{"status":"ok","message-type":"work","message-version":"1.0.0","message":{"institution":[{"name":"medRxiv"}],"indexed":{"date-parts":[[2026,1,16]],"date-time":"2026-01-16T13:41:07Z","timestamp":1768570867592,"version":"3.49.0"},"posted":{"date-parts":[[2025,10,7]]},"group-title":"Health Informatics","reference-count":40,"publisher":"openRxiv","license":[{"start":{"date-parts":[[2025,10,7]],"date-time":"2025-10-07T00:00:00Z","timestamp":1759795200000},"content-version":"vor","delay-in-days":0,"URL":"http:\/\/creativecommons.org\/licenses\/by\/4.0\/"}],"content-domain":{"domain":[],"crossmark-restriction":false},"short-container-title":[],"accepted":{"date-parts":[[2025,11,14]]},"abstract":"<jats:title>Abstract<\/jats:title>\n                <jats:sec>\n                  <jats:title>Background<\/jats:title>\n                  <jats:p>Pancreatic cancer (PANC) is often diagnosed at late stages due to the absence of specific early symptoms, resulting in one of the highest cancer mortality rates. While imaging modalities such as MRI and CT offer high diagnostic accuracy, their population-wide application is however impractical due to the cost. Electronic health records (EHRs) provide a routine, easily accessible, longitudinal and scalable data source for risk prediction, particularly for diseases with no specific symptom such as PANC.<\/jats:p>\n                <\/jats:sec>\n                <jats:sec>\n                  <jats:title>Method<\/jats:title>\n                  <jats:p>We introduce PANCDetect, a multimodal framework that leverages large language model (LLM)-derived embeddings of diagnoses, procedures, medications, and laboratory tests, and integrates these data modalities through a Transformer-based architecture. We train the model on MarketScan (\u2248250M patients), and validate it externally on additional large real-world EHR datasets of University of Michigan Precision Health, or UMPH data (n\u22486M patients) and OneFlorida+ data (n\u224826M patients). We then fine-tuned the general model on UMPH EHR data. We evaluated the performance of both models using metrics including area under the receiver operating characteristic curve (AUROC) and area under the precision-recall-gain curve (AUPRG). We assessed the top predictive features with integrated gradients (IG).<\/jats:p>\n                <\/jats:sec>\n                <jats:sec>\n                  <jats:title>Result<\/jats:title>\n                  <jats:p>In the MarketScan cohort, PANCDetect achieved an AUROC of 0.812 and AUPRG of 0.851 at the 6-month prediction window, and an AUROC of 0.735 and AUPRG of 0.629 for 60-month prediction, significantly outperforming CancerRiskNet. External validation on UMPH and OneFlorida+ demonstrated good generalizability, with 6-months AUROC scores of 0.711 and 0.793, respectively. Fine-tuning on UMPH with laboratory data further improved performance, reaching an AUROC of 0.927 and an AUPRG of 0.979 at 6 months. Even at the 60-month horizon, the refined PANCDetect model maintained strong performance, with an AUROC of 0.835 and AUPRG of 0.911. Attribution analysis highlighted type 2 diabetes, pancreatic diseases, personal and family cancer history as the most important risk factors.<\/jats:p>\n                <\/jats:sec>\n                <jats:sec>\n                  <jats:title>Conclusion<\/jats:title>\n                  <jats:p>PANCDetect is the state-of-the-art method integrating multimodal EHR data with LLM embeddings for accurate, interpretable, and generalizable early prediction of pancreatic cancer. This framework holds promise for precision screening of high-risk patients, with the potential to improve survival outcomes without increasing healthcare costs.