A large language model-based tool for identifying relationships to industry in research on the carcinogenicity of benzene, cobalt, and aspartame
Bibliographic record
Abstract
BACKGROUND: Industry-funded research poses a threat to the validity of scientific inference on carcinogenic hazards. Scientists require tools to better identify and characterize industry sponsored research across bodies of evidence to reduce the possible influence of industry bias in evidence synthesis reviews. We applied a novel large language model (LLM)-based tool named InfluenceMapper to demonstrate and evaluate its performance in identifying relationships to industry in research on the carcinogenicity of benzene, cobalt, and aspartame. METHODS: All epidemiological, animal cancer, and mechanistic studies included in systematic reviews on the carcinogenicity of the three agents by the IARC Monographs programme. Selected agents were recently evaluated by the Monographs and are of commercial interest by major industries. InfluenceMapper extracted disclosed entities in study publications and classified up to 40 possible disclosed relationship types between each entity and the study and between each entity and author. A human classified entities as 'industry or industry-funded' and assessed relationships with industry for potential conflicts of interest. Positive predictive values described the extent of true positive relationships identified by InfluenceMapper compared to human assessment. RESULTS: Analyses included 2,046 studies for all three agents. We identified 320 disclosed industry or industry-funded entities from InfluenceMapper output that were involved in 770 distinct study-entity and author-entity relationships. For each agent, between 4 and 8% of studies disclosed funding by industry and 1-4% of studies had at least one author who disclosed receiving industry funding directly. Industry trade associations for all three agents funded 22 studies published in 16 journals over a 37-year span. Aside from funding, the most prevalent disclosed relationships with industry were receiving data, holding employment, paid consulting, and providing expert testimony. Positive predictive values were excellent (≥ 98%) for study-entity relationships but declined for relationships with individual authors. CONCLUSIONS: LLM-based tools can significantly expedite and bolster the detection of disclosed conflicts of interest from industry sponsored research in cancer prevention. Possible use cases include facilitating the assessment of bias from industry studies in evidence synthesis reviews and alerting scientists to the influence of industry on scientific inference. Persistent challenges in ascertaining conflicts of interest underscore the urgent need for standardized, transparent, and enforceable disclosures in biomedical journals.
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 | 0.001 | 0.000 |
| Meta-epidemiology (narrow) | 0.000 | 0.000 |
| Meta-epidemiology (broad) | 0.000 | 0.000 |
| Bibliometrics | 0.000 | 0.000 |
| Science and technology studies | 0.000 | 0.000 |
| Scholarly communication | 0.000 | 0.000 |
| Open science | 0.000 | 0.000 |
| Research integrity | 0.000 | 0.000 |
| Insufficient payload (model declined to judge) | 0.000 | 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".