Design-by-treatment interaction model
This method checks whether different types of studies in a network meta-analysis give conflicting results, especially when some studies compare more than two treatments at once. It helps identify inconsistencies that might bias the conclusions about which treatment works best.
At a glance
Use when
Conducting network meta-analysis with multi-arm trials where design-related bias or inconsistency is suspected.
Avoid when
When the evidence network contains only two-treatment comparisons or very few designs, making interaction modeling unstable.
Inputs
Network meta-analysis data including multi-arm trials, with information on treatment sets and pairwise comparisons.
Outputs
Estimates of inconsistency parameters, statistical fit of the model, identification of designs contributing to inconsistency.
How it works
The design-by-treatment interaction model is a statistical framework used in network meta-analysis to assess inconsistency in evidence networks, particularly arising from multi-arm trials. It distinguishes between loop inconsistency and design inconsistency by modeling interactions between study design (i.e., set of treatments compared) and treatment effects. It generalizes the Lu-Ades model and allows detection of inconsistencies not limited to closed loops, especially where multi-arm trials introduce complex dependencies.
- HTA domains
- Clinical Effectiveness
- Categories
- Indirect Comparisons
- Assumptions
- Treatment effects can vary by study design; inconsistency is explainable by design-by-treatment interaction; random or fixed effects assumptions apply within designs.
- Strengths
- Explicitly accounts for multi-arm trials; distinguishes design-related inconsistency from loop inconsistency; provides a comprehensive framework for inconsistency assessment.
- Limitations
- Requires sufficient network complexity and design variation; may be underpowered in sparse networks; complex interpretation due to increased number of parameters.
- Also known as
- Design-by-treatment interaction, DBT interaction model
Questions this answers
- › Are results from different study designs in the network meta-analysis consistent?
- › Is there evidence of bias due to differences in trial designs involving multi-arm studies?
- › How can inconsistency be modeled when multi-arm trials are present?
- › Does the choice of treatment set in a trial influence the estimated effect sizes?
- › What is the source of statistical inconsistency in a network of evidence?
- › Can inconsistency be distinguished between loop-specific and design-related causes?
References & sources
Similar by meaning
Beta record. Generated from the primary source via AI extraction and independent audit, pending final human review.