Matched Forest

Matched Forest identifies exposure variables and interaction effects in high-dimensional matched case-control studies by transforming matched data with counterfactuals and applying a modified variable importance score derived from supervised learning.

Key Features:

• High-Dimensional Data Handling: Operates on datasets with large numbers of explanatory variables typical of modern biomedical studies.
• Interaction Detection: Detects interaction effects among variables to capture complex joint associations.
• Potential Outcome Model Basis: Leverages the potential outcome model framework to accommodate varying numbers of matching and exposure variables.
• Counterfactual Data Transformation: Transforms matched case-control data by adding counterfactuals that preserve both case and control values for each instance.
• Modified Variable Importance Score: Computes a modified variable importance score derived from supervised learning to prioritize significant predictors, including main effects and interactions.

Scientific Applications:

• Variable selection in matched case-control studies: Identifies exposure variables associated with health outcomes within matched case-control designs.
• Interaction discovery for exposure-outcome relationships: Detects interaction effects among exposures to support analyses of complex biological relationships and potential causal pathways.

Methodology:

The method transforms matched case-control data by adding counterfactuals within a potential outcome framework and then applies a supervised-learning-derived modified variable importance score to identify significant variables and interactions.