Effects of cluster sampling on epidemiologic analysis in population-based case-control studies

Area of Science:

• Epidemiology
• Biostatistics
• Survey Methodology

Background:

• Population-based case-control studies are crucial for epidemiological research.
• Cluster sampling is often used in large-scale surveys but can introduce bias.
• Classical statistical procedures may not adequately account for cluster sampling effects.

Purpose of the Study:

• To investigate the impact of cluster sampling on standard epidemiologic procedures.
• To develop and present modified statistical methods for cluster-based case-control studies.
• To ensure accurate estimation and testing of odds ratios in the presence of cluster sampling.

Main Methods:

• Consideration of three cluster sampling plans for control selection in population-based case-control designs.
• Development of modified procedures for testing homogeneity of odds ratios across strata.
• Presentation of modified methods for estimating and testing a common odds ratio.
• Simulation studies using data from the 1970 Health Interview Survey and a mixed multinomial model.

Main Results:

• Classical procedures may be robust under mild cluster sampling but can fail in extreme cases.
• The Mantel-Haenszel and Woolf-Haldane tests may exhibit inflated Type I error rates (sizes) and reduced confidence interval coverage under cluster sampling.
• Classical odds ratio estimates can be biased with non-self-weighting cluster samples.

Conclusions:

• Cluster sampling significantly impacts the validity of classical epidemiologic procedures.
• The proposed modified procedures effectively address the biases introduced by cluster sampling.
• These modified methods ensure more reliable odds ratio estimation and hypothesis testing in cluster-based studies.