Node-splitting models for inconsistency assessment
This method helps detect inconsistencies in network meta-analyses by automatically comparing treatment effects from direct and indirect evidence, making it easier to identify potential biases or conflicts in the data.
At a glance
Use when
Assessing the robustness of network meta-analysis results by identifying local inconsistencies in treatment effect estimates.
Avoid when
When the network has very few direct comparisons or when all evidence is indirect for a given contrast.
Inputs
Network of randomized clinical trials with multi-treatment comparisons, typically structured as a graph of treatments and studies.
Outputs
Estimates of inconsistency for each split comparison, contrasting direct and indirect effect estimates, along with measures of statistical uncertainty.
How it works
The node-splitting approach assesses inconsistency in network meta-analysis by separating direct and indirect evidence for specific treatment comparisons. This work introduces an automated method with an unambiguous decision rule to select which comparisons to split, ensuring that only comparisons within potentially inconsistent loops are evaluated. The rule avoids complications from multi-arm trials and allows for systematic, efficient model generation across all relevant comparisons.
- HTA domains
- Clinical Effectiveness
- Categories
- HeterogeneityIndirect Comparisons
- Assumptions
- The underlying network structure allows separation of direct and indirect evidence; multi-arm trials are correctly parameterized; inconsistency, if present, is localized to specific loops.
- Strengths
- Provides a clear, interpretable contrast between direct and indirect evidence,Automates model generation, reducing manual effort,Systematically covers all potentially inconsistent loops
- Limitations
- Relies on sufficient direct evidence for split comparisons,May have low power to detect inconsistency in sparse networks,Assumes consistency in the rest of the network when splitting a single comparison
- Also known as
- Node-splitting method, Automated node-splitting
Questions this answers
References & sources
Related methods
Similar by meaning
Beta record. Generated from the primary source via AI extraction and independent audit, pending final human review.