Double machine learning (DoubleML)

Double machine learning estimates the effect of one variable, such as drive time to care, on an outcome while adjusting for many confounders. It fits flexible machine-learning models to predict both the outcome and the variable of interest from the confounders, then estimates the effect from what those models cannot explain.

Cross-fitting, where each model is trained on one part of the data and applied to another, prevents overfitting from biasing the estimate and allows valid confidence intervals.

The method still depends on measuring the important confounders. It makes adjustment more flexible; it does not turn observational data into a randomized trial.