Bibliographic record
Abstract
The Cochrane Collaboration should be congratulated for its dedication to documenting continuing clinical trials, teaching critical appraisal, and supporting research into new methods of reviewing the literature. However, Cochrane Reviews can be created by untrained people who simply follow an algorithmic approach and are unaware of important methodological issues. Therefore, the objective of this article is to highlight important limitations of Cochrane Reviews, including the Review Manager software1 that is required,2 the inappropriate use of a summary statistic, and finally the restriction to only randomised controlled trial (RCT) data. To illustrate these points, I have used a 1% random sample of Cochrane Reviews—that is, 16 studies numbered 1, 101, 201…1501 of 1596 of the Cochrane Database on 3 April 2003.3–18 There are important limitations to the software required by the Cochrane Collaboration (Review Manager). Most important is that Review Manager cannot include ( a ) results based on survival analyses—for example, most appropriate analysis for time to next injury—and instead calculates relative risks based on simple proportions (this leads to inappropriate estimates when patients have different lengths of follow up19) and ( b ) analyses adjusted for confounding—that is, multiple regression analysis. Software is available, but requires statistical expertise that would preclude the algorithmic approach. Whereas the Cochrane Collaboration could improve its software, the other two major problems are process oriented. The Cochrane Collaboration promotes wide participation, and this leads to inexperienced authors and peer reviewers. For example, it is not always appropriate to pool data into one overall summary statistic,20 and, even when it is, there are different methods—that is, fixed and random effects models—to be used depending on the structure of the data.20,21 In the …
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 machine prediction
Teacher imitationNot calibrated prevalence, not ground truth. Human validation pending. The Gemma side is a direct model label for every work in the frame, read from the title-only record. The Codex side is a classifier learned from the 10,348 direct Codex labels and calibrated to design-weighted sample rates; fields without enough sample support carry no Codex call. Candidate is the union of the two sides; consensus is their intersection. These outputs are machine_predicted_unvalidated and are not human labels.
Distilled classifier scores by category (both heads)
| Category | Codex | Gemma |
|---|---|---|
| Metaresearch | 0.083 | 0.314 |
| Meta-epidemiology (narrow) | 0.006 | 0.009 |
| Meta-epidemiology (broad) | 0.023 | 0.012 |
| Bibliometrics | 0.035 | 0.040 |
| Science and technology studies | 0.003 | 0.009 |
| Scholarly communication | 0.029 | 0.031 |
| Open science | 0.011 | 0.013 |
| Research integrity | 0.019 | 0.025 |
| Insufficient payload (model declined to judge) | 0.099 | 0.055 |
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 source (direct Gemma or distilled Codex), 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".