Estimation of individual treatment effect in observational data iscomplicated due to the challenges of confounding and selection bias. A usefulinferential framework to address this is the counterfactual (potentialoutcomes) model which takes the hypothetical stance of asking what if anindividual had received both treatments. Making use of random forests (RF)within the counterfactual framework we estimate individual treatment effects bydirectly modeling the response. We find accurate estimation of individualtreatment effects is possible even in complex heterogeneous settings but thatthe type of RF approach plays an important role in accuracy. Methods designedto be adaptive to confounding, when used in parallel with out-of-sampleestimation, do best. One method found to be especially promising iscounterfactual synthetic forests. We illustrate this new methodology byapplying it to a large comparative effectiveness trial, Project Aware, in orderto explore the role drug use plays in sexual risk. The analysis revealsimportant connections between risky behavior, drug usage, and sexual risk.
展开▼
