Exact inference for disease prevalence based on a test with unknown specificity and sensitivity.

Area of Science:

• Epidemiology
• Biostatistics
• Public Health

Background:

• Accurate COVID-19 prevalence estimation is crucial for public policy.
• Challenges include imperfect test accuracy (sensitivity/specificity) and low initial disease prevalence.
• Traditional statistical methods may fail with small sample sizes or low prevalence.

Purpose of the Study:

• To develop robust statistical methods for estimating disease prevalence.
• To provide valid confidence intervals for prevalence estimation, especially in low-prevalence settings.
• To improve the reliability of seroprevalence studies.

Main Methods:

• Proposed novel confidence intervals for disease prevalence estimation.
• Developed methods applicable to both unweighted and weighted data settings.
• Utilized hybrid bootstrap methods for improved inference in weighted settings.

Main Results:

• The proposed confidence intervals are valid regardless of sample size in the unweighted setting.
• Hybrid bootstrap methods demonstrate robust performance, outperforming asymptotic approximations.
• Reanalyzed existing seroprevalence data, including a Santa Clara County antibody study.

Conclusions:

• The new statistical methods offer more reliable disease prevalence estimates.
• These methods address limitations of traditional approaches in challenging epidemiological scenarios.
• Improved prevalence estimation supports better-informed public health policy decisions.