<\/jats:p>\n                <\/jats:sec>","DOI":"10.1101\/2025.10.03.25337210","type":"posted-content","created":{"date-parts":[[2025,10,7]],"date-time":"2025-10-07T22:50:19Z","timestamp":1759877419000},"source":"Crossref","is-referenced-by-count":0,"title":["PANCDetect: Early Detection of Pancreatic Cancer from Multimodal EHR data with LLM Embeddings"],"prefix":"10.64898","author":[{"given":"Zicheng","family":"Jin","sequence":"first","affiliation":[],"role":[{"role":"author","vocabulary":"crossref"}]},{"given":"Xuhui","family":"Guo","sequence":"additional","affiliation":[],"role":[{"role":"author","vocabulary":"crossref"}]},{"given":"Zehua","family":"Wang","sequence":"additional","affiliation":[],"role":[{"role":"author","vocabulary":"crossref"}]},{"given":"Qiang","family":"Yang","sequence":"additional","affiliation":[],"role":[{"role":"author","vocabulary":"crossref"}]},{"given":"Xiaotong","family":"Yang","sequence":"additional","affiliation":[],"role":[{"role":"author","vocabulary":"crossref"}]},{"given":"Xinyu","family":"Zhang","sequence":"additional","affiliation":[],"role":[{"role":"author","vocabulary":"crossref"}]},{"ORCID":"https:\/\/orcid.org\/0000-0002-1403-0396","authenticated-orcid":false,"given":"Rui","family":"Yin","sequence":"additional","affiliation":[],"role":[{"role":"author","vocabulary":"crossref"}]},{"ORCID":"https:\/\/orcid.org\/0000-0002-4654-2126","authenticated-orcid":false,"given":"Lana X.","family":"Garmire","sequence":"additional","affiliation":[],"role":[{"role":"author","vocabulary":"crossref"}]}],"member":"54368","reference":[{"key":"2025111701200517000_2025.10.03.25337210v3.1","doi-asserted-by":"crossref","first-page":"102193","DOI":"10.1016\/j.ctrv.2021.102193","article-title":"Blood biomarkers for differential diagnosis and early detection of pancreatic cancer","volume":"96","year":"2021","journal-title":"Cancer Treat. Rev"},{"key":"2025111701200517000_2025.10.03.25337210v3.2","doi-asserted-by":"crossref","first-page":"698","DOI":"10.1016\/S2468-1253(19)30416-9","article-title":"Early detection of pancreatic cancer","volume":"5","year":"2020","journal-title":"Lancet Gastroenterol. Hepatol"},{"key":"2025111701200517000_2025.10.03.25337210v3.3","doi-asserted-by":"publisher","DOI":"10.1053\/j.gastro.2020.10.007"},{"key":"2025111701200517000_2025.10.03.25337210v3.4","doi-asserted-by":"crossref","first-page":"1257","DOI":"10.1002\/cac2.12204","article-title":"Early screening and diagnosis strategies of pancreatic cancer: a comprehensive review","volume":"41","year":"2021","journal-title":"Cancer Commun"},{"key":"2025111701200517000_2025.10.03.25337210v3.5","doi-asserted-by":"crossref","unstructured":"S\u00f8reide, K. , Ismail, W. , Roals\u00f8, M. , Ghotbi, J. & Zaharia, C . Early Diagnosis of Pancreatic Cancer: Clinical Premonitions, Timely Precursor Detection and Increased Curative-Intent Surgery. Cancer Control 30, 10732748231154711 (2023).","DOI":"10.1177\/10732748231154711"},{"key":"2025111701200517000_2025.10.03.25337210v3.6","doi-asserted-by":"publisher","DOI":"10.1186\/s13244-020-00861-y"},{"key":"2025111701200517000_2025.10.03.25337210v3.7","doi-asserted-by":"crossref","unstructured":"Wittram, R. , Kreis, L. , K\u00f6nig, H.-H. & Brettschneider, C . Economic evaluations of early detection strategies for pancreatic cancer: a systematic review. Eur. J. Health Econ. (2025).","DOI":"10.1007\/s10198-025-01793-4"},{"key":"2025111701200517000_2025.10.03.25337210v3.8","doi-asserted-by":"publisher","DOI":"10.14309\/ajg.0000000000002870"},{"key":"2025111701200517000_2025.10.03.25337210v3.9","doi-asserted-by":"crossref","first-page":"171","DOI":"10.1038\/s41746-024-01166-w","article-title":"A multi-center study on the adaptability of a shared foundation model for electronic health records","volume":"7","year":"2024","journal-title":"Npj