Confidence interval methods for antimicrobial resistance surveillance data.

Area of Science:

• Epidemiology
• Biostatistics
• Public Health

Background:

• Antimicrobial resistance (AMR) poses a significant global health threat.
• Accurate burden assessment is limited by uncertain AMR prevalence estimates.
• Geographical pooling of AMR data can introduce bias due to population heterogeneity.

Purpose of the Study:

• To evaluate methods for estimating uncertainty in AMR prevalence.
• To compare methods accounting for data clustering versus assuming independence.
• To assess the impact of geographical coverage on confidence interval accuracy.

Main Methods:

• Utilized AMR data from up to 381 US laboratories.
• Constructed confidence intervals using cluster-robust methods and standard independence-assuming methods.
• Analyzed confidence interval accuracy with increasing facility coverage.

Main Results:

• Cluster-robust methods were more likely to include the population mean than independence-assuming methods.
• Increased geographical coverage improved accuracy but did not fully correct for independence assumption violations.
• Bias persists when the clustered data structure is ignored.

Conclusions:

• Standard methods assuming data independence likely yield biased AMR prevalence estimates.
• Accounting for clustered data structure and intra-cluster variation is crucial for accurate AMR confidence intervals.
• Improved uncertainty capture enhances global health burden assessment for AMR.