Scalable Counterfactual Risk Estimation for Rare Events in Longitudinal Data
Researchers have developed a new subsampling and reweighting strategy to make causal effect estimation more computationally efficient for rare events in longitudinal data. This method addresses the challenges of high computational cost and class imbalance that plague existing g-formula-based techniques like the ICE estimator. The proposed approach enhances estimation stability and reduces computational burden, as demonstrated in simulations and a large-scale EHR cohort study on suicide risk. AI
IMPACT Improves the accuracy and efficiency of causal inference models, particularly for rare events in healthcare data.