Federated Analysis With Differential Privacy in Oncology Research: Longitudinal Observational Study Across Hospital Data Warehouses
Bibliographic record
Abstract
Background: Federated analytics in health care allows researchers to perform statistical queries on remote datasets without access to the raw data. This method arose from the need to perform statistical analysis on larger datasets collected at multiple health care centers while avoiding regulatory, governance, and privacy issues that might arise if raw data were collected at a central location outside the health care centers. Despite some pioneering work, federated analytics is still not widely used on real-world data, and to our knowledge, no real-world study has yet combined it with other privacy-enhancing techniques such as differential privacy (DP). Objective: The first objective of this study was to deploy a federated architecture in a real-world setting. The oncology study used for this deployment compared the medical health care management of patients with metastatic non-small cell lung cancer before and after the first wave of COVID-19 pandemic. The second goal was to test DP in this real-world scenario to assess its practicality and use as a privacy-enhancing technology. Methods: A federated architecture platform was set up in the Toulouse, Reims, and Foch centers. After harmonization of the data in each center, statistical analyses were performed using DataSHIELD (Data aggregation through anonymous summary-statistics from harmonized individual-level databases), a federated analysis R library, and a new open-source DP DataSHIELD package was implemented (dsPrivacy). Results: A total of 50 patients were enrolled in the Toulouse and Reims centers and 49 in the Foch center. We have shown that DataSHIELD is a practical tool to efficiently conduct our study across all 3 centers without exposing data on a central node, once a sufficient setup has been established to configure a secure network between hospitals. All planned aggregated results were successfully generated. We also observed that DP can be implemented in practice with promising trade-offs between privacy and accuracy, and we built a library that will prove useful for future work. Conclusions: The federated architecture platform made it possible to run a multicenter study on real-world oncology data while ensuring strong privacy guarantees using differential privacy.
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 | 0.027 | 0.053 |
| Meta-epidemiology (narrow) | 0.000 | 0.001 |
| Meta-epidemiology (broad) | 0.001 | 0.002 |
| Bibliometrics | 0.002 | 0.005 |
| Science and technology studies | 0.001 | 0.001 |
| Scholarly communication | 0.002 | 0.002 |
| Open science | 0.001 | 0.003 |
| Research integrity | 0.001 | 0.002 |
| Insufficient payload (model declined to judge) | 0.002 | 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 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".