Nonparametric Bayesian Meta-Analysis Model With Change Point Detection: A Case Study on Time-Varying...

Area of Science:

• Environmental Epidemiology
• Biostatistics
• Climate Change Research

Background:

• Established U- or J-shaped associations between ambient temperature and mortality exist.
• Climate change and adaptive behaviors (e.g., air conditioning use) alter these temperature-mortality relationships over time.
• Conventional models often assume linear temporal changes, which may not accurately reflect reality.

Purpose of the Study:

• To develop a novel two-stage modeling framework to analyze time-varying temperature-mortality associations.
• To address limitations of conventional models, including assumptions of linear temporal change and sensitivity to outliers.
• To accurately capture non-linear and non-gradual temporal shifts in temperature-mortality associations.

Main Methods:

• A two-stage modeling framework was employed.
• Stage 1: Distributed lag nonlinear models estimated temperature-mortality associations for annual sub-periods in Japan.
• Stage 2: A nonparametric Bayesian meta-analysis model with change-point detection (Probit Stick-Breaking Process) pooled associations, incorporating robust estimation for outliers.

Main Results:

• Preliminary analysis of Japanese data suggested non-linear temporal changes in temperature-mortality associations.
• The proposed Bayesian meta-analysis model effectively detected temporal shifts and provided robust estimates.
• Simulation studies validated the framework's performance.

Conclusions:

• The novel nonparametric Bayesian meta-analysis model offers a more flexible and robust approach to studying time-varying temperature-mortality associations.
• This framework challenges the assumption of linear temporal changes in existing models.
• Accurate assessment of temperature-mortality dynamics is crucial for public health adaptation strategies under climate change.