Quantifying previous SARS-CoV-2 infection through mixture modelling of antibody levels

Area of Science:

• Epidemiology
• Immunology
• Biostatistics

Background:

• Accurate estimation of the proportion of the population previously infected with SARS-CoV-2 (COVID-19) is vital for informing vaccination strategies and easing movement restrictions.
• Current methods rely on serosurvey data, involving a two-step process of calculating the proportion above an antibody threshold and adjusting with external sensitivity and specificity estimates.
• A key limitation is the potential non-representativeness of PCR-confirmed cases used for sensitivity estimation, possibly leading to biased results.

Purpose of the Study:

• To compare the standard threshold-based approach with a mixture modeling approach for estimating SARS-CoV-2 previous infection prevalence.
• To illustrate the potential bias in the standard method using real-world serosurvey data from Kenya.

Main Methods:

• The study employed mixture modeling as an alternative to the traditional threshold-based method for analyzing serosurvey data.
• This approach bypasses the need for external data from PCR-confirmed cases to estimate test sensitivity and specificity.
• Data from multiple Kenyan serosurveys were utilized for comparative analysis.

Main Results:

• The mixture modeling analysis yielded estimates of previous SARS-CoV-2 infection that were frequently and substantially higher than those obtained using the standard threshold analysis.
• This highlights a significant bias inherent in the conventional method when applied to the studied population.

Conclusions:

• The findings suggest that the standard threshold-based method may underestimate the true prevalence of previous SARS-CoV-2 infection.
• Mixture modeling offers a more robust and potentially more accurate alternative for estimating population-level infection burden, crucial for effective public health decision-making.