Rethinking statistical approaches for serological data analysis for viral surveillance

Area of Science:

• * Epidemiology and Biostatistics
• * Infectious Disease Surveillance
• * Zoonotic Pathogen Research

Background:

• * Accurate serological surveillance is vital for early detection and understanding emerging zoonotic diseases.
• * Choosing appropriate statistical methods is critical for differentiating seronegative/seropositive samples and estimating prevalence and risk factors.
• * Standard cutoff methods have limitations in propagating uncertainty, unlike advanced approaches like Gaussian mixture models.

Purpose of the Study:

• * To compare the performance of cutoff and clustering approaches for analyzing serological data.
• * To evaluate methods using simulated datasets reflecting real-world epidemiological, biological, and immunological processes.
• * To provide empirical guidance for selecting robust analysis methods for understudied pathogens with limited assay validation.

Main Methods:

• * Simulation of serological datasets for understudied pathogens, mimicking data generation processes.
• * Comparison of standard deviation-based cutoff methods with Bayesian mixture models (clustering approach).
• * Quantification of model performance using coverage and bias metrics for serostatus, prevalence, and risk factor estimates.

Main Results:

• * Bayesian mixture models consistently demonstrated higher coverage and lower bias across most scenarios.
• * Cutoff methods showed acceptable performance only in specific cases of very low seroprevalence (<3%) with distinct signal differences.
• * Cutoff methods exhibited poor coverage for risk factor regression coefficients, limiting their utility in determinant analysis.

Conclusions:

• * Bayesian mixture models are recommended for robust serological data analysis, especially for emerging zoonotic diseases.
• * The study advises against using cutoff approaches for quantifying determinants of seropositivity due to inadequate coverage.
• * Findings offer empirical evidence to guide the selection of statistical methods in serological surveillance systems.