Maximum likelihood inference for left-censored HIV RNA data.

Area of Science:

• Biostatistics
• Virology
• Clinical Trials

Background:

• Bioassays frequently encounter left-censored data due to assay limits.
• Plasma HIV RNA measurements in studies like the Hemophilia Growth and Development Study are often left-censored.
• This censoring impacts viral load quantification and outcome association assessments.

Purpose of the Study:

• To evaluate the impact of left-censoring on plasma HIV RNA measurements.
• To compare statistical methods for analyzing left-censored bioassay data.
• To assess viral load quantification and its association with outcomes under censoring.

Main Methods:

• Comparison of maximum likelihood estimation (MLE) with substitution methods (LOQ, LOQ/2) and multiple imputation (MI).
• Analysis of left-censored plasma HIV RNA data from the Hemophilia Growth and Development Study.
• Simulations under Gaussian and mixture of Gaussian distributions to assess robustness and sensitivity.

Main Results:

• Maximum likelihood estimation demonstrated the least bias in quantifying viral load and its associations.
• Multiple imputation using a censored Gaussian model showed competitive performance to MLE.
• Substitution methods were generally less accurate than MLE and MI.

Conclusions:

• Maximum likelihood estimation is the preferred method for analyzing left-censored HIV RNA data.
• Multiple imputation offers a computationally simpler, yet competitive, alternative.
• Accurate handling of left-censored data is crucial for reliable bioassay results.