Times are displayed in (UTC-08:00) Pacific Time (US & Canada) Change
11/11/2024 |
8:30 AM – 10:00 AM |
Franciscan B
S22: AI Fairness and Ethics - Justice League
Presentation Type: Oral
Session Chair:
Shaun Grannis, MD, MS, FAAFP, FACMI, FAMIA - Regenstrief Institute
The Impact of Race, Ethnicity, and Gender on Fairness in a Multicenter Model Predicting Glaucoma Outcomes Using Electronic Health Records
Presentation Time: 08:30 AM - 08:45 AM
Abstract Keywords: Fairness and Elimination of Bias, Racial Disparities, Diversity, Equity, Inclusion, Accessibility, and Health Equity, Clinical Decision Support, Machine Learning
Primary Track: Applications
Programmatic Theme: Clinical Informatics
In this study of 39,090 glaucoma patients within the Sight Outcomes Research Collaborative, a large multicenter registry of electronic health records, we predicted which patients will need glaucoma surgery, comparing modeling approaches that included the sensitive attributes of gender, race, and ethnicity, excluded them, or trained separate models for subgroups. We found that the most fair or best performing modeling approaches in the internal test set were not robust to evaluation on patients from two external test sites held-out from the training process.
Speaker(s):
Sophia Wang, MD, MS
Stanford University
Author(s):
Rohith Ravindranath, M.S. - Stanford University; Tina Hernandez-Boussard - Stanford University; Joshua Stein, MD, MS - University of Michigan;
Presentation Time: 08:30 AM - 08:45 AM
Abstract Keywords: Fairness and Elimination of Bias, Racial Disparities, Diversity, Equity, Inclusion, Accessibility, and Health Equity, Clinical Decision Support, Machine Learning
Primary Track: Applications
Programmatic Theme: Clinical Informatics
In this study of 39,090 glaucoma patients within the Sight Outcomes Research Collaborative, a large multicenter registry of electronic health records, we predicted which patients will need glaucoma surgery, comparing modeling approaches that included the sensitive attributes of gender, race, and ethnicity, excluded them, or trained separate models for subgroups. We found that the most fair or best performing modeling approaches in the internal test set were not robust to evaluation on patients from two external test sites held-out from the training process.
Speaker(s):
Sophia Wang, MD, MS
Stanford University
Author(s):
Rohith Ravindranath, M.S. - Stanford University; Tina Hernandez-Boussard - Stanford University; Joshua Stein, MD, MS - University of Michigan;
Ensuring Fairness in Detecting Mild Cognitive Impairment with MRI
Presentation Time: 08:45 AM - 09:00 AM
Abstract Keywords: Fairness and Elimination of Bias, Machine Learning, Imaging Informatics
Primary Track: Applications
Programmatic Theme: Translational Bioinformatics
Machine learning (ML) algorithms play a crucial role in the early and accurate diagnosis of Alzheimer's Disease (AD), which is essential for effective treatment planning. However, existing methods are not well-suited for identifying Mild Cognitive Impairment (MCI), a critical transitional stage between normal aging and AD. This inadequacy is primarily due to label imbalance and bias from different sensitve attributes in MCI classification. To overcome these challenges, we have designed an end-to-end fairness-aware approach for label-imbalanced classification, tailored specifically for neuroimaging data. This method, built on the recently developed FACIMS framework, integrates into STREAMLINE, an automated ML environment. We evaluated our approach against nine other ML algorithms and found that it achieves comparable balanced accuracy to other methods while prioritizing fairness in classifications with five different sensitive attributes. This analysis contributes to the development of equitable and reliable ML diagnostics for MCI detection.
