Two-sample density-based empirical likelihood tests for incomplete data in application to a pneumonia study

Area of Science:

• Biostatistics
• Clinical Trials
• Epidemiology

Background:

• Pneumonia incidence is often measured using invasive and non-invasive procedures in clinical trials.
• A recent trial comparing pneumonia treatments had missing data for invasive procedures based on non-invasive thresholds.
• This missingness pattern created bivariate data dependent on observed non-invasive values.

Purpose of the Study:

• To develop a statistical methodology for comparing treatments with bivariate data exhibiting missingness in one variable.
• To address the challenge of analyzing data where invasive procedure measurements were selectively omitted.
• To provide an efficient and practical approach for handling such missing data patterns in clinical research.

Main Methods:

• Developed a semi-parametric methodology using a density-based empirical likelihood approach.
• Created a novel empirical likelihood method with both parametric and non-parametric components.
• The non-parametric part uses observed data; the parametric part handles missing invasive variable data.

Main Results:

• The methodology was applied to actual data from a pneumonia clinical trial.
• The empirical likelihood approach provided a non-parametric approximation to Neyman-Pearson-type test statistics.
• The method demonstrated efficiency and practicality for analyzing the complex dataset.

Conclusions:

• The developed semi-parametric empirical likelihood method is effective for comparing treatments with bivariate data and missingness.
• This approach offers a robust solution for handling selectively missing invasive data in pneumonia studies.
• The method is a valuable tool for biostatisticians and researchers conducting clinical trials with similar data structures.