An approach to nonparametric inference on the causal dose-response function.

Area of Science:

• Biostatistics
• Epidemiology
• Statistical Inference

Background:

• Continuous exposure analysis often relies on parametric models, which can lead to invalid inference due to unmet distributional assumptions.
• Nonparametric methods offer a more robust alternative by requiring only mild assumptions about the data-generating mechanism.

Purpose of the Study:

• To propose a novel nonparametric test for the null hypothesis that a dose-response function is constant.
• To develop a method for constructing simultaneous confidence bands for the dose-response function.

Main Methods:

• The proposed test assesses the variance of the dose-response function, hypothesizing zero variance under the null.
• A novel variance estimator is developed with a characterized null limiting distribution for well-calibrated hypothesis testing.
• Confidence bands are constructed by inverting the proposed hypothesis test.

Main Results:

• The simulation study validates the proposed nonparametric method's performance.
• The approach allows for valid statistical inference on dose-response functions in continuous exposure settings.
• The method is applied to assess the impact of travel distance on HIV retention in care.

Conclusions:

• The developed nonparametric approach provides a valid and robust method for statistical inference on continuous dose-response functions.
• This method addresses limitations of traditional parametric approaches, improving reliability in real-world applications.
• The findings have implications for public health research, particularly in understanding factors affecting patient adherence and retention in care.