Accounting for misclassification bias of binary outcomes due to underscreening: a sensitivity analysis

Area of Science:

• Epidemiology
• Biostatistics
• Medical Informatics

Background:

• Diagnostic testing is often limited to high-risk groups, leading to undiagnosed cases.
• This selective testing complicates accurate disease prevalence estimation and risk factor analysis.
• Missing disease status data presents a significant challenge in population health studies.

Purpose of the Study:

• To develop a statistical model for disease prevalence and risk factor estimation in populations with incomplete diagnostic testing.
• To address the challenge of missing outcome data due to selective testing.
• To provide a robust method for epidemiological studies with non-randomized testing.

Main Methods:

• Formulated the problem as a missing data issue.
• Proposed a Bayesian selection model to jointly analyze disease outcome and testing status.
• Conducted sensitivity analyses to evaluate the impact of the sensitivity parameter on results.

Main Results:

• Applied the model to a retrospective cohort of pediatric asthma exacerbation patients evaluated for pneumonia.
• Identified female gender, fever, and severe hypoxia as significant predictors of radiographic pneumonia.
• Simulation studies confirmed model performance even with low disease prevalence and screening proportions.

Conclusions:

• The Bayesian selection model offers a viable approach for estimating disease prevalence.
• This model is effective for studying disease risk factors when testing is not universally applied.
• The methodology provides a valuable tool for epidemiological research with selective diagnostic testing.