Mixed effects models with censored data with application to HIV RNA levels

Area of Science:

• Biostatistics
• Longitudinal Data Analysis
• Clinical Laboratory Science

Background:

• Mixed effects models are standard for analyzing continuous longitudinal data.
• Laboratory measurements often have detection limits, leading to censored data.
• Standard estimation procedures may be biased with censored outcomes.

Purpose of the Study:

• To adapt the Expectation-Maximization (EM) algorithm for mixed effects models.
• To incorporate handling of left and/or right censored data in longitudinal studies.
• To improve parameter estimation accuracy for laboratory measurements with detection limits.

Main Methods:

• Modification of the standard EM estimation procedure for mixed effects models.
• Inclusion of algorithms to address both lower and upper censoring (detection limits).
• Application to longitudinal studies with continuous, potentially censored, outcomes.

Main Results:

• The modified EM procedure effectively accounts for left and/or right censoring.
• Accurate estimation of fixed effects and variance components is achieved.
• The method provides reliable results even with data subject to detection limits.

Conclusions:

• The proposed modified EM estimation is a robust approach for longitudinal data with censoring.
• This method enhances the analysis of laboratory measurements affected by detection limits.
• It offers a valuable tool for biostatisticians and researchers in related fields.