This is the authors' abstract. We don't add key points for this paper.
Machine learning (ML) has become an essential tool in healthcare, supporting diagnosis, prognosis, and treatment decisions. However, biases present in pre-training data can compromise both model performance and fairness, disproportionately affecting underrepresented groups. This study systematically examines the impact of four pre-training bias metrics on the accuracy of three ML models across four health-related datasets. Our findings show that more data does not necessarily translate to better performance, particularly when data imbalance and bias are present. Moreover, pre-training bias metrics are associated with accuracy disparities, underscoring the importance of proactive bias assessment to develop more equitable ML models in healthcare.
Answers are grounded in this paper and cite their sources.
No takes yet. Share an insight, caveat, or question.
Rodrigues et al. (2025) studied this question.
Synapse has enriched 8 closely related papers on similar clinical questions. Consider them for comparative context: