Complexity and bias in cross-sectional data with binary disease outcome in observational studies

Area of Science:

• Epidemiology
• Biostatistics
• Population Health

Background:

• Cross-sectional studies are common for analyzing binary disease outcomes in observational research.
• Existing understanding suggests cross-sectional data is less informative than longitudinal data for disease progression.
• There is a lack of clarity on potential biases in cross-sectional data and their relation to population risk.

Purpose of the Study:

• To investigate the connection between binary disease outcomes in cross-sectional populations and the underlying birth-illness-death process.
• To determine if and how bias arises in cross-sectional data when inferring population risk.
• To propose and illustrate a method for analyzing population risk using age-specific risk probability (ARP).

Main Methods:

• Conceptualizing disease progression as a birth-illness-death process.
• Deriving explicit formulas to link cross-sectional outcome distribution to population risk.
• Developing and applying the age-specific risk probability (ARP) model.
• Illustrating the ARP model with Alzheimer's disease data.

Main Results:

• The distribution of cross-sectional binary outcomes differs significantly from the true population risk distribution.
• Bias is almost always present when using cross-sectional data to estimate population risk.
• The cross-sectional risk probability is influenced by both population risk and the relative durations of disease states.

Conclusions:

• Inference on population risk from cross-sectional data is inherently biased.
• The proposed age-specific risk probability (ARP) model offers a potentially compromised, yet still biased, approach to understanding population risk.
• Careful consideration and critique are necessary when interpreting results from cross-sectional studies, even when using methods like ARP.