Why federated learning will do little to overcome the deeply embedded biases in clinical medicine
Bibliographic record
Abstract
We read with great interest the article by van Genderen et al. [1] which provides a contemporary and comprehensive overview of the potential of federating data access and data sharing in intensive care.Importantly, the authors list the perpetuation of biases encoded in clinical care practice as a major potential shortcoming.Furthermore, they state that this could be mitigated by "ensuring an adequate representation of hospitals from various regions worldwide could lead to more diverse and inclusive health datasets."We agree that the use of diverse and inclusive health datasets should be promoted as a necessary first step to build fair machine learning algorithms.However, we do not believe that this will be sufficient to overcome the deeply embedded biases in medicine from a knowledge system that is designed around a majoritized few.Even with high quality data from the intensive care units from across the world, the social patterning of the data generation process can still produce artificial intelligence (AI) that is bound to preserve and even scale existing disparities in care with resulting inequities in patient outcomes.There are numerous examples of data issues that stem from the social patterning of the data capture and data generation process (Fig. 1).These include, but are certainly not limited to, (1) the differential performance of medical devices used to measure physiologic signals
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 | 0.023 | 0.113 |
| Meta-epidemiology (narrow) | 0.001 | 0.001 |
| Meta-epidemiology (broad) | 0.001 | 0.001 |
| Bibliometrics | 0.001 | 0.001 |
| Science and technology studies | 0.003 | 0.010 |
| Scholarly communication | 0.005 | 0.012 |
| Open science | 0.002 | 0.003 |
| Research integrity | 0.031 | 0.052 |
| Insufficient payload (model declined to judge) | 0.004 | 0.006 |
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".