[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"detail-sidebar-cat-0-en-105":3,"doc-seo-351258-105":59,"doc-detail-351258-en":134},{"code":4,"msg":5,"data":6},0,"success",[7,13,18,23,28,33,38,43,48,51,55],{"id":8,"doc_module":4,"doc_module_name":9,"category_name":10,"show_sort_weight":11,"slug":12},1,"Document","Story & Novel",90,"story-novel",{"id":14,"doc_module":4,"doc_module_name":9,"category_name":15,"show_sort_weight":16,"slug":17},2,"Literature",80,"literature",{"id":19,"doc_module":4,"doc_module_name":9,"category_name":20,"show_sort_weight":21,"slug":22},4,"Exam",70,"exam",{"id":24,"doc_module":4,"doc_module_name":9,"category_name":25,"show_sort_weight":26,"slug":27},5,"Comic",60,"comic",{"id":29,"doc_module":4,"doc_module_name":9,"category_name":30,"show_sort_weight":31,"slug":32},6,"Technology",50,"technology",{"id":34,"doc_module":4,"doc_module_name":9,"category_name":35,"show_sort_weight":36,"slug":37},7,"Healthcare",40,"healthcare",{"id":39,"doc_module":4,"doc_module_name":9,"category_name":40,"show_sort_weight":41,"slug":42},8,"Research & Report",30,"research-report",{"id":44,"doc_module":4,"doc_module_name":9,"category_name":45,"show_sort_weight":46,"slug":47},9,"Religion & Spirituality",20,"religion-spirituality",{"id":46,"doc_module":4,"doc_module_name":9,"category_name":49,"show_sort_weight":46,"slug":50},"World Cup","world-cup",{"id":52,"doc_module":4,"doc_module_name":9,"category_name":53,"show_sort_weight":52,"slug":54},10,"Lifestyle","lifestyle",{"id":56,"doc_module":4,"doc_module_name":9,"category_name":57,"show_sort_weight":24,"slug":58},19,"General","general",{"code":4,"msg":60,"data":61},"ok",{"site_id":62,"language":63,"slug":64,"title":65,"keywords":66,"description":67,"schema_data":68,"social_meta":127,"head_meta":129,"extra_data":131,"updated_unix":133},105,"en","mitigating-algorithmic-bias-in-cancer-site-classification-models","Mitigating Algorithmic Bias in Cancer Site Classification Models","","Integrating artificial intelligence into cancer diagnostics improves tumor classification beyond rule-based systems, yet models may still encode demographic biases. This large-scale applied study probes a deep learning cancer site classifier to quantify race information captured in document embeddings, then tests performance and fairness changes after removing race-correlated embedding dimensions via post-training sensitivity pruning. Results assess macro-F1, accuracy, race-stratified performance, and demographic parity and equalized odds before versus after pruning.",{"@graph":69,"@context":126},[70,84,105],{"@type":71,"itemListElement":72},"BreadcrumbList",[73,77,79,82],{"item":74,"name":75,"@type":76,"position":8},"https://docshare.wps.com","Home","ListItem",{"item":78,"name":9,"@type":76,"position":14},"https://docshare.wps.com/document/",{"item":80,"name":40,"@type":76,"position":81},"https://docshare.wps.com/document/research-report/",3,{"item":83,"name":65,"@type":76,"position":19},"https://docshare.wps.com/document/mitigating-algorithmic-bias-in-cancer-site-classification-models/351258/",{"url":83,"name":65,"@type":85,"image":86,"author":91,"headline":65,"publisher":94,"fileFormat":97,"inLanguage":63,"description":67,"dateModified":98,"datePublished":99,"encodingFormat":97,"isAccessibleForFree":100,"interactionStatistic":101},"DigitalDocument",{"url":87,"@type":88,"width":89,"height":90},"https://docshare.wps.com/thumbnails/mitigating-algorithmic-bias-in-cancer-site-classification-models/351258.png","ImageObject",300,407,{"name":92,"@type":93},"Rhys","Person",{"url":74,"name":95,"@type":96},"DocShare","Organization","application/pdf","2026-10-30","2026-09-22",true,{"@type":102,"interactionType":103,"userInteractionCount":52},"InteractionCounter",{"@type":104},"ViewAction",{"@type":106,"mainEntity":107},"FAQPage",[108,114,118,122],{"name":109,"@type":110,"acceptedAnswer":111},"What is the main objective of the study on cancer site classification models?","Question",{"text":112,"@type":113},"To determine whether document embeddings encode racial information and whether pruning race-correlated embedding dimensions changes