|
Abstract:
|
Covariate-adjusted dose-response curves are helpful to describe the effect of a continuous exposure on an outcome while adjusting for potential confounders. Classical methods for estimating such curves often rely on restrictive parametric assumptions, which carry significant risk of model misspecification. It is therefore of interest to estimate and draw inference about these curves without strong modeling assumptions. Nonparametric estimation in this context is challenging because in a nonparametric model these curves cannot be estimated at regular rates and available estimators will generally be sensitive to the selection of tuning parameters. In this work, we show that if the curve is monotone, nonparametric estimation and inference is possible without the need to select tuning parameters and under minimal smoothness conditions. Our estimator builds upon tools from shape-constrained inference and targeted learning. We describe theoretical properties of our estimator, illustrate its practical performance via numerical studies, and apply our method to estimate the covariate-adjusted continuous effect of BMI on immune response in data from multiple HIV vaccine trials.
|
