Estimation of design effects in cluster surveys

Area of Science:

• Epidemiology
• Biostatistics

Background:

• Cluster sampling can inflate disease prevalence estimates compared to simple random sampling.
• This inflation, known as the design effect, is influenced by disease prevalence, cluster size, and within-cluster disease association.
• Prior survey design effects may not apply to new surveys if these factors differ.

Purpose of the Study:

• To evaluate pairwise odds ratios as a portable measure of within-cluster disease association.
• To compare the utility of pairwise odds ratios with traditional design effects for epidemiological surveys.
• To assess factors influencing design effects in cluster sampling for disease prevalence.

Main Methods:

• Estimated within-village pairwise odds ratios for fever and cough from four studies in Africa and Asia.
• Calculated design effects for fever and cough prevalence estimates.
• Analyzed the impact of cluster size, odds ratio, and cluster size variation on design effects.

Main Results:

• Pairwise odds ratios for fever ranged from 1.04 to 1.34; for cough, they ranged from 1.03 to 1.24.
• Design effects for fever ranged from 2.35 to 6.80; for cough, they ranged from 1.99 to 7.39.
• Design effects were more sensitive to cluster size and odds ratio magnitude than to cluster size variation.

Conclusions:

• Pairwise odds ratios provide a more adaptable metric for within-cluster disease association than design effects.
• Understanding factors influencing design effects is crucial for accurate disease prevalence estimation in cluster surveys.
• These findings aid in planning more robust epidemiological studies using cluster sampling.