Wasserstein GAN-based estimation for conditional distribution function with current status data.

Area of Science:

• Statistics
• Machine Learning
• Biostatistics

Background:

• Current status data present unique analytical challenges in medicine, econometrics, and social science.
• Existing statistical methods often fail when models are misspecified.
• Accurate analysis of current status data is crucial for reliable predictions.

Purpose of the Study:

• To propose a model-free, two-stage generative approach for estimating conditional distribution functions.
• To address the limitations of existing methods in analyzing current status data.
• To provide a robust method for handling complex data structures in various scientific fields.

Main Methods:

• A nonparametric, two-stage generative model was developed.
• The approach involves learning a conditional generator for joint distributions.
• Nonparametric maximum likelihood estimators were constructed for conditional distribution functions.

Main Results:

• The proposed estimator demonstrates consistency and favorable convergence properties.
• Simulation studies confirmed superior performance compared to classical modeling approaches.
• The method yielded reasonable predictions in a Parvovirus B19 seroprevalence data application.

Conclusions:

• The deep conditional generative approach offers a robust and accurate method for current status data analysis.
• This model-free technique overcomes limitations of traditional statistical modeling.
• The approach has practical implications for fields like epidemiology and social science research.