[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"doc-detail-126727-en":3,"doc-seo-126727-105":31,"detail-sidebar-cat-0-en-105":96},{"code":4,"msg":5,"data":6},0,"success",{"doc_id":7,"user_id":8,"nickname":9,"user_avatar":10,"doc_module":4,"category_id":11,"category_name":12,"doc_title":13,"doc_description":14,"doc_content":15,"file_id":16,"file_url":17,"file_type":18,"file_size":19,"view_count":20,"is_deleted":4,"is_public":21,"is_downloadable":21,"audit_status":21,"page_count":22,"language":23,"language_code":24,"site_id":25,"html_lang":24,"table_of_contents":26,"faqs":27,"seo_title":28,"seo_description":14,"update_tm":29,"read_time":30},126727,962084925782,"Ava Thompson","https://ap-avatar.wpscdn.com/davatar_9964176cb1d06d4a9deccf72a44ae3dc",7,"Healthcare","An adversarial training framework for mitigating algorithmic biases in clinical machine learning","Machine learning in healthcare can inherit and amplify sensitive-group biases from training data, leading to unfair prediction and potentially worse clinical decisions. An adversarial training framework is proposed to mitigate biases acquired during data collection by learning model parameters that avoid inferring sensitive features. The framework is evaluated on rapid COVID-19 prediction, targeting hospital site and patient ethnicity biases, using equality of odds to improve outcome fairness while retaining clinically effective screening performance (negative predictive values above 0.98) with validation across multiple cohorts.","ARTICLE OPEN  \n[www.nature.com/npjdigitalmed](www.nature.com/npjdigitalmed)  \nAn adversarial training framework for mitigating algorithmic biases in clinical machine learning  \nJenny Yang 1 ✉ , Andrew A. S. Soltan 2,3, David W. Eyre 4, Yang Yang5,7 and David A. Clifton1,6,7  \n\n|  | Machine learning is becoming increasingly prominent in healthcare. Although its benefits are clear, growing attention is being given to how these tools may exacerbate existing biases and disparities. In this study, we introduce an adversarial training framework that is capable of mitigating biases that may have been acquired through data collection. We demonstrate this proposed framework on the real-world task of rapidly predicting COVID-19, and focus on mitigating site-specific (hospital) and demographic (ethnicity) biases. Using the statistical definition of equalized odds, we show that adversarial training improves outcome fairness, while still achieving clinically-effective screening performances (negative predictive values >0.98) . We compare our method to previous benchmarks, and perform prospective and external validation across four independent hospital cohorts |  |\n| --- | --- | --- |\n|  |  |  |\n|  |  |  |\n|  |  |  |\n|  |  |  |\n|  |  |  |\n|  |  |  |\n|  | Our method can be generalized to any outcomes, models, and definitions of fairness. |  |\n|  | npj Digital Medicine (2023)6:55; [https://doi.org/10.1038/s41746-023-00805-y](https://doi.org/10.1038/s41746-023-00805-y) |  |\n|  |  |  |\n\n.  \nINTRODUCTION  \nA fundamental observation in machine learning (ML) research is that models can become biased based on the samples used during training. This can lead to poorer predictive performance and unfair decision-making. Here, we define “bias” as a difference in performance between subgroups for a predictive task1,2; and similarly, define an “unfair” decision as any result that is skewed towards a particular group or population2–4. In other words, given a classifier which predicts labels yi from features xi for samples i, bias arises when a statistical property for the distribution of {yi, i ϵ Z} differs from the distribution of {yi, i ϵ Z’}, where Z is considered a sensitive subgroup (i.e., a group that a model may be biased against) and Z’ is its non-sensitive complement. With respect to fairness, previous machine learning works have evaluated statistical properties such as demographic parity, equality of odds, and equal opportunity2–6.  \nIf a machine learning model acquires unintentional biases, it may be unable to capture the true relationship between the features and the target outcome. This is particularly harmful insensitive domains such as healthcare because: (1) a biased model can lead to inaccurate predictions for critical and, potentially, lifealtering decisions; (2) a bias against a particular group can result in those patients receiving poorer care compared to those in other groups; and (3) a biased model can exacerbate and propagate existing inequities in healthcare and society. Thus, in our study, we propose a framework for bias mitigation using adversarial debiasing, whereby a model is trained to learn parameters that do not infer sensitive features. We consider a classifier which predicts yi from features xi, while remaining unbiased with respect to some sensitive feature, Z. To evaluate group outcome fairness, we use the statistical metric of equality of odds, which states that a classifier Ŷ is fair if Ŷ and Z are conditionally independent given Y2–5. For binary classification, this is equivalent to P(Ŷ = 1| Y = y, Z = 0) = P(Ŷ = 1| Y = y, Z = 1), y ϵ {0, 1} . Using the real-world clinical task of COVID-19 screening, we demonstrate the  \neffectiveness of this technique for two sensitive features-patient ethnicity and hospital location.  \nPrevious works on training fair machine learning systems have shown that ML models can be trained to reduce demographicbased biases. Such biases are highly relevant in clinical settings, as they can unintention","cbCailpcIgvue4nu","https://ap.wps.com/l/cbCailpcIgvue4nu","pdf",1544482,2,1,10,"English","en",105,"# Introduction\n## Bias in machine learning models and clinical impact\n## Fairness definitions and equality of odds\n## Related work: demographic and location-related biases\n## Study focus: adversarial debiasing for COVID-19 screening","[{\"question\":\"What problem does the adversarial training framework address in clinical machine learning?