Differential Misclassification of Disease under Partial-Mouth Sampling.

Area of Science:

• Biostatistics
• Epidemiology
• Dental Public Health

Background:

• Partial-cluster sampling, common in periodontal research, involves examining a subset of observations (e.g., tooth sites) within a cluster (e.g., full mouth).
• Accurate disease classification is crucial for valid epidemiological studies, especially when assessing associations with exposures.
• Previous studies have not fully addressed the impact of partial-mouth sampling on disease misclassification and subsequent exposure association bias.

Purpose of the Study:

• To investigate the effect of cluster-level disease misclassification, arising from partial-cluster sampling, on the association with a cluster-level dichotomous exposure.
• To analyze how partial-mouth sampling influences disease misclassification probabilities, including sensitivity and negative predictive values.
• To evaluate the potential for bias in disease-exposure associations estimated using partial-recording protocols compared to full-cluster sampling.

Main Methods:

• Utilized conditional probability arguments to model disease misclassification under partial-cluster sampling.
• Investigated the impact of misclassification on the estimated association between a cluster-level dichotomous outcome (disease) and a dichotomous exposure.
• Employed a numeric example to demonstrate bias in odds ratios under partial-cluster random sampling versus full-cluster sampling.

Main Results:

• Disease misclassification at the cluster level is generally differential under partial-cluster sampling when disease probability varies by exposure status.
• The degree of misclassification depends on the sampling protocol, joint probability structure of the condition within clusters, and the disease criterion.
• Disease-exposure odds ratios can be biased in either direction (toward or away from the null) compared to gold-standard estimates from full-cluster sampling.

Conclusions:

• Partial-mouth sampling leads to differential misclassification of disease, where sensitivity and negative predictive values are exposure-dependent.
• This differential misclassification results in biased inference for disease-exposure associations.
• Standard analysis procedures may under- or overestimate associations in periodontal data; methods addressing bias in partial-recording protocols are necessary.