JAMIA Open最新文献

{"title":"A tabular prior-data fitted network-based prediction model integrating ferritin and total iron-binding capacity for 30-day all-cause mortality in critically ill cancer patients.","authors":"Lin-Lin Liu, Lei Han, Zuo-Lin Xiang","doi":"10.1093/jamiaopen/ooag148","DOIUrl":"10.1093/jamiaopen/ooag148","url":null,"abstract":"<p><strong>Objectives: </strong>Accurately predicting outcomes for critically ill cancer patients remains challenging. This study aimed to integrate biologically relevant iron metabolism markers into short-term mortality prediction and to develop an interpretable, externally validated model based on a tabular prior-data fitted network (TabPFN) for risk stratification in this population.</p><p><strong>Materials and methods: </strong>We conducted a retrospective cohort study using data from Medical Information Mart for Intensive Care IV (MIMIC-IV) (training and internal validation) and eICU Collaborative Research Database (eICU-CRD) (external validation), including critically ill adult patients with cancer. Associations between iron metabolism markers (ferritin, serum iron, and total iron-binding capacity [TIBC]) and 30-day all-cause mortality were assessed using the Kaplan-Meier curves, multivariable Cox regression, and restricted cubic splines. A TabPFN-based prediction model was developed and compared with 8 other machine-learning algorithms, with the least absolute shrinkage and selection operator used for feature selection. Model performance was evaluated using the area under the receiver operating characteristic curve (AUROC), area under the precision-recall curve (AUPRC), calibration, and decision curve analysis. SHapley Additive exPlanation values provided model interpretability.</p><p><strong>Results: </strong>Among 1137 patients in the MIMIC-IV cohort, 293 (26.0%) died within 30 days. Elevated ferritin (hazard ratio [HR] = 1.18, 95% CI, 1.09-1.27) and decreased TIBC (HR = 0.92, 95% CI, 0.85-0.99) were independently associated with mortality and exhibited non-linear relationships. Serum iron showed no prognostic value. The TabPFN model achieved the best performance, with an AUROC of 0.865 (95% CI, 0.812-0.919) in the internal validation set and 0.772 (95% CI, 0.744-0.814) in the external validation set, along with good calibration (Brier score = 0.121) and clinical net benefit. SHapley Additive exPlanation analysis identified lymphocyte count, platelet count, lactate, and ferritin as the most influential predictors.</p><p><strong>Conclusion: </strong>Iron metabolism dysregulation-particularly altered ferritin and TIBC levels-has important prognostic value in critically ill cancer patients. By integrating these biomarkers with a TabPFN framework, this study provides an accurate, interpretable, and externally validated tool for 30-day all-cause mortality prediction that may support clinical decision-making.</p>","PeriodicalId":36278,"journal":{"name":"JAMIA Open","volume":"9 4","pages":"ooag148"},"PeriodicalIF":4.7,"publicationDate":"2026-07-31","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://www.ncbi.nlm.nih.gov/pmc/articles/PMC13425414/pdf/","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"148654242","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"Real-world evaluation of a transformer-based natural language processing system for identifying social determinants of health from routine clinical documentation.","authors":"Xiangren Wang, Jingchuan Guo, Yi Guo, Yao An Lee, Mattia Prosperi, Jiang Bian, Noah Hammarlund","doi":"10.1093/jamiaopen/ooag147","DOIUrl":"10.1093/jamiaopen/ooag147","url":null,"abstract":"<p><strong>Objectives: </strong>To evaluate the real-world performance of a transformer-based natural language processing (NLP) system for extracting Social Determinants of Health (SDoH) from clinical notes, using survey-based SDoH measures a reference comparators.</p><p><strong>Materials and methods: </strong>This study was conducted at the University of Florida Health in adults with at least 2 clinical encounters in the prior year. A research survey was completed by 1001 participants; sampling targeted 50% Black patients to support subgroup analyses. Comparative analyses were restricted to the 414 participants who also had Epic SDoH survey data and clinical notes available for NLP extraction. We compared concepts extracted by the SOcial DeterminAnts (SODA) NLP pipeline against the independently administered research survey, which served as the primary reference standard, and against the structured Epic-embedded SDoH questionnaire. Nine domains were evaluated: abuse, alcohol use, drug use, education, financial constraints, housing, physical activity, social cohesion, and transportation. Sensitivity, specificity, positive predictive value, negative predictive value, and F1 scores were calculated by domain.</p><p><strong>Results: </strong>The NLP pipeline more consistently aligned with negative survey responses than with patient-reported social needs, although performance was lower in some domains, especially alcohol use and financial constraints. Sensitivity was higher only for alcohol use (55%); the lowest values were for abuse (5%), drug use (0%), and financial constraints (16%). These results cannot be attributed to SODA extraction alone. They reflect some combination of social information not being recorded in clinical notes, content that was recorded but not extracted, and mismatches between extracted concepts and survey definitions, and the present analysis cannot separate these contributions. The 2 surveys agreed only modestly with each other, so no single instrument provides a definitive ground truth.</p><p><strong>Conclusion: </strong>The pipeline more consistently aligned with negative survey responses than with patient-reported social needs. Because the 2 surveys agreed only modestly, the reference itself is imperfect, and apparent NLP performance depends in part on which survey is used as the comparator. Apparent gaps in NLP performance reflect both how social risks are recorded in clinical notes and how patients disclose them across different survey settings, in addition to limits of the extraction pipeline. Improving documentation practices, integrating locally tuned large language models, and monitoring subgroup performance may all be needed to make SDoH identification tools reliably detect social needs across patient populations.</p>","PeriodicalId":36278,"journal":{"name":"JAMIA Open","volume":"9 4","pages":"ooag147"},"PeriodicalIF":4.7,"publicationDate":"2026-07-31","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://www.ncbi.nlm.nih.gov/pmc/articles/PMC13427761/pdf/","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"148654309","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"Comparison of classification models for adolescent's emotional and behavioral problems.","authors":"Akhyt Tilyeubai, Javzmaa Tsend, Nyamdavaa Uugandavaa, Bayarmaa Vaanchindorj, Purevdolgor Luvsantseren, Ajnai Luvsan-Ish, Baasandorj Chilkhaasuren, Galbadrakh Chuluunbaatar, