Analysis of non-ignorable missing and left-censored longitudinal data using a weighted random effects tobit model

Area of Science:

• Biostatistics
• Longitudinal Data Analysis
• Biomarker Research

Background:

• Longitudinal studies often encounter non-ignorable missing data due to patient discharge or death.
• Response measurements can be left-censored due to detection limits.
• Existing methods may not adequately handle both non-ignorable missingness and left-censoring.

Purpose of the Study:

• To extend the random effects tobit regression model for analyzing longitudinal data with non-ignorable missingness and left-censoring.
• To introduce a weighted random effects tobit regression model using augmented inverse probability weighting.
• To evaluate the performance of the proposed model against existing methods.

Main Methods:

• Development of a weighted random effects tobit regression model.
• Computation of weights using an augmented inverse probability weighted methodology.
• Extensive simulation studies to compare model performance.
• Application to real-world interleukin-6 biomarker data from a sepsis study.

Main Results:

• The proposed weighted random effects tobit model provides consistent estimates.
• The augmented inverse probability weights lead to minimal root mean square errors.
• The model effectively handles non-ignorable missing and left-censored longitudinal data.
• Successful application to interleukin-6 biomarker data.

Conclusions:

• The weighted random effects tobit regression model is a robust approach for analyzing complex longitudinal data.
• The augmented inverse probability weighted methodology improves estimation accuracy.
• This method offers a valuable tool for biomarker studies with missing and censored data.