Bayesian approach to cancer-trend analysis using age-stratified Poisson regression models

Area of Science:

• Epidemiology
• Biostatistics
• Statistical Modeling

Background:

• Traditional methods for analyzing cancer rate trends, such as Annual Percentage Change (APC), assume log-rate linearity, which may be inaccurate over longer periods.
• Average Annual Percentage Change (AAPC) offers an alternative by averaging APC over piecewise linear segments, but existing calculations have limitations.

Purpose of the Study:

• To propose and validate a novel Bayesian approach for calculating APC and AAPC from age-adjusted cancer rate data.
• To provide a more robust statistical framework for estimating cancer trend changes, particularly at joinpoints.

Main Methods:

• Utilized age-specific Poisson regression models with a log-link function to model cancer counts.
• Incorporated unknown joinpoints with slope-change mixture distributions and Dirichlet process priors for nonparametric age-specific intercepts.
• Developed a Bayesian framework for constructing credible intervals for AAPC, improving upon frequentist methods that condition on joinpoint locations.

Main Results:

• The proposed Bayesian method accurately captures trend-changes in cancer rates, as demonstrated by simulation studies.
• Bayesian credible intervals for AAPC offer a more comprehensive uncertainty assessment compared to traditional frequentist approaches.
• The method was successfully illustrated using prostate cancer incidence data.

Conclusions:

• The novel Bayesian approach provides a significant advancement for calculating APC and AAPC in cancer surveillance.
• This method offers improved accuracy and more reliable uncertainty quantification for epidemiological trend analysis.
• The application to prostate cancer incidence highlights its practical utility in public health research.