Bias correction via outcome reassignment for cross-sectional data with binary disease outcome

Area of Science:

• Epidemiology
• Biostatistics
• Observational Studies

Background:

• Cross-sectional data with binary disease outcomes are common in observational studies.
• Existing methods may introduce bias when inferring population risk due to differing outcome distributions.
• The bias in cross-sectional data analysis and its relation to population risk requires further understanding.

Purpose of the Study:

• To address the bias in estimating population risk using cross-sectional data.
• To propose and validate an outcome reassignment (OR) approach for biased binary outcome data.
• To develop statistical methods for analyzing cross-sectional data with the OR approach.

Main Methods:

• Proposed an outcome reassignment approach to adjust binary disease outcomes.
• Developed a sign test and a semiparametric pseudo-likelihood method for OR approach analysis.
• Utilized simulations and real-world Alzheimer's Disease data for validation.

Main Results:

• Demonstrated that commonly used age-specific risk probabilities are biased for population risk estimation.
• The outcome reassignment approach effectively corrects for bias in cross-sectional data.
• Proposed methods showed reliable performance in simulations and Alzheimer's Disease data analysis.

Conclusions:

• Cross-sectional data analysis for population risk estimation is prone to bias.
• The proposed outcome reassignment approach offers a statistically sound method to mitigate this bias.
• This work provides valuable tools for more accurate inference from observational cross-sectional studies.