Linear mixed models for investigating effect modification in subgroup meta-analysis.

Area of Science:

• Biostatistics
• Clinical Epidemiology
• Medical Research Methodology

Background:

• Subgroup meta-analysis compares treatment effects across patient subgroups to identify differential treatment effects, potentially enabling personalized medicine.
• Existing aggregate data subgroup meta-analysis methods have limitations in flexibility and handling complex data structures.
• Individual participant data meta-analysis is often preferred but resource-intensive and may not always be feasible.

Approach:

• Proposes linear mixed models (LMMs) as a robust method for aggregate data subgroup meta-analysis.
• LMMs extend current methods by offering greater flexibility in modeling heterogeneity and accommodating studies with missing subgroup information.
• Compares LMMs against existing methods using simulations and two case studies to evaluate performance.

Key Points:

• Linear mixed models provide a more adaptable framework for aggregate data subgroup meta-analysis compared to traditional methods.
• The LMM approach effectively handles heterogeneity and missing subgroup data, which are common challenges in meta-analyses.
• Simulation and case study results validate the advantages of LMMs for exploring treatment effect modification in aggregate data.

Conclusions:

• Linear mixed models represent an advantageous advancement for aggregate data subgroup meta-analysis.
• This approach facilitates a more nuanced understanding of treatment effect modification across subgroups.
• The findings support the use of LMMs for preliminary exploration of treatment effect modifiers before undertaking individual participant data meta-analysis.