A hierarchical zero-inflated log-normal model for skewed responses

Area of Science:

• Biostatistics
• Genomics
• Reproductive Biology

Background:

• Zero-inflated count data is well-researched, but zero-inflated lognormal data presents unique challenges.
• Human sperm cell DNA damage, measured by the Comet assay, often exhibits lognormal distribution with excess zeros.
• Existing statistical methods for zero-inflated lognormal data do not adequately handle complex designs with multiple observations per subject across visits.

Purpose of the Study:

• To develop and validate a novel statistical model for zero-inflated lognormal data with hierarchical structures.
• To address inter- and intra-subject variations in longitudinal studies with multiple observations per visit.
• To analyze human sperm cell DNA damage data from Comet assay experiments.

Main Methods:

• A zero-inflated hierarchical model incorporating latent random variables was developed.
• The Expectation-Maximization (EM) algorithm was employed for parameter estimation.
• Parametric bootstrap methods were used to estimate standard errors.
• The model was applied to human sperm cell DNA damage data from Comet assay.

Main Results:

• The proposed hierarchical model effectively accommodates both inter- and intra-subject variations in zero-inflated lognormal data.
• The EM algorithm provided reliable Maximum Likelihood estimates for model parameters.
• Standard errors were accurately estimated using parametric bootstrap, ensuring statistical rigor.
• The model demonstrated utility in analyzing complex human sperm DNA damage data.

Conclusions:

• The developed zero-inflated hierarchical model offers a robust solution for analyzing complex zero-inflated lognormal data, particularly in biological and medical research.
• This approach advances the statistical methodology for handling longitudinal data with multiple observations and significant zero inflation.
• The findings provide a valuable tool for researchers studying DNA damage and other related biological markers.