Navigating Fairness in Healthcare: A Comparative Analysis of Single-Stage and Multi-Stage Bias Mitigation Strategies
Bibliographic record
Abstract
Despite machine learning’s potential to revolutionize healthcare decision-making, systematic biases in algorithms and training data can worsen already-existing health inequities among different populations. Employing bias mitigation strategies can enhance the reliability and fairness of AI healthcare systems across diverse patient populations. The aim of this study is to develop a framework for evaluating various bias mitigation techniques and quantitatively measuring fairness in binary classification. We examined these techniques across pre-processing, in-processing, and post-processing stages, utilizing both single- and multi-stage intervention approaches. Our evaluation focused on their impact on model reliability, employing group and individual fairness metrics with a particular emphasis on the sensitive attribute of age. Our analysis focused on two healthcare datasets—stroke prediction and breast cancer diagnosis—highlighting the trade-offs between fairness and accuracy. Results show that Adversarial Debiasing significantly improves fairness by 95% in the breast cancer dataset without compromising accuracy. For stroke prediction, Reweighing is the most effective, enhancing fairness by 41% with minimal accuracy impact. These findings indicate that no single classifier or bias mitigation strategy fits all datasets; understanding context and experimenting with various methods is essential. While multistage bias mitigation can reduce bias, it may not always be effective, and efforts to minimize bias often come at the cost of accuracy, potentially hindering overall model performance in classification tasks.
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 | 0.031 | 0.081 |
| Meta-epidemiology (narrow) | 0.001 | 0.000 |
| Meta-epidemiology (broad) | 0.001 | 0.001 |
| Bibliometrics | 0.001 | 0.001 |
| Science and technology studies | 0.001 | 0.002 |
| Scholarly communication | 0.002 | 0.003 |
| Open science | 0.001 | 0.002 |
| Research integrity | 0.001 | 0.001 |
| Insufficient payload (model declined to judge) | 0.001 | 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".