Incorporating external evidence on between-trial heterogeneity in network meta-analysis

Area of Science:

• Biostatistics
• Medical Informatics
• Evidence-Based Medicine

Background:

• Network meta-analysis (NMA) often suffers from imprecise estimation of between-study heterogeneity variances due to sparse data.
• This imprecision leads to unstable standard errors for treatment differences, impacting the reliability of NMA results.
• External evidence can be leveraged to provide informative prior distributions for heterogeneity, thereby enhancing NMA inferences.

Purpose of the Study:

• To explore and present methods for specifying informative priors for multiple heterogeneity variances within NMA.
• To address the challenge of incorporating external evidence for heterogeneity in both homogeneous and heterogeneous NMA models.
• To improve the precision and stability of treatment effect estimates in NMA, particularly in data-sparse situations.

Main Methods:

• Approach 1: Assumes equal heterogeneity variances across all pairwise comparisons, allowing straightforward incorporation of a common prior.
• Approaches 2-4: Handle unequal heterogeneity variances, ensuring valid variance-covariance matrices.
• Approach 2: Uses different priors per comparison type, assuming proportionality across types and equality within types.
• Approach 3: Specifies separate priors for variances and correlations for each heterogeneity variance.
• Approach 4: Employs an informative inverse Wishart distribution for multiple unequal heterogeneity variances.
• Methods are exemplified using two NMA applications with priors derived from published evidence-based distributions.

Main Results:

• Demonstrated successful incorporation of relevant prior information on between-study heterogeneity into NMAs.
• Showcased methods that do not require assuming equal heterogeneity across all treatment comparisons.
• The proposed approaches enhance NMAs with sparse data, yielding more appropriate intervals for treatment differences compared to methods relying on imprecise heterogeneity estimates.

Conclusions:

• Informative priors derived from external evidence can significantly improve network meta-analysis, especially in sparse data settings.
• The presented strategies offer flexible ways to incorporate heterogeneity information, accommodating both common and varying heterogeneity across comparisons.
• These methods lead to more robust and reliable estimates of treatment effects in complex NMA scenarios.