A Bayesian approach estimating treatment effects on biomarkers containing zeros with detection limits

Area of Science:

• Occupational and Environmental Health
• Biostatistics
• Toxicology

Background:

• Biomarker measurements in occupational and environmental health studies can be bimodally distributed with a spike at zero.
• Left-censoring due to assay detection limits complicates distinguishing true zeros from low measurements.
• Intervention effects should be estimated using only exposed individuals.

Purpose of the Study:

• To present novel Bayesian approaches for analyzing biomarker data with true zeros and left-censoring.
• To provide a statistical framework that accurately accounts for unexposed individuals with zero biomarker values.
• To compare the performance of Bayesian methods against existing maximum likelihood approaches.

Main Methods:

• Development of Bayesian methods utilizing non-standard mixture distributions.
• Application of methods to simulation studies for performance evaluation.
• Analysis of a randomized chemoprevention trial in Qidong, China.

Main Results:

• The proposed Bayesian methods effectively handle true zeros and left-censored data.
• Simulation studies demonstrate the advantages of the Bayesian approach.
• The methods were successfully applied to a real-world chemoprevention trial.

Conclusions:

• Bayesian approaches with non-standard mixture distributions offer a robust solution for biomarker data with true zeros.
• Accurate analysis of biomarker data is crucial for estimating intervention effects in environmental health.
• The study provides a valuable statistical tool for researchers in occupational and environmental health.