(133) A novel estimation method for adjusting virtual inconsistency in a star-shaped network meta-analysis: an illustration of a case study

Background: In a star-shaped network, comparative evaluations among treatments relying solely on indirect evidence were based on an unidentifiable consistency assumption, limiting the reliability of the results. A method has been devised to assess the robustness of results from a star-shaped network meta-analysis (NMA). However, the method, being established through the numerical procedures, could require the lengthy computation time.

Objectives: To explain the existing analysis method for evaluating the robustness of results in a star-shaped NMA, propose an estimation approach to reduce the running time, and illustrate the method’s usability.

Methods:

Design: Method development and illustration
Setting: The existing estimation method consists of two main steps: (1) analyzing each imputed network using the NMA model; (2) pooling the results. An illustrative example was of a NMA assessing the benefits and harms of Dipeptidyl peptidase-4 inhibitors (DPP4i), glucagon-like peptide 1 receptor agonists (GPP-1RA), sodium-glucose co-transporter-2 inhibitors (SGLT2i) in patients with cardiovascular disease. Outcomes data from 17 placebo-controlled studies (6 DPP4i, 6 GLP-1RA, and 5 SGLT2i studies) were abstracted.
Exposures or Interventions: We constructed a new three-level hierarchical NMA model to simultaneously analyze data from multiple complete networks.
Main outcome measures: The outcomes were cardiovascular mortality, stroke, and myocardial infarction.
Statistical analysis: The imputed data simulated a situation that would allow both indirect and direct comparisons, with a statistically acceptable extent of inconsistency. By comparing the agreement between the results obtained from the original star-shaped NMA and the results after incorporating the imputed data, the robustness of the results of the original star-shaped NMA was quantified and assessed. We compared the estimates produced by both approaches and their running time.

Results: In the analysis that incorporated imputed data, we observed that the treatment rankings for the cardiovascular mortality and stroke remained consistent to the original rankings derived from the star-shaped network. For the myocardial infarction, the treatment rankings differed in 75% of the results from the analysis incorporating imputed data compared to the original rankings. While the estimates of all the parameters obtained from the new three-level hierarchical model method were identical to those from the existing two-step estimation method, the running time was decreased ten-fold.

Conclusions: When applying the method for evaluating the robustness of results in a star-shaped NMA, the developed three-level hierarchical NMA model may enhance time efficiency, thereby improving the method’s usability.