Reliability of environmental sampling culture results using the negative binomial intraclass correlation coefficient

Area of Science:

• Veterinary epidemiology
• Statistical modeling
• Environmental microbiology

Background:

• The Intraclass Correlation Coefficient (ICC) is vital for assessing measurement reliability across different sources.
• Traditional methods often require data transformation (e.g., natural logarithm) for linear mixed model (LMM) based ICC estimation, particularly for overdispersed data.
• Previous studies utilized log-transformed culture results for estimating Mycobacterium avium subsp. paratuberculosis reliability in environmental samples.

Purpose of the Study:

• To evaluate the performance of a newly defined negative binomial ICC for analyzing overdispersed count data from environmental samples.
• To compare the accuracy of the negative binomial ICC with the traditional LMM-based ICC, including transformed data.
• To assess the reliability of environmental sampling for Mycobacterium avium subsp. paratuberculosis using the negative binomial ICC.

Main Methods:

• Defined and applied a negative binomial ICC based on a generalized linear mixed model for negative binomial distributed data.
• Included fixed effects in the negative binomial ICC model for analyzing culture results from environmental samples.
• Conducted simulations with diverse inputs and negative binomial distribution parameters (r; p) to compare performance against LMM-based ICC with various data transformations (natural logarithm, square root).

Main Results:

• The negative binomial ICC demonstrated superior performance compared to the LMM-based ICC, even when data was transformed.
• Simulations indicated that the negative binomial ICC provided a more accurate estimation of reliability for overdispersed data.
• A secondary comparison confirmed that the mean of estimated ICC values closely approximated the true ICC across a broad range of values.

Conclusions:

• The negative binomial ICC offers a more robust and accurate method for estimating reliability with overdispersed count data from environmental samples.
• This advanced statistical approach eliminates the need for data transformations often required by traditional LMM-based methods.
• The negative binomial ICC enhances the reliability assessment of environmental sampling, particularly for pathogens like Mycobacterium avium subsp. paratuberculosis.