Bringing Code to Data: Do Not Forget Governance
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
Developing or independently evaluating algorithms in biomedical research is difficult because of restrictions on access to clinical data. Access is restricted because of privacy concerns, the proprietary treatment of data by institutions (fueled in part by the cost of data hosting, curation, and distribution), concerns over misuse, and the complexities of applicable regulatory frameworks. The use of cloud technology and services can address many of the barriers to data sharing. For example, researchers can access data in high performance, secure, and auditable cloud computing environments without the need for copying or downloading. An alternative path to accessing data sets requiring additional protection is the model-to-data approach. In model-to-data, researchers submit algorithms to run on secure data sets that remain hidden. Model-to-data is designed to enhance security and local control while enabling communities of researchers to generate new knowledge from sequestered data. Model-to-data has not yet been widely implemented, but pilots have demonstrated its utility when technical or legal constraints preclude other methods of sharing. We argue that model-to-data can make a valuable addition to our data sharing arsenal, with 2 caveats. First, model-to-data should only be adopted where necessary to supplement rather than replace existing data-sharing approaches given that it requires significant resource commitments from data stewards and limits scientific freedom, reproducibility, and scalability. Second, although model-to-data reduces concerns over data privacy and loss of local control when sharing clinical data, it is not an ethical panacea. Data stewards will remain hesitant to adopt model-to-data approaches without guidance on how to do so responsibly. To address this gap, we explored how commitments to open science, reproducibility, security, respect for data subjects, and research ethics oversight must be re-evaluated in a model-to-data context.
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 | 0.058 | 0.098 |
| Meta-epidemiology (narrow) | 0.000 | 0.000 |
| Meta-epidemiology (broad) | 0.000 | 0.000 |
| Bibliometrics | 0.000 | 0.001 |
| Science and technology studies | 0.000 | 0.000 |
| Scholarly communication | 0.001 | 0.000 |
| Open science | 0.010 | 0.007 |
| Research integrity | 0.000 | 0.001 |
| Insufficient payload (model declined to judge) | 0.002 | 0.001 |
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