Evidence synthesis for count distributions based on heterogeneous and incomplete aggregated data

Area of Science:

• Biostatistics
• Epidemiology
• Medical Data Analysis

Background:

• Poisson models are standard for count data analysis.
• Overdispersed data, where variance exceeds the mean, requires advanced modeling.
• Heterogeneous data reporting (e.g., counts, proportions, rates) complicates evidence synthesis.

Purpose of the Study:

• To develop a joint modeling framework for synthesizing count data with heterogeneous reporting formats.
• To address the challenges of analyzing overdispersed data in evidence synthesis.
• To enable coherent statistical inference from diverse data sources.

Main Methods:

• Utilizing negative binomial models as a generalization of Poisson models for overdispersion.
• Developing a joint modeling strategy to integrate different data types (mean counts, proportions, rates).
• Applying the methods to a systematic review of chronic obstructive pulmonary disease (COPD) data.

Main Results:

• Demonstrated the feasibility of joint modeling for heterogeneous count data.
• Showcased how integrating diverse data sources enhances statistical inference.
• Provided a robust method for evidence synthesis in the presence of overdispersion and varied reporting.

Conclusions:

• Joint modeling offers a powerful approach for evidence synthesis with heterogeneous count data.
• Negative binomial models effectively handle overdispersion in statistical analyses.
• The proposed methods are applicable to various medical research areas, including COPD studies.