Systematically missing confounders in individual participant data meta-analysis of observational cohort studies

Area of Science:

• Epidemiology
• Biostatistics
• Cardiovascular Research

Background:

• Meta-analyses of observational studies face challenges due to inconsistent confounder adjustment across cohorts.
• Existing methods either exclude valuable data or fail to fully account for confounding factors.
• This limits the precision and generalizability of findings in epidemiological research.

Purpose of the Study:

• To propose a novel bivariate random-effects meta-analysis model.
• To effectively utilize all available cohort data, irrespective of confounder adjustment completeness.
• To provide a robust method for adjusting for all potential confounders in meta-analyses.

Main Methods:

• Development of a bivariate random-effects meta-analysis model.
• Incorporation of both fully and partially adjusted effect estimates from cohorts with complete confounder data.
• Estimation of within-cohort correlation to leverage all available information.

Main Results:

• The proposed method allows for the inclusion of all cohorts while fully adjusting for confounders.
• Demonstrated application in estimating the association between fibrinogen levels and coronary heart disease (CHD) incidence.
• Utilized data from 154,012 participants across 31 cohorts for the CHD analysis.

Conclusions:

• The bivariate random-effects model offers a superior approach to meta-analysis when confounder data is heterogeneous.
• This method enhances the ability to accurately estimate exposure-disease associations using all available observational data.
• The findings provide a more reliable estimate of the link between fibrinogen and coronary heart disease risk.