Bayesian inference for asymptomatic COVID-19 infection rates

Area of Science:

• Biostatistics
• Epidemiology
• Infectious Disease

Background:

• Meta-analyses are crucial for summarizing independent studies but rely on specific assumptions.
• Violations of these assumptions can compromise the validity of meta-analytic inferences.
• Accurate estimation of COVID-19 asymptomatic infection rates is vital for public health.

Purpose of the Study:

• To evaluate the statistical justification for pooling data in meta-analyses with unmet assumptions.
• To reanalyze COVID-19 asymptomatic infection rate data using advanced Bayesian techniques.
• To provide guidance on cautious data pooling when effect sizes are heterogeneous.

Main Methods:

• Employed three Bayesian statistical methods with more general structures than traditional meta-analytic approaches.
• Reanalyzed existing review data focused on estimating the COVID-19 asymptomatic infection rate.
• Assessed the extent and nature of statistically justified data pooling.

Main Results:

• Demonstrated the utility of Bayesian methods in handling meta-analytic assumptions violations.
• Identified situations where pooling data from all studies may be statistically inappropriate due to heterogeneous effect sizes.
• Provided a quantitative assessment of justified data pooling for COVID-19 asymptomatic infection rates.

Conclusions:

• Bayesian methods provide a robust alternative for meta-analysis when standard assumptions are not met.
• Researchers should exercise caution when pooling data if true effect sizes likely originate from different sources.
• The developed methodology is applicable to various COVID-19 outcomes and other research areas.