneous effects, but doesn't directly identify effect modifiers.

Purpose of the Study:

• To summarize and explain machine learning methods for effect measure modification analyses.
• To discuss the application of these methods for discovering vulnerable subgroups.
• To provide a reference for public health researchers implementing these techniques.

Main Methods:

• Review and explanation of machine learning techniques for estimating heterogeneous exposure effects.
• Discussion on how these methods can be adapted for effect measure modification.
• Demonstration using R implementation and a case study on drought and child stunting.

Main Results:

• Machine learning methods can estimate heterogeneous exposure effects, aiding in the discovery of vulnerable subgroups.
• These data-driven approaches can supplement traditional methods in high-dimensional settings.
• The case study illustrates the practical application of these methods in public health research.

Conclusions:

• Machine learning methods offer a valuable, data-driven approach to supplement traditional analyses for effect measure modification.
• These techniques can assist in identifying previously unknown vulnerable populations.
• Further research and implementation in R are encouraged for public health applications.