Comparing machine and human reviewers to evaluate the risk of bias in randomized controlled trials
Why this work is in the frame
A frame that forgets how it found something cannot be audited. These are the routes that admitted this work.
Bibliographic record
Abstract
BACKGROUND: Evidence from new health technologies is growing, along with demands for evidence to inform policy decisions, creating challenges in completing health technology assessments (HTAs)/systematic reviews (SRs) in a timely manner. Software can decrease the time and burden by automating the process, but evidence validating such software is limited. We tested the accuracy of RobotReviewer, a semi-autonomous risk of bias (RoB) assessment tool, and its agreement with human reviewers. METHODS: Two reviewers independently conducted RoB assessments on a sample of randomized controlled trials (RCTs), and their consensus ratings were compared with those generated by RobotReviewer. Agreement with the human reviewers was assessed using percent agreement and weighted kappa (κ). The accuracy of RobotReviewer was also assessed by calculating the sensitivity, specificity, and area under the curve in comparison to the consensus agreement of the human reviewers. RESULTS: The study included 372 RCTs. Inter-rater reliability ranged from κ = -0.06 (no agreement) for blinding of participants and personnel to κ = 0.62 (good agreement) for random sequence generation (excluding overall RoB). RobotReviewer was found to use a high percentage of "irrelevant supporting quotations" to complement RoB assessments for blinding of participants and personnel (72.6%), blinding of outcome assessment (70.4%), and allocation concealment (54.3%). CONCLUSION: RobotReviewer can help with risk of bias assessment of RCTs but cannot replace human evaluations. Thus, reviewers should check and validate RoB assessments from RobotReviewer by consulting the original article when not relevant supporting quotations are provided by RobotReviewer. This consultation is in line with the recommendation provided by the developers.
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.
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 | 0.961 | 0.974 |
| Meta-epidemiology (narrow) | 0.000 | 0.000 |
| Meta-epidemiology (broad) | 0.030 | 0.005 |
| Bibliometrics | 0.001 | 0.003 |
| 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) | 0.004 | 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 it