classification performance or fairness.","Answer",{"name":115,"@type":110,"acceptedAnswer":116},"How were the cancer site classiﬁer embeddings generated and used in the analysis?",{"text":117,"@type":113},"A hierarchical self-attention network (HiSAN) generates 400-dimensional document embeddings from 3.5 million pathology reports, which then train downstream gradient-boosted decision trees for site and for racial category prediction.",{"name":119,"@type":110,"acceptedAnswer":120},"What were the key findings regarding demographic bias and model performance after pruning?",{"text":121,"@type":113},"Feature overlap between cancer-site and race prediction models was minimal, cumulative importance of race information was negligible, and post-training pruning caused no meaningful accuracy drop (0.07%).",{"name":123,"@type":110,"acceptedAnswer":124},"Which fairness and performance metrics were compared before and after pruning?",{"text":125,"@type":113},"The study compared overall macro-F1 and accuracy, race-stratiﬁed macro-F1, and group fairness metrics based on demographic parity and equalized odds before versus after pruning.","https://schema.org",{"og:url":83,"og:type":128,"og:title":65,"og:site_name":95,"og:description":67},"article",{"robots":130,"canonical":83},"index,follow",{"doc_id":132,"site_id":62},351258,1790442863,{"code":4,"msg":5,"data":135},{"doc_id":132,"user_id":136,"nickname":92,"user_avatar":137,"doc_module":4,"category_id":39,"category_name":40,"doc_title":65,"doc_description":67,"doc_content":138,"file_id":139,"file_url":140,"file_type":141,"file_size":142,"view_count":52,"is_deleted":4,"is_public":8,"is_downloadable":8,"audit_status":8,"page_count":52,"language":143,"language_code":63,"site_id":62,"html_lang":63,"table_of_contents":144,"faqs":145,"seo_title":146,"seo_description":67,"update_tm":147,"read_time":148,"preview_page_count":19,"preview_urls":149},687207024643,"https://ap-avatar.wpscdn.com/davatar_3d24733baf745e90a7e4bdd5f77d97b2","Original Reports | Artiﬁcial Intelligence  \nMitigating Algorithmic Bias in Cancer Site Classiﬁcation Models  \nAbhishek Shivanna, PhD1  ; Adam Spannaus, PhD1 ; Jordan Tschida, PhD1 ; John Gounley, PhD1  ; Patrycja Krawczuk, MS1  ; and Heidi Hanson, PhD1  \nDOI [https://doi.org/10.1200/CCI-25-00250](https://doi.org/10.1200/CCI-25-00250)  \n\n| ABSTRACT |  |\n| --- | --- |\n| PURPOSE | Integrating artiﬁcial intelligence in cancer diagnostics has improved tumor classiﬁcation beyond rule-based systems. Despite these advancements, these models may still encode demographic biases. We conducted a large-scale, applied bias-probing study of a deep learning–based cancer site classiﬁer to quantify race information encoded in document embeddings. We then evaluated how performance changes when race-correlated embedding dimensions are removed in a post-training sensitivity analysis. |\n| METHODS | The cancer site classiﬁer was trained using 3.5 million electronic cancer pathology reports from six of the National Cancer Institute’s SEER registries. We trained a hierarchical self-attention network to generate 400-dimensional document embeddings. These embeddings were used to train two downstream, gradient-boosted decision tree classiﬁers: one to classify the cancer sites and another to predict racial categories. We identiﬁed overlapping features by intersecting the top 50 feature-importance rankings from the site and race models and computed their cumulative feature importance in each model. As a post hoc sensitivity analysis, we progressively pruned these overlapping dimensions, retrained the site model, and compared overall macro-F1 and accuracy, race-stratiﬁed macro-F1, and group fairness metrics on the basis of demographic parity and equalized odds before and after pruning. |\n| RESULTS | The analysis revealed minimal feature overlap between the cancer site and race prediction models, and the cumulative importance scores indicated a negligible inﬂuence of racial information on clinical predictions. Post-training pruning of overlapping features did not compromise the models’ diagnostic accuracy, with a 0.07% loss