A conditional approach for inference in multivariate age-period-cohort models.

Area of Science:

• Epidemiology
• Biostatistics
• Public Health

Background:

• Age-period-cohort (APC) models analyze disease data by age and time.
• Multivariate APC (MAPC) models extend this to analyze heterogeneous trends across strata (e.g., geography).
• Existing MAPC models share parameters (e.g., age effects) while allowing others to vary by stratum.

Purpose of the Study:

• To propose a conditional approach for direct modeling of relative time trends.
• To demonstrate the conditional approach's ability to handle unmeasured confounding for more precise relative risk estimation.
• To extend the methodology for data with multiple stratification levels.

Main Methods:

• Development of a conditional inference approach for relative time trends.
• Application of maximum likelihood estimation using multinomial logistic regression software.
• Suggestion of smoothing splines for stabilizing relative time trend estimates.
• Validation using chronic obstructive pulmonary disease mortality data in England & Wales, stratified by area and gender.

Main Results:

• The conditional approach allows direct modeling of relative time trends.
• Potential for improved precision in relative risk estimation due to handling of unmeasured confounding.
• Successful application to complex, multi-stratified disease mortality data.

Conclusions:

• The proposed conditional approach offers a valuable alternative for analyzing stratified epidemiological data.
• This method enhances the understanding of disease trends across different population subgroups.
• The approach is robust and applicable to real-world public health challenges.