Fairness without Demographics on Electronic Health Records
Track: Traditional track
Keywords: Electronic Health Records, Fairness Without Demographics, Clinical Model
TL;DR: We address fairness concerns in machine learning models for electronic health records without demographic information.
Abstract: Machine learning systems are notoriously prone to biased predictions about certain demographic groups, leading to algorithmic fairness issues. Due to patient privacy and social inequity concerns, some demographic information may not be available for training a clinical algorithm. Moreover, the complex interaction of different demographics can lead to a lot of unknown minority subpopulations. These challenges greatly limit the applicability of existing group fairness algorithms. To improve the fairness-without-demographics algorithm in the clinical regime, we argue that the gradients of clinical models can provide insights for relieving inequities. In this paper, we adopt an adversarial weighting architecture and leverage the correlation between model gradients and demographic groups to improve the identification and increase the exposure of underrepresented groups. We learn the weights of different samples by constructing a graph where samples with similar gradients are connected. Unlike the surrogate grouping methods that cluster groups by proxy sensitive attribute like features and labels, which can be inaccurate, our method provides a soft grouping mechanism that is more robust. The results show that our method can improve fairness significantly without sacrificing too much of the overall accuracy.
Presentation And Attendance Policy: I have read and agree with the symposium's policy on behalf of myself and my co-authors.
Ethics Board Approval: Yes, we have/will include(d) information about IRB approval or its equivalent, in the manuscript.
Data And Code Availability: Yes, we will make data and code available upon acceptance.
Primary Area: Challenges limiting the adoption of modern ML in healthcare
Student First Author: Yes, the primary author of the manuscript is a student.
Submission Number: 29
Loading