Related Experiment Videos
On estimation of vaccine efficacy using validation samples with selection bias
Daniel O Scharfstein1, M Elizabeth Halloran, Haitao Chu
1Department of Biostatistics, Johns Hopkins Bloomberg School of Public Health, Baltimore, MD 21205, USA. dscharf@jhsph.edu
Biostatistics (Oxford, England)
|March 25, 2006
Summary
Estimating vaccine efficacy (VE) using validation sets can be biased if these sets are not randomly selected. New frequentist and Bayesian methods address this validation bias, providing more accurate VE estimates, especially for influenza vaccines.
Area of Science:
- Epidemiology
- Biostatistics
Background:
- Validation sets are crucial for improving vaccine efficacy (VE) estimations.
- Standard methods assume outcomes are missing at random (MAR), which is often not true in practice.
- Non-random validation set selection can lead to biased VE estimates.
Purpose of the Study:
- To develop novel frequentist and Bayesian statistical approaches for estimating VE in the presence of validation bias.
- To address the limitations of existing methods that rely on the MAR assumption.
- To provide a framework for studies with missing binary outcomes and categorical covariates.
Main Methods:
- Proposed frequentist and Bayesian methods to account for non-random validation set selection.
- Incorporated expert opinion on the nature and extent of validation selection bias.
- Re-analyzed an influenza vaccine study using the proposed methods.
Main Results:
- The developed methods effectively estimate VE even with validation bias.
- In an influenza vaccine study, estimates using validation sets were significantly higher than those using only non-specific case definitions, within plausible bias ranges.
- The approach is robust to varying degrees of selection bias.
Conclusions:
- The proposed frequentist and Bayesian methods offer a robust solution for estimating VE with non-random validation sets.
- Accounting for validation bias is essential for accurate VE assessment, particularly in infectious disease surveillance.
- These methods enhance the reliability of VE studies with missing outcome data.