Digit. Med"},{"key":"2025111701200517000_2025.10.03.25337210v3.10","doi-asserted-by":"crossref","first-page":"521","DOI":"10.1038\/s41746-025-01872-z","article-title":"Transformer patient embedding using electronic health records enables patient stratification and progression analysis","volume":"8","year":"2025","journal-title":"Npj Digit. Med"},{"key":"2025111701200517000_2025.10.03.25337210v3.11","doi-asserted-by":"crossref","first-page":"1407","DOI":"10.1038\/s41597-024-04168-1","article-title":"Cross-Standard Health Data Harmonization using Semantics of Data Elements","volume":"11","year":"2024","journal-title":"Sci. Data"},{"key":"2025111701200517000_2025.10.03.25337210v3.12","doi-asserted-by":"crossref","first-page":"7155","DOI":"10.1038\/s41598-020-62922-y","article-title":"BEHRT: Transformer for Electronic Health Records","volume":"10","year":"2020","journal-title":"Sci. Rep"},{"key":"2025111701200517000_2025.10.03.25337210v3.13","doi-asserted-by":"crossref","first-page":"1106","DOI":"10.1109\/JBHI.2022.3224727","article-title":"Hi-BEHRT: Hierarchical Transformer-Based Model for Accurate Prediction of Clinical Events Using Multimodal Longitudinal Electronic Health Records","volume":"27","year":"2023","journal-title":"IEEE J. Biomed. Health Inform"},{"key":"2025111701200517000_2025.10.03.25337210v3.14","doi-asserted-by":"crossref","first-page":"86","DOI":"10.1038\/s41746-021-00455-y","article-title":"Med-BERT: pretrained contextualized embeddings on large-scale structured electronic health records for disease prediction","volume":"4","year":"2021","journal-title":"Npj Digit. Med"},{"key":"2025111701200517000_2025.10.03.25337210v3.15","doi-asserted-by":"publisher","DOI":"10.48550\/ARXIV.2502.04397"},{"key":"2025111701200517000_2025.10.03.25337210v3.16","doi-asserted-by":"publisher","DOI":"10.48550\/ARXIV.2504.03964"},{"key":"2025111701200517000_2025.10.03.25337210v3.17","doi-asserted-by":"publisher","DOI":"10.1093\/bib\/bbac409"},{"key":"2025111701200517000_2025.10.03.25337210v3.18","doi-asserted-by":"crossref","first-page":"357","DOI":"10.1038\/s41586-023-06160-y","article-title":"Health system-scale language models are all-purpose prediction engines","volume":"619","year":"2023","journal-title":"Nature"},{"key":"2025111701200517000_2025.10.03.25337210v3.19","doi-asserted-by":"crossref","first-page":"104888","DOI":"10.1016\/j.ebiom.2023.104888","article-title":"A pancreatic cancer risk prediction model (Prism) developed and validated on large-scale US clinical data","volume":"98","year":"2023","journal-title":"eBioMedicine"},{"key":"2025111701200517000_2025.10.03.25337210v3.20","doi-asserted-by":"publisher","DOI":"10.1016\/j.ejca.2020.10.019"},{"key":"2025111701200517000_2025.10.03.25337210v3.21","doi-asserted-by":"publisher","DOI":"10.1038\/s41591-023-02332-5"},{"key":"2025111701200517000_2025.10.03.25337210v3.22","doi-asserted-by":"crossref","first-page":"465","DOI":"10.1038\/s41746-025-01869-8","article-title":"Enhancing EHR-based pancreatic cancer prediction with LLM-derived embeddings","volume":"8","year":"2025","journal-title":"Npj Digit. Med"},{"key":"2025111701200517000_2025.10.03.25337210v3.23","doi-asserted-by":"publisher","DOI":"10.1007\/978-3-030-51455-6_20"},{"key":"2025111701200517000_2025.10.03.25337210v3.24","doi-asserted-by":"crossref","first-page":"100257","DOI":"10.1016\/j.xgen.2023.100257","article-title":"The Michigan Genomics Initiative: A biobank linking genotypes and electronic clinical records in Michigan Medicine patients","volume":"3","year":"2023","journal-title":"Cell Genomics"},{"key":"2025111701200517000_2025.10.03.25337210v3.25","doi-asserted-by":"publisher","DOI":"10.1097\/ACM.0000000000002029"},{"key":"2025111701200517000_2025.10.03.25337210v3.26","unstructured":"Flach, P. & Kull, M . Precision-Recall-Gain Curves: PR Analysis Done Right. In Proc. 28th Conference on Neural Information Processing Systems (NeurIPS) (2015)."