Measurement error in meta-analysis (MEMA)-A Bayesian framework for continuous outcome data subject to...

Area of Science:

• Biostatistics
• Epidemiology
• Statistical Modeling

Background:

• Meta-analyses ideally use unbiased study data.
• Non-differential measurement error in exposure variables can compromise meta-analysis results.
• Accurate estimation of associations is crucial in research synthesis.

Purpose of the Study:

• To develop a statistical framework for meta-analysis when exposure variables have non-differential measurement error.
• To provide methods for obtaining reliable point and interval estimates in such scenarios.
• To explore adjustments for covariates using individual-participant data (IPD).

Main Methods:

• A flexible Bayesian meta-analysis framework is proposed.
• The model accommodates varying degrees of prior knowledge about measurement error magnitude.
• Methods are demonstrated for continuous outcomes and continuous exposure variables using regression coefficients.

Main Results:

• The Bayesian framework yields appropriate point and interval estimates despite measurement error.
• The model's performance is evaluated across different levels of prior information on measurement error.
• Demonstration of adjusting for covariates using IPD in the meta-analysis model.

Conclusions:

• The developed Bayesian approach effectively handles non-differential measurement error in meta-analyses.
• This method enhances the accuracy of estimated associations in research synthesis.
• The framework offers flexibility and can incorporate individual-participant data for robust analysis.