Cross-NMA/NMR
Cross-NMA/NMR is a method that combines different types of medical studies—like randomized trials and observational studies—and uses both detailed individual data and summary data to compare treatments. It also accounts for study quality and bias, and can show how treatment effects vary by patient characteristics like age.
At a glance
Use when
When synthesizing heterogeneous evidence from multiple study designs and data formats; when risk of bias varies across studies; when exploring effect modification by patient characteristics
Avoid when
When data on risk of bias or participant-level covariates are unavailable; when computational resources or statistical expertise are limited; when simplicity and transparency are prioritized over comprehensiveness
Inputs
Randomized and non-randomized studies, individual participant data (IPD), aggregate data (AD), risk of bias assessments, participant-level covariates
Outputs
Bias-adjusted relative treatment effects, estimates of treatment-covariate interactions, individualized treatment effect predictions, network-level summaries of efficacy and safety
How it works
Cross-NMA/NMR refers to a suite of Bayesian network meta-analysis (NMA) and network meta-regression (NMR) models that enable synthesis of cross-design evidence (randomized and non-randomized studies) and cross-format data (individual participant data and aggregate data). It uses a three-level hierarchical model to integrate IPD and AD, and includes four approaches to account for differences in study design and risk of bias (RoB). These include models that ignore RoB differences, use NRS to inform penalized priors, or apply bias-adjustment methods that downweight high RoB studies. The method allows for estimation of relative treatment effects adjusted for bias and effect modification by participant-level covariates via meta-regression.
- Project
- HTx
- Funding
- Horizon 2020
- Project status
- Completed 2024
- HTA domains
- Clinical Effectiveness
- Technology
- Non-specific
- Assumptions
- Differences in risk of bias can be modeled and adjusted for; IPD and AD are compatible; treatment effects can be modified by participant-level covariates; exchangeability of study effects holds within design and bias strata
- Strengths
- Enables inclusion of all relevant evidence regardless of design or data format; accounts for risk of bias in both RCTs and NRS; allows for personalized treatment effect estimation via meta-regression; uses Bayesian framework to incorporate prior knowledge and uncertainty
- Limitations
- Requires detailed data and risk of bias assessments; computationally intensive; model complexity may hinder interpretability; results depend on prior specifications and model assumptions
- Also known as
- Cross-design NMA, Cross-format NMA, Bayesian NMA/NMR with bias adjustment, Integrated IPD-AD NMR
Questions this answers
- › How can I combine randomized and observational studies in a treatment comparison?
- › How do I account for bias in different types of studies when comparing treatments?
- › Can I use both individual patient data and summary data together in one analysis?
- › How does treatment effectiveness change with patient age or other characteristics?
- › What happens to treatment estimates when we adjust for study quality?
- › How can I make treatment comparisons more comprehensive by including all available evidence?
References & sources
Related methods
- Alternativecrossnma R Package
- RelatedTwo-stage network meta-regression for heterogeneous treatment effects
- RelatedThree-stage network meta-regression for heterogeneous treatment effects
- Builds onConducting Indirect-Treatment-Comparison and Network-Meta-Analysis Studies
- AlternativeMultilevel Network Meta-Regression (ML-NMR)
- Used withBUGSnet
- RelatedCINeMA
Similar by meaning
Beta record. Generated from the primary source via AI extraction and independent audit, pending final human review.