Statistical approaches to adjusting weights for dependent arms in network meta-analysis

Area of Science:

• Biostatistics
• Clinical Trial Design
• Evidence Synthesis

Background:

• Network meta-analysis (NMA) synthesizes evidence from multiple randomized controlled trials (RCTs).
• Standard NMA assumes independence between treatment arms, which is violated in within-person trial designs (e.g., split-mouth, crossover).
• Dependent treatment arms in these designs require specialized statistical handling to avoid biased conclusions.

Purpose of the Study:

• To develop and demonstrate statistical approaches for adjusting weights of dependent arms in within-person trial designs for NMA.
• To provide methods applicable in standard statistical software (R, STATA).

Main Methods:

• The study presents three distinct approaches: data augmentation, adjusting variance, and reducing weight.
• These methods are designed to account for correlations between dependent treatment arms.
• A case study on periodontal regeneration illustrates the implementation and comparison of these approaches.

Main Results:

• The adjusting variance approach is compatible with existing STATA network packages.
• The reducing weight approach necessitates custom programming for the within-study variance-covariance matrix.
• All three methods offer valid ways to incorporate dependent arm data into NMA.

Conclusions:

• Accurate statistical adjustment for dependent arms is crucial for reliable NMA results from within-person trials.
• The presented methods offer practical solutions for researchers using specialized trial designs.
• These approaches enhance the validity and applicability of network meta-analysis in complex clinical research settings.