Speaker(s):
Boning Tong
University of Pennsylvania
Author(s):
Boning Tong - University of Pennsylvania; Travyse Edwards, BA - University of Pennsylvania; Shu Yang; Bojian Hou, PhD - University of Pennsylvania; Davoud Tarzanagh, PhD - University of Pennsylvania; Ryan Urbanowicz, PhD - Cedars-Sinai Medical Center; Jason Moore, PhD, FACMI - Cedars-Sinai; Marylyn Ritchie, PhD - University of Pennsylvania, Perelman School of Medicine; Christos Davatzikos, PhD - University of Pennsylvania; Li Shen, Ph.D. - University of Pennsylvania;
Presentation Time: 08:45 AM - 09:00 AM
Abstract Keywords: Fairness and Elimination of Bias, Machine Learning, Imaging Informatics
Primary Track: Applications
Programmatic Theme: Translational Bioinformatics
Machine learning (ML) algorithms play a crucial role in the early and accurate diagnosis of Alzheimer's Disease (AD), which is essential for effective treatment planning. However, existing methods are not well-suited for identifying Mild Cognitive Impairment (MCI), a critical transitional stage between normal aging and AD. This inadequacy is primarily due to label imbalance and bias from different sensitve attributes in MCI classification. To overcome these challenges, we have designed an end-to-end fairness-aware approach for label-imbalanced classification, tailored specifically for neuroimaging data. This method, built on the recently developed FACIMS framework, integrates into STREAMLINE, an automated ML environment. We evaluated our approach against nine other ML algorithms and found that it achieves comparable balanced accuracy to other methods while prioritizing fairness in classifications with five different sensitive attributes. This analysis contributes to the development of equitable and reliable ML diagnostics for MCI detection.
Speaker(s):
Boning Tong
University of Pennsylvania
Author(s):
Boning Tong - University of Pennsylvania; Travyse Edwards, BA - University of Pennsylvania; Shu Yang; Bojian Hou, PhD - University of Pennsylvania; Davoud Tarzanagh, PhD - University of Pennsylvania; Ryan Urbanowicz, PhD - Cedars-Sinai Medical Center; Jason Moore, PhD, FACMI - Cedars-Sinai; Marylyn Ritchie, PhD - University of Pennsylvania, Perelman School of Medicine; Christos Davatzikos, PhD - University of Pennsylvania; Li Shen, Ph.D. - University of Pennsylvania;
Emerging Algorithmic Bias: Fairness Drift and Model Sustainability
Presentation Time: 09:00 AM - 09:15 AM
Abstract Keywords: Fairness and Elimination of Bias, Machine Learning, Evaluation
Primary Track: Foundations
Programmatic Theme: Clinical Informatics
While the tendency of clinical AI model accuracy to drift over time is well-documented, the potential for algorithmic biases to emerge over time has yet to be characterized. We explored temporal fairness in models predicting post-surgical complications in a national population over 11 years. We observed complex, variable patterns of fairness drift and interactions between updating and fairness using population-level maintenance strategies. Responsible deployment of clinical AI requires evaluation of and strategies for fairness sustainability.
Speaker(s):
Sharon Davis, PhD
Vanderbilt University Medical Center
Author(s):
Chad Dorn, PSM - Vanderbilt University Medical Center; Daniel Park, BA - Vanderbilt University Medical Center; Michael Matheny, MD, MS, MPH, FACMI, FAMIA - Vanderbilt University Medical Center;
Presentation Time: 09:00 AM - 09:15 AM
Abstract Keywords: Fairness and Elimination of Bias, Machine Learning, Evaluation
Primary Track: Foundations
Programmatic Theme: Clinical Informatics
While the tendency of clinical AI model accuracy to drift over time is well-documented, the potential for algorithmic biases to emerge over time has yet to be characterized. We explored temporal fairness in models predicting post-surgical complications in a national population over 11 years. We observed complex, variable patterns of fairness drift and interactions between updating and fairness using population-level maintenance strategies. Responsible deployment of clinical AI requires evaluation of and strategies for fairness sustainability.
