Empirical evaluation of SUCRA-based treatment ranks in network meta-analysis: quantifying robustness using Cohen’s kappa
Bibliographic record
Abstract
OBJECTIVE: To provide a framework for quantifying the robustness of treatment ranks based on Surface Under the Cumulative RAnking curve (SUCRA) in network meta-analysis (NMA) and investigating potential factors associated with lack of robustness. METHODS: We propose the use of Cohen's kappa to quantify the agreement between SUCRA-based treatment ranks estimated through NMA of a complete data set and a subset of it. We illustrate our approach using five published NMA data sets, where robustness was assessed by removing studies one at a time. RESULTS: Overall, SUCRA-based treatment ranks were robust to individual studies in the five data sets we considered. We observed more incidences of disagreement between ranks in the networks with larger numbers of treatments. Most treatments moved only one or two ranks up or down. The lowest quadratic weighted kappa estimate observed across all networks was in the network with the smallest number of treatments (4), where weighted kappa=40%. In the network with the largest number of treatments (12), the lowest observed quadratic weighted kappa=89%, reflecting a small shift in this network's treatment ranks overall. Preliminary observations suggest that a study's size, the number of studies making a treatment comparison, and the agreement of a study's estimated treatment effect(s) with those estimated by other studies making the same comparison(s) may explain the overall robustness of treatment ranks to studies. CONCLUSIONS: Investigating robustness or sensitivity in an NMA may reveal outlying rank changes that are clinically or policy-relevant. Cohen's kappa is a useful measure that permits investigation into study characteristics that may explain varying sensitivity to individual studies. However, this study presents a framework as a proof of concept and further investigation is required to identify potential factors associated with the robustness of treatment ranks using more extensive empirical evaluations.
Fetched live from OpenAlex and de-inverted. Abstracts are not stored in this database: the inverted indexes are 8.6 GB of the frame’s 9.3 GB of text, and the host has 13 GB free.
How this classification was reachedexpand
Full frame distilled prediction
Teacher imitationNot calibrated prevalence, not ground truth. Human validation pending. Learned from the 10,348 direct Codex labels and 10,348 direct Gemma labels. Candidate is the union of thresholded teacher heads; consensus is their intersection. These outputs are machine_predicted_unvalidated and are not human labels or direct frontier model labels.
Codex and Gemma teacher scores by category
| Category | Codex | Gemma |
|---|---|---|
| Metaresearch | 0.287 | 0.006 |
| Meta-epidemiology (narrow) | 0.000 | 0.000 |
| Meta-epidemiology (broad) | 0.009 | 0.005 |
| Bibliometrics | 0.001 | 0.005 |
| Science and technology studies | 0.000 | 0.000 |
| Scholarly communication | 0.001 | 0.000 |
| Open science | 0.002 | 0.000 |
| Research integrity | 0.000 | 0.000 |
| Insufficient payload (model declined to judge) | 0.029 | 0.000 |
Machine scores (provisional)
The two teacher heads of the student model, read on this work. A score orders the frame for review; it never asserts a category, and the validation status ships verbatim with every row.
Baseline scores from an immature model (maturity gate not passed, 7 training rounds). Scores rank; they never assert a category.
score_only:v0-immature-baseline · verbatim from the scoring run: score_only means the number may rank works, and no category label ships from itClassification
machine, unvalidatedMachine predicted; a candidate call from one teacher head, not a consensus.
How this classification was reached, model by model and score by score, is at the end of the page under "How this classification was reached".