Statistical analysis of single-copy assays when some observations are zero

Area of Science:

• Biostatistics
• Virology
• Immunology

Background:

• Single-copy assays are crucial for quantifying rare entities in HIV cure research, but statistical analysis is challenged by sampling variability and zero-copy findings.
• Common methods for handling zero-copy observations include imputation (replacing zeros with one or adding one) or left-censoring, which may introduce bias.

Discussion:

• This study evaluated four statistical approaches for analyzing data from single-copy assays in HIV cure research.
• Methods involving alteration of zero-copy observations (imputation or censoring) demonstrated more attenuation of intervention effects and reduced power to detect true effects compared to analyzing zeros as unaltered data.
• Negative binomial regression, a method for count data, is proposed as a suitable approach for analyzing unaltered zero-copy observations, preserving the integrity of the data.

Key Insights:

• Altering zero-copy observations in single-copy assay data can lead to biased estimates of intervention effects and reduced statistical power in HIV cure research.
• Analyzing zero-copy observations as unaltered data, particularly with negative binomial regression, provides more accurate and reliable estimates of treatment effects.
• The choice of statistical method significantly impacts the interpretation of results from studies involving rare entities quantified by single-copy assays.

Outlook:

• The findings support the adoption of negative binomial regression for analyzing rare event data in HIV cure research, enhancing the reliability of study outcomes.
• Further research could explore the application of negative binomial regression in other fields dealing with rare event quantification and zero-inflated data.
• Standardizing statistical analysis methods for single-copy assay data will improve the comparability and reproducibility of HIV cure research findings.