Speaker(s):
Sharon Davis, PhD
Vanderbilt University Medical Center
Author(s):
Chad Dorn, PSM - Vanderbilt University Medical Center; Daniel Park, BA - Vanderbilt University Medical Center; Michael Matheny, MD, MS, MPH, FACMI, FAMIA - Vanderbilt University Medical Center;
Causal Fairness for Decomposing Racial and Sex Disparities in Treatment Allocation Using Real-World Data
Presentation Time: 09:15 AM - 09:30 AM
Abstract Keywords: Causal Inference, Health Equity, Data Mining
Primary Track: Foundations
Programmatic Theme: Clinical Research Informatics
Recent efforts toward equitable healthcare have highlighted disparities in treatment allocation based on race, sex, socioeconomic status, and health insurance access, emphasizing the need for fairness in clinical decisions. However, the mechanisms behind these disparities are not well understood due to limited datasets and the complexity of causal inference methods. To address this, we introduce a causal fairness analysis (CFA) framework to assess the impact of race and sex on treatment decisions using electronic health record (EHR) data. This framework combines causal mediation analysis and double machine learning to explore how protected attributes influence clinical decisions. Our study focuses on decomposing treatment disparities for coronary artery disease (CAD), examining direct, indirect (through social and clinical health determinants), and confounded effects (by other demographics). This innovative approach aims to provide a detailed understanding of healthcare disparities, shedding light on potential biases and discrimination, and advancing the quest for more equitable healthcare solutions.
Speaker(s):
Linying Zhang, PhD
Washington University in St. Louis
Author(s):
Linying Zhang, PhD - Washington University in St. Louis; Xinzhuo Jiang, MS - Columbia University Department of Biomedical Informatics; Karthik Natarajan, PhD - Columbia University Dept of Biomedical Informatics; George Hripcsak, MD - Columbia University Irving Medical Center;
Presentation Time: 09:15 AM - 09:30 AM
Abstract Keywords: Causal Inference, Health Equity, Data Mining
Primary Track: Foundations
Programmatic Theme: Clinical Research Informatics
Recent efforts toward equitable healthcare have highlighted disparities in treatment allocation based on race, sex, socioeconomic status, and health insurance access, emphasizing the need for fairness in clinical decisions. However, the mechanisms behind these disparities are not well understood due to limited datasets and the complexity of causal inference methods. To address this, we introduce a causal fairness analysis (CFA) framework to assess the impact of race and sex on treatment decisions using electronic health record (EHR) data. This framework combines causal mediation analysis and double machine learning to explore how protected attributes influence clinical decisions. Our study focuses on decomposing treatment disparities for coronary artery disease (CAD), examining direct, indirect (through social and clinical health determinants), and confounded effects (by other demographics). This innovative approach aims to provide a detailed understanding of healthcare disparities, shedding light on potential biases and discrimination, and advancing the quest for more equitable healthcare solutions.
Speaker(s):
Linying Zhang, PhD
Washington University in St. Louis
Author(s):
Linying Zhang, PhD - Washington University in St. Louis; Xinzhuo Jiang, MS - Columbia University Department of Biomedical Informatics; Karthik Natarajan, PhD - Columbia University Dept of Biomedical Informatics; George Hripcsak, MD - Columbia University Irving Medical Center;
Fairness of AI Collaboration and Suppression in Emergency Triage
Presentation Time: 09:30 AM - 09:45 AM
Abstract Keywords: Fairness and Elimination of Bias, Machine Learning, Clinical Decision Support
Primary Track: Foundations
Programmatic Theme: Clinical Informatics
Both AI collaboration and suppression to reduce automation bias pose the risk of perpetuating unfairness present in AI models. In this work, we evaluate the impact of AI in a simulated emergency department triage scenario with an auditor model to suppress AI predictions upon algorithmic fairness. We find that unfairness present in the AI is perpetuated into human-AI collaboration; however, our findings also suggest that AI suppression can aid in reducing this unfairness.