},{"key":"2025111701200517000_2025.10.03.25337210v3.27","unstructured":"Sundararajan, M. , Taly, A. & Yan, Q . Axiomatic Attribution for Deep Networks. In Proc. 34th International Conference on Machine Learning (ICML) (2017)."},{"key":"2025111701200517000_2025.10.03.25337210v3.28","doi-asserted-by":"crossref","first-page":"194","DOI":"10.1038\/s41746-022-00742-2","article-title":"A Large Clinical Language Model to Unlock Patient Information from Unstructured Electronic Health Records","volume":"5","year":"2022","journal-title":"Npj Digit. Med"},{"key":"2025111701200517000_2025.10.03.25337210v3.29","unstructured":"Excoffier, J.-B. et al. Generalist embedding models are better at short-context clinical semantic search than specialized embedding models. Preprint at https:\/\/doi.org\/10.48550\/ARXIV.2401.01943 (2024)."},{"key":"2025111701200517000_2025.10.03.25337210v3.30","doi-asserted-by":"publisher","DOI":"10.48550\/ARXIV.2502.17403"},{"key":"2025111701200517000_2025.10.03.25337210v3.31","doi-asserted-by":"publisher","DOI":"10.1111\/1475-6773.12819"},{"key":"2025111701200517000_2025.10.03.25337210v3.32","doi-asserted-by":"publisher","DOI":"10.2337\/dc20-0207"},{"key":"2025111701200517000_2025.10.03.25337210v3.33","doi-asserted-by":"publisher","DOI":"10.1038\/sj.bjc.6602619"},{"key":"2025111701200517000_2025.10.03.25337210v3.34","doi-asserted-by":"publisher","DOI":"10.4093\/dmj.2025.0237"},{"key":"2025111701200517000_2025.10.03.25337210v3.35","doi-asserted-by":"crossref","first-page":"e361","DOI":"10.3399\/BJGP.2021.0552","article-title":"Intra-abdominal cancer risk with abdominal pain: a prospective cohort primary care study","volume":"72","year":"2022","journal-title":"Br. J. Gen. Pract. J. R. Coll. Gen. Pract"},{"key":"2025111701200517000_2025.10.03.25337210v3.36","first-page":"169","article-title":"Surveillance of Individuals with a Family History of Pancreatic Cancer and Inherited Cancer Syndromes: A Strategy for Detecting Early Pancreatic Cancers","volume":"9","year":"2019","journal-title":"Diagn. Basel Switz"},{"key":"2025111701200517000_2025.10.03.25337210v3.37","doi-asserted-by":"crossref","first-page":"zrad023","DOI":"10.1093\/bjsopen\/zrad023","article-title":"Hereditary colorectal, gastric, and pancreatic cancer: comprehensive review","volume":"7","year":"2023","journal-title":"BJS Open"},{"key":"2025111701200517000_2025.10.03.25337210v3.38","first-page":"803","article-title":"Risk of pancreatic cancer in breast cancer families from the breast cancer family registry","volume":"22","year":"2013","journal-title":"Cancer Epidemiol. Biomark. Prev. Publ. Am. Assoc. Cancer Res. Cosponsored Am. Soc. Prev. Oncol"},{"key":"2025111701200517000_2025.10.03.25337210v3.39","doi-asserted-by":"publisher","DOI":"10.1002\/ijc.25148"},{"key":"2025111701200517000_2025.10.03.25337210v3.40","doi-asserted-by":"crossref","first-page":"132","DOI":"10.1038\/s41746-021-00489-2","article-title":"Privacy-first health research with federated learning","volume":"4","year":"2021","journal-title":"Npj Digit. Med"}],"container-title":[],"original-title":[],"link":[{"URL":"https:\/\/syndication.highwire.org\/content\/doi\/10.1101\/2025.10.03.25337210","content-type":"unspecified","content-version":"vor","intended-application":"similarity-checking"}],"deposited":{"date-parts":[[2026,1,15]],"date-time":"2026-01-15T20:47:42Z","timestamp":1768510062000},"score":1,"resource":{"primary":{"URL":"http:\/\/medrxiv.org\/lookup\/doi\/10.1101\/2025.10.03.25337210"}},"subtitle":[],"short-title":[],"issued":{"date-parts":[[2025,10,7]]},"references-count":40,"URL":"https:\/\/doi.org\/10.1101\/2025.10.03.25337210","relation":{},"subject":[],"published":{"date-parts":[[2025,10,7]]},"subtype":"preprint"}}