\",\"answer\":\"It addresses performance disparities caused by biases learned from training data. These biases can lead to unfair and less accurate clinical predictions across sensitive subgroups.\"},{\"question\":\"How is fairness evaluated in the study?\",\"answer\":\"The study uses the statistical metric of equality of odds. For binary classification, fairness is expressed as equal conditional probabilities of positive predictions across sensitive groups given the true outcome.\"},{\"question\":\"Which sensitive features are tested in the COVID-19 screening task?\",\"answer\":\"The evaluation targets patient ethnicity and hospital location (site-specific/hospital) biases using real-world data.\"},{\"question\":\"Does adversarial training reduce bias without harming clinical performance?\",\"answer\":\"Yes. The framework improves outcome fairness while maintaining clinically effective screening performance, reported with negative predictive values greater than 0.98.\"}]","An adversarial training framework for mitigating algorithmic biases in clinical machine learning | PDF",1785934440,25,{"code":4,"msg":32,"data":33},"ok",{"site_id":25,"language":24,"slug":34,"title":13,"keywords":35,"description":14,"schema_data":36,"social_meta":91,"head_meta":93,"extra_data":95,"updated_unix":29},"an-adversarial-training-framework-for-mitigating-algorithmic-biases-in-clinical-machine-learning","",{"@graph":37,"@context":90},[38,54,69],{"@type":39,"itemListElement":40},"BreadcrumbList",[41,45,48,51],{"item":42,"name":43,"@type":44,"position":21},"https://docshare.wps.com","Home","ListItem",{"item":46,"name":47,"@type":44,"position":20},"https://docshare.wps.com/document/","Document",{"item":49,"name":12,"@type":44,"position":50},"https://docshare.wps.com/document/healthcare/",3,{"item":52,"name":13,"@type":44,"position":53},"https://docshare.wps.com/document/an-adversarial-training-framework-for-mitigating-algorithmic-biases-in-clinical-machine-learning/126727/",4,{"url":52,"name":13,"@type":55,"author":56,"headline":13,"publisher":58,"fileFormat":61,"inLanguage":24,"description":14,"dateModified":62,"datePublished":63,"encodingFormat":61,"isAccessibleForFree":64,"interactionStatistic":65},"DigitalDocument",{"name":9,"@type":57},"Person",{"url":42,"name":59,"@type":60},"DocShare","Organization","application/pdf","2026-08-24","2026-08-05",true,{"@type":66,"interactionType":67,"userInteractionCount":20},"InteractionCounter",{"@type":68},"ViewAction",{"@type":70,"mainEntity":71},"FAQPage",[72,78,82,86],{"name":73,"@type":74,"acceptedAnswer":75},"What problem does the adversarial training framework address in clinical machine learning?","Question",{"text":76,"@type":77},"It addresses performance disparities caused by biases learned from training data. These biases can lead to unfair and less accurate clinical predictions across sensitive subgroups.","Answer",{"name":79,"@type":74,"acceptedAnswer":80},"How is fairness evaluated in the study?",{"text":81,"@type":77},"The study uses the statistical metric of equality of odds. For binary classification, fairness is expressed as equal conditional probabilities of positive predictions across sensitive groups given the true outcome.",{"name":83,"@type":74,"acceptedAnswer":84},"Which sensitive features are tested in the COVID-19 screening task?",{"text":85,"@type":77},"The evaluation targets patient ethnicity and hospital location (site-specific/hospital) biases using real-world data.",{"name":87,"@type":74,"acceptedAnswer":88},"Does adversarial training reduce bias without harming clinical performance?",{"text":89,"@type":77},"Yes. The framework improves outcome fairness while maintaining clinically effective screening performance, reported with negative predictive values greater than 0.98.","https://schema.org",{"og:url":52,"og:type":92,"og:title":13,"og:site_name":59,"og:description":14},"article",{"robots":94,"canonical":52},"index,follow",{"doc_id":7,"site_id":25},{"code":4,"msg":5,"data":97},[98,102,106,110,115,120,123,128,133,136,139],{"id":21,"doc_module":4,"doc_module_name":47,"category_name":99,"show_sort_weight":100,"slug":101},"Story & Novel",90,"story-novel",{"id":20,"doc_module":4,"doc_module_name":47,"category_name":103,"show_sort_weight":104,"slug":105},"Literature",80,"literature",{"id":53,"doc_module":4,"doc_module_name":47,"category_name":107,"show_sort_weight":108,"slug":109},"Exam",70,"exam",{"id":111,"doc_module":4,"doc_module_name":47,"category_name":112,"show_sort_weight":113,"slug":114},5,"Comic",60,"comic",{"id":116,"doc_module":4,"doc_module_name":47,"category_name":117,"show_sort_weight":118,"slug":119},6,"Technology",50,"technology",{"id":11,"doc_module":4,"doc_module_name":47,"category_name":12,"show_sort_weight":121,"slug":122},40,"healthcare",{"id":124,"doc_module":4,"doc_module_name":47,"category_name":125,"show_sort_weight":126,"slug":127},8,"Research & Report",30,"research-report",{"id":129,"doc_module":4,"doc_module_name":47,"category_name":130,"show_sort_weight":131,"slug":132},9,"Religion & Spirituality",20,"religion-spirituality",{"id":131,"doc_module":4,"doc_module_name":47,"category_name":134,"show_sort_weight":131,"slug":135},"World Cup","world-cup",{"id":22,"doc_module":4,"doc_module_name":47,"category_name":137,"show_sort_weight":22,"slug":138},"Lifestyle","lifestyle",{"id":140,"doc_module":4,"doc_module_name":47,"category_name":141,"show_sort_weight":111,"slug":142},19,"General","general"]