Speaker(s):
Katherine (Katie) Brown, PhD
Vanderbilt University Medical Center
Author(s):
Katherine (Katie) Brown, PhD - Vanderbilt University Medical Center; Michael Cauley, PhD - Vanderbilt University Medical Center; Benjamin Collins, MD - Vanderbilt University Medical Center; Nicholas Jackson, PhD Student Biomedical Informatics - Vanderbilt University; Bradley Malin, PhD - Vanderbilt University Medical Center; Jessica Ancker, MPH, PhD, FACMI - Vanderbilt University Medical Center;
Presentation Time: 09:30 AM - 09:45 AM
Abstract Keywords: Fairness and Elimination of Bias, Machine Learning, Clinical Decision Support
Primary Track: Foundations
Programmatic Theme: Clinical Informatics
Both AI collaboration and suppression to reduce automation bias pose the risk of perpetuating unfairness present in AI models. In this work, we evaluate the impact of AI in a simulated emergency department triage scenario with an auditor model to suppress AI predictions upon algorithmic fairness. We find that unfairness present in the AI is perpetuated into human-AI collaboration; however, our findings also suggest that AI suppression can aid in reducing this unfairness.
Speaker(s):
Katherine (Katie) Brown, PhD
Vanderbilt University Medical Center
Author(s):
Katherine (Katie) Brown, PhD - Vanderbilt University Medical Center; Michael Cauley, PhD - Vanderbilt University Medical Center; Benjamin Collins, MD - Vanderbilt University Medical Center; Nicholas Jackson, PhD Student Biomedical Informatics - Vanderbilt University; Bradley Malin, PhD - Vanderbilt University Medical Center; Jessica Ancker, MPH, PhD, FACMI - Vanderbilt University Medical Center;
Enhancement of Fairness in AI for Chest X-ray Classification
Presentation Time: 09:45 AM - 10:00 AM
Abstract Keywords: Fairness and Elimination of Bias, Machine Learning, Racial disparities
Primary Track: Foundations
The use of artificial intelligence (AI) in medicine has shown promise to improve the quality of healthcare decisions. However, AI can be biased in a manner that produces unfair predictions for certain demographic subgroups. In MIMIC-CXR, a publicly available dataset of over 300,000 chest X-ray images, diagnostic AI has been shown to have a higher false negative rate for racial minorities. We evaluated the capacity of synthetic data augmentation, oversampling, and demographic-based corrections to enhance the fairness of AI predictions. We show that adjusting unfair predictions for demographic attributes, such as race, is ineffective at improving fairness or predictive performance. However, using oversampling and synthetic data augmentation to modify disease prevalence reduced such disparities by 65.4% and 10.6%, respectively. Moreover, such fairness gains were accomplished without reduction in performance (95% CI AUC: [0.816, 0.820] versus [0.817, 0.820] versus [0.817, 0.821] for baseline, oversampling, and augmentation, respectively).
Speaker(s):
Nicholas Jackson, PhD Student Biomedical Informatics
Vanderbilt University
Author(s):
Nicholas Jackson, PhD Student Biomedical Informatics - Vanderbilt University; Chao Yan, PhD - Vanderbilt University Medical Center; Bradley Malin, PhD - Vanderbilt University Medical Center;
Presentation Time: 09:45 AM - 10:00 AM
Abstract Keywords: Fairness and Elimination of Bias, Machine Learning, Racial disparities
Primary Track: Foundations
The use of artificial intelligence (AI) in medicine has shown promise to improve the quality of healthcare decisions. However, AI can be biased in a manner that produces unfair predictions for certain demographic subgroups. In MIMIC-CXR, a publicly available dataset of over 300,000 chest X-ray images, diagnostic AI has been shown to have a higher false negative rate for racial minorities. We evaluated the capacity of synthetic data augmentation, oversampling, and demographic-based corrections to enhance the fairness of AI predictions. We show that adjusting unfair predictions for demographic attributes, such as race, is ineffective at improving fairness or predictive performance. However, using oversampling and synthetic data augmentation to modify disease prevalence reduced such disparities by 65.4% and 10.6%, respectively. Moreover, such fairness gains were accomplished without reduction in performance (95% CI AUC: [0.816, 0.820] versus [0.817, 0.820] versus [0.817, 0.821] for baseline, oversampling, and augmentation, respectively).
Speaker(s):
Nicholas Jackson, PhD Student Biomedical Informatics
Vanderbilt University
Author(s):
Nicholas Jackson, PhD Student Biomedical Informatics - Vanderbilt University; Chao Yan, PhD - Vanderbilt University Medical Center; Bradley Malin, PhD - Vanderbilt University Medical Center;