Optimal method for determining the intraclass correlation coefficients of urinary biomarkers such as...

Area of Science:

• Environmental Epidemiology
• Biomarker Validation
• Statistical Methods

Background:

• Urinary biomarkers are vital in epidemiological studies for assessing chemical exposure.
• Reliability of these biomarkers is critical to prevent misclassification and bias.
• Standardized methods for calculating reliability, such as the intraclass correlation coefficient (ICC), for urinary biomarkers are lacking.

Purpose of the Study:

• To evaluate different imputation methods for left-censored urinary biomarker data before calculating ICCs.
• To assess the impact of the left-censoring proportion on estimated ICCs.
• To propose a method for estimating true ICCs.

Main Methods:

• Comparison of five methods for handling left-censored data: Gibbs sampler imputation, univariate distribution fitting, multiple imputation by chained equation, bootstrap expectation-maximization, and single value substitution.
• Calculation of ICCs using these imputation methods.
• Mathematical assessment of the relationship between censoring proportion and ICC.
• Organophosphate pesticide metabolites (dialkylphosphates) were used as example biomarkers.

Main Results:

• The Gibbs sampler-based imputation method demonstrated superior performance for values below reporting limits, indicated by lower Kolmogorov-Smirnov test statistics.
• Across all tested imputation methods, the estimated ICCs decreased as the proportion of left-censored data increased.
• The performance of imputation methods varied, with Gibbs sampling showing the best imputation quality.

Conclusions:

• The Gibbs sampler approach is recommended for imputing left-censored urinary biomarker data.
• Increased censoring rates significantly attenuate ICC estimates, highlighting the need for robust imputation strategies.
• A mathematical method is proposed to estimate true ICCs, accounting for censoring effects.