in accuracy. |\n| CONCLUSION | Our ﬁndings demonstrate that HiSAN-generated embeddings from SEER data can be used effectively in cancer site classiﬁcation without signiﬁcant demographic bias inﬂuencing the outcomes. Post-training pruning therefore functions as a practical audit and sensitivity check. |\n\nACCOMPANYING CONTENT  \n Data Supplement  \nAccepted January 23, 2026 Published March 11, 2026  \nJCO Clin Cancer Inform 10:e2500250  \nPublished by American Society of Clinical Oncology  \nLicensed under the Creative Commons Attribution 4.0 License  \nINTRODUCTION  \nArtiﬁcial intelligence (AI) has enormous potential for advancing precision medicine and public health surveillance. In oncology, AI models that use machine learning (ML) and deep learning techniques have shown proﬁciency in deciphering complex clinical data for accurate cancer case identiﬁcation and classiﬁcation.1-4 However, as more AI tools have been integrated into clinical decision making, concerns have emerged regarding their potential to inherit or amplify biases present in their training data.5-10  \nHistorically, health care data and algorithms have often reﬂected biases present in their training data, thereby  \nperpetuating demographic differences in health outcomes if not addressed.11-14 Such biases can originate from imbalanced data sets, algorithmic design choices, or the erroneous interpretation of sociodemographic variables as clinical predictors.15-23 Developing robust methods for identifying and mitigating these biases is essential to ensure that AI technologies beneﬁt all populations.  \nPronounced differences between races in incidence rates, prognosis, and treatment outcomes are seen in colon, lung, breast, ovarian, and prostate cancers. For example, prostate cancer mortality among Black/African American men is nearly twice that among White men.24-28 Biases in algori","cbCaii4vBe1JfSP0","https://ap.wps.com/l/cbCaii4vBe1JfSP0","pdf",1670207,"English","# Abstract\n## Purpose\n## Methods\n## Results\n## Conclusion\n# Introduction\n# Context\n## Key Objective\n## Knowledge Generated\n## Relevance","[{\"question\":\"What is the main objective of the study on cancer site classification models?\",\"answer\":\"To determine whether document embeddings encode racial information and whether pruning race-correlated embedding dimensions changes classification performance or fairness.\"},{\"question\":\"How were the cancer site classiﬁer embeddings generated and used in the analysis?\",\"answer\":\"A hierarchical self-attention network (HiSAN) generates 400-dimensional document embeddings from 3.5 million pathology reports, which then train downstream gradient-boosted decision trees for site and for racial category prediction.\"},{\"question\":\"What were the key findings regarding demographic bias and model performance after pruning?\",\"answer\":\"Feature overlap between cancer-site and race prediction models was minimal, cumulative importance of race information was negligible, and post-training pruning caused no meaningful accuracy drop (0.07%).\"},{\"question\":\"Which fairness and performance metrics were compared before and after pruning?\",\"answer\":\"The study compared overall macro-F1 and accuracy, race-stratiﬁed macro-F1, and group fairness metrics based on demographic parity and equalized odds before versus after pruning.\"}]","Mitigating Algorithmic Bias in Cancer Site Classification Models | PDF",1790093247,25,[150,154,158,162],{"page_no":8,"url":151,"width":152,"height":153},"https://res-academy.cache.wpscdn.com/upload/kdocshare/2026-11-16/dd666c46e5b0a5f65463c00136815f01.webp",983,1315,{"page_no":14,"url":155,"width":156,"height":157},"https://res-academy.cache.wpscdn.com/upload/kdocshare/2026-11-16/00c614574ceb67410735abd9a32b6535.webp",982,1314,{"page_no":81,"url":159,"width":160,"height":161},"https://res-academy.cache.wpscdn.com/upload/kdocshare/2026-11-16/50991ea01263f7dc7c6be28132d5dc9b.webp",981,1312,{"page_no":19,"url":163,"width":164,"height":165},"https://res-academy.cache.wpscdn.com/upload/kdocshare/2026-11-16/8b372e605b6a6ff3e78b72def64facbf.webp",980,1311]