Analysis of multicentre epidemiological studies: contrasting fixed or random effects modelling and meta-analysis

Area of Science:

• Epidemiology
• Biostatistics

Background:

• Multicentre studies enhance statistical power in epidemiological research by increasing participant numbers.
• Analyzing multicentre data presents challenges, particularly in controlling for confounding by study centre.

Purpose of the Study:

• To compare and contrast three common methods for analysing multicentre epidemiological data: fixed effects, random effects, and meta-analysis.
• To provide guidance on selecting the most appropriate modelling approach based on study characteristics.
• To highlight the importance of distinguishing within-centre from between-centre associations and assessing effect heterogeneity.

Main Methods:

• Comparison of three statistical modelling approaches: pooling data with fixed effects, random effects modelling, and meta-analysis of centre-specific models.
• Illustration using a real-world epidemiological study and a synthetic dataset.
• Provision of R and Stata code for reproducibility.

Main Results:

• Different modelling approaches can yield divergent conclusions in multicentre studies.
• The choice of model significantly impacts the interpretation of exposure-associated risks and confounder effects.
• Key analytical considerations include differentiating within-centre and between-centre effects and assessing heterogeneity.

Conclusions:

• The selection of an appropriate statistical model is crucial for valid risk factor identification in multicentre epidemiological studies.
• Understanding the nuances between fixed effects, random effects, and meta-analysis is essential for researchers.
• Proper handling of centre-specific effects and heterogeneity ensures robust epidemiological findings.