INFORMATION-THRESHOLDED NETWORK META-ANALYSIS: FEASIBILITY AND IMPACT IN A MATURE EVIDENCE NETWORK
Author(s)
Grace G. Goldsmith-Martin, BSc, PhD1, Kathleen Bowes, PhD1, Rachel Kewley, BSc1, Xin Zhao, MS2, Teuta Gjuladin-Hellon, PhD1.
1Genesis Research Group, London, United Kingdom, 2Genesis Research Group, Hoboken, NJ, USA.
1Genesis Research Group, London, United Kingdom, 2Genesis Research Group, Hoboken, NJ, USA.
OBJECTIVES: Network meta-analysis (NMA) is an established approach for estimating relative treatment effects across multiple interventions and studies, but inclusion of small studies may disproportionately affect treatment effect (TE) estimates and uncertainty. Therefore, this study evaluated how excluding smaller studies affects overall robustness of NMA findings while preserving network connectivity.
METHODS: A published NMA of LDL-lowering agents was reanalyzed (Toth 2022); selection was based on the number of trials included, established network geometry, and comprehensive data availability. For all studies, floored base-10 logarithms of sample size were calculated and studies were classified “large” if this value matched the largest study within each direct comparison, to maintain network geometry. Two random-effects NMAs were conducted (netmeta [Balduzzi 2023]): the reconstructed original network and a restricted large-study-only network.
RESULTS: In the restricted NMA some direct comparisons were informed by a single study. TE estimates for LDL-lowering agents were broadly similar between the reconstructed original (48 studies) and restricted (24 studies) networks. The magnitude of TE estimate percentage change between the original and restricted networks ranged from 2.2% to 12.9% across all comparisons, with most treatments (75%) remaining within 10% of the original TE. Point estimate rankings of the top four agents changed after exclusion of small studies, while less effective agents remained consistently ranked at the bottom. Confidence intervals widened for all treatments. High heterogeneity persisted in both networks (I² = 85.2% [95% CI: 81.3%, 88.2%] for reconstructed NMA and I² = 82.4% [95% CI: 73.8%, 88.1%] for restricted NMA).
CONCLUSIONS: Excluding small studies in information-thresholded NMA increased uncertainty and impacted point-estimate ranking among the most effective agents, while less effective treatments remained consistently ranked. Small-study exclusion did not resolve heterogeneity. Information-thresholded sensitivity analyses may help assess the robustness of findings in mature evidence networks, but further research is warranted.
METHODS: A published NMA of LDL-lowering agents was reanalyzed (Toth 2022); selection was based on the number of trials included, established network geometry, and comprehensive data availability. For all studies, floored base-10 logarithms of sample size were calculated and studies were classified “large” if this value matched the largest study within each direct comparison, to maintain network geometry. Two random-effects NMAs were conducted (netmeta [Balduzzi 2023]): the reconstructed original network and a restricted large-study-only network.
RESULTS: In the restricted NMA some direct comparisons were informed by a single study. TE estimates for LDL-lowering agents were broadly similar between the reconstructed original (48 studies) and restricted (24 studies) networks. The magnitude of TE estimate percentage change between the original and restricted networks ranged from 2.2% to 12.9% across all comparisons, with most treatments (75%) remaining within 10% of the original TE. Point estimate rankings of the top four agents changed after exclusion of small studies, while less effective agents remained consistently ranked at the bottom. Confidence intervals widened for all treatments. High heterogeneity persisted in both networks (I² = 85.2% [95% CI: 81.3%, 88.2%] for reconstructed NMA and I² = 82.4% [95% CI: 73.8%, 88.1%] for restricted NMA).
CONCLUSIONS: Excluding small studies in information-thresholded NMA increased uncertainty and impacted point-estimate ranking among the most effective agents, while less effective treatments remained consistently ranked. Small-study exclusion did not resolve heterogeneity. Information-thresholded sensitivity analyses may help assess the robustness of findings in mature evidence networks, but further research is warranted.
Conference/Value in Health Info
2026-11, ISPOR Europe 2026, Vienna, Austria
Value in Health, Volume 29, Issue 12S
Code
SA65
Topic
Study Approaches
Topic Subcategory
Literature Review & Synthesis, Meta-Analysis & Indirect Comparisons
Disease
Cardiovascular Disorders (including MI, Stroke, Circulatory)