Simulation Study Compares DBTI and Loop Models in Network Meta-Analysis

Researchers have published a simulation study comparing two frequentist models used to assess global inconsistency in network meta-analysis, specifically the design-by-treatment interaction (DBTI) model and the Loop model. The study, published in BMC Medical Research Methodology, examines the theoretical relationship between these two parameterizations and evaluates their performance across various network configurations and inconsistency-generating mechanisms.

The team characterized the relationship between the DBTI and Loop models by matching their design-specific mean effects. A key metric in this comparison was the difference in degrees of freedom, defined as Δdf, which equals the degrees of freedom of the DBTI model minus those of the Loop model. Simulation scenarios were constructed with varying values of Δdf in networks with three, five, and eight treatments.

The simulation design involved generating binary data under two distinct inconsistency specifications: DBTI-based and Loop-based. Both models were fitted using restricted maximum likelihood. Global inconsistency was assessed using joint Wald tests. The performance of each model was evaluated through three primary metrics: empirical size, power, and bias in heterogeneity variance estimation.

The results indicated that when Δdf equals zero, the two models are equivalent up to reparameterization under the same covariance assumptions. In these scenarios, the DBTI and Loop models produced nearly identical results. However, when Δdf is greater than zero, the Loop model imposes additional constraints on the inconsistency patterns that can be represented by the DBTI model.

Under the Loop-based generating mechanism, power differences were generally small. There was no consistent advantage for the Loop model.

In contrast, under the selected DBTI-based configurations, the DBTI test demonstrated higher rejection rates compared to the Loop model. However, the Loop model exhibited upward bias in heterogeneity variance estimates at larger inconsistency values.

The study also noted that empirical size was elevated in some scenarios, which complicates direct comparisons of power between the two models.

The conclusions of the study emphasize that having fewer degrees of freedom for inconsistency did not consistently yield higher empirical power in the simulated networks. This finding underscores the importance of considering both model complexity and the specific inconsistency structures each model can represent when selecting a global inconsistency test. The choice between DBTI and Loop models should not be based solely on degrees of freedom but should also take into account the network configuration and heterogeneity estimates.

Subject of Research: Simulation comparison of DBTI and Loop models for global inconsistency testing in network meta-analysis

Article Title: Degrees of freedom and inconsistency-generating mechanisms in global inconsistency testing: a simulation comparison of DBTI and Loop models in NMA

Article References: Liang, Q., Zhang, Z., Guo, W., Qin, L., & Yang, K. (2026). Degrees of freedom and inconsistency-generating mechanisms in global inconsistency testing: a simulation comparison of DBTI and Loop models in NMA. BMC Medical Research Methodology. https://doi.org/10.1186/s12874-026-03031-8

Image Credits: AI Generated

DOI: 10.1186/s12874-026-03031-8

Keywords: network meta-analysis, inconsistency testing, DBTI model, Loop model, degrees of freedom, simulation study, statistical methods, BMC Medical Research Methodology, Degrees, freedom, inconsistency-generating, mechanisms