Estimating Effects with Rare Outcomes and High Dimensional Covariates: Knowledge is Power

Area of Science:

• Epidemiology
• Biostatistics
• Causal Inference

Background:

• Rare outcomes are common in clinical research, but analytical methods for estimating causal effects are limited.
• Existing methods may lack power and stability when dealing with rare events, hindering scientific investigation.

Purpose of the Study:

• To develop a novel targeted minimum loss-based estimator (TMLE) for estimating causal effects and associations with rare outcomes.
• To improve the statistical power and stability of estimations for rare events.

Main Methods:

• Constructed a new TMLE incorporating bounds on the conditional mean of the outcome.
• Focused on causal risk difference and statistical models with exposure and confounders.
• Evaluated performance through finite sample simulations comparing with existing estimators.

Main Results:

• The proposed TMLE demonstrated comparable or superior performance to alternative methods like propensity score matching and IPTW.
• The estimator provided consistent estimates when either the conditional mean outcome or propensity score was consistently estimated.
• TMLE ensured point estimates remained within the valid parameter range, enhancing reliability.

Conclusions:

• The developed TMLE offers a robust and efficient semiparametric approach for analyzing rare events, even with high-dimensional covariates.
• This method enhances the investigation of associations between exposures and rare outcomes, such as neighborhood norms and alcohol use disorder.