Meta-analysis of continuous outcome data from individual patients

Area of Science:

• Biostatistics
• Clinical Trials
• Epidemiology

Background:

• Individual patient data (IPD) meta-analyses are increasingly prevalent, offering advantages over traditional summary statistics meta-analyses.
• Multilevel or hierarchical models are suitable for analyzing continuous individual patient outcome data from clinical trials.

Purpose of the Study:

• To develop a general framework for IPD meta-analysis using multilevel models.
• To explore the investigation of heterogeneity using patient-level covariates and meta-regression.
• To compare different statistical software packages for implementing these methods.

Main Methods:

• Utilized multilevel/hierarchical models for continuous IPD meta-analysis.
• Developed a general framework encompassing traditional meta-analysis, meta-regression, and patient-level covariate inclusion.
• Considered unexplained variation in treatment differences as random effects.
• Illustrated methods using fixed trial effects models with an extension to random trial effects.

Main Results:

• Demonstrated the application of multilevel models for IPD meta-analysis in an Alzheimer's disease example.
• Compared the use of SAS PROC MIXED, MLwiN (classical framework), and BUGS (Bayesian framework).
• Discussed the relative merits of the software packages and model assumption assessment.

Conclusions:

• Multilevel models offer a robust and flexible approach for individual patient data meta-analysis.
• The developed framework facilitates the investigation of heterogeneity and inclusion of covariates.
• Software choice depends on specific analytical needs and user expertise.