Related Experiment Videos
Inference for case-control studies when exposure status is both informatively missing and misclassified.
Robert H Lyles1, Andrew S Allen, W Dana Flanders
1Department of Biostatistics, The Rollins School of Public Health of Emory University, Atlanta, GA 30322, USA. rlyles@sph.emory.edu
Statistics in Medicine
|February 8, 2006
Summary
This study addresses bias in case-control studies caused by exposure misclassification and missing data. Combining these issues can inflate odds ratio estimates, even with non-differential misclassification.
Area of Science:
- Epidemiology
- Biostatistics
Background:
- Case-control studies frequently encounter misclassified categorical exposure variables.
- Exposure status is often informatively missing, meaning missingness probability relates to the actual exposure.
Purpose of the Study:
- To introduce novel study designs and analytic procedures for simultaneously addressing exposure misclassification and informative missingness in 2x2 analyses.
- To quantify the combined impact of these biases on exposure odds ratio estimates.
Main Methods:
- Development of integrated methods to handle both misclassification and missing data.
- Utilizing validation data for misclassification correction and supplemental sampling for missing data.
- Analysis based on convergence in probability.
Main Results:
- The combined effects of missingness and misclassification can significantly bias naive odds ratio estimates.
- Bias can lead to inflated estimates or estimates on the wrong side of the null, even with non-differential misclassification.
- Demonstrated impact using a case-control study of low birth weight and breast cancer risk.
Conclusions:
- Naïve analyses in case-control studies are unreliable when facing both exposure misclassification and informative missingness.
- The proposed methods offer a robust approach to mitigate combined biases.
- Accurate estimation of exposure-disease associations requires addressing both data quality issues concurrently.