Accuracy-fairness trade-off in ML for healthcare: A quantitative evaluation of bias mitigation strategies
Bibliographic record
Abstract
Although machine learning (ML) has significant potential to improve healthcare decision-making, embedded biases in algorithms and datasets risk exacerbating health disparities across demographic groups. To address this challenge, it is essential to rigorously evaluate bias mitigation strategies to ensure fairness and reliability across patient populations. The aim of this research is to propose a comprehensive evaluation framework that systematically assesses a wide range of bias mitigation techniques at pre-processing, in-processing, and post-processing stages, using both single- and multi-stage intervention approaches. This study evaluates bias mitigation strategies across three clinical prediction tasks: breast cancer diagnosis, stroke prediction, and Alzheimer’s disease detection. Our evaluation employs group- and individual-level fairness metrics, contextualized for specific sensitive attributes relevant to each dataset. Beyond fairness-accuracy trade-offs, we demonstrate how metric selection must align with clinical goals (e.g., parity metrics for equitable access, confusion-matrix metrics for diagnostics). Our results reinforce that no single classifier or mitigation strategy is universally optimal, underscoring the value of our proposed framework for evaluating fairness and accuracy throughout the bias mitigation process. According to the results, Adversarial Debiasing improved fairness by 95% in breast cancer diagnosis without compromising accuracy. Reweighing was most effective in stroke prediction, boosting fairness by 41%, and Reject Option Classification yielded nearly 50% fairness improvement in Alzheimer’s detection. Multi-stage bias mitigation did not consistently lead to better outcomes, and in many cases, fairness gains came at the expense of accuracy. These findings provide practical guidance for selecting fairness-aware machine learning strategies in healthcare, aiding both model development and benchmarking across diverse clinical applications. • We propose a comprehensive evaluation framework that systematically compares single- and multi-stage bias mitigation strategies across pre-, in-, and post-processing stages in healthcare machine learning. • The framework assesses accuracy–fairness trade-off using task-sensitive fairness metrics across three clinical prediction tasks, breast cancer diagnosis, stroke prediction, and Alzheimer’s disease detection, demonstrating real-world applicability. • We empirically validate multiple bias mitigation strategies, showing that significant fairness improvements can be achieved while maintaining high diagnostic accuracy.
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 | 0.055 | 0.174 |
| Meta-epidemiology (narrow) | 0.002 | 0.000 |
| Meta-epidemiology (broad) | 0.001 | 0.001 |
| Bibliometrics | 0.002 | 0.001 |
| Science and technology studies | 0.001 | 0.003 |
| Scholarly communication | 0.003 | 0.004 |
| Open science | 0.001 | 0.003 |
| Research integrity | 0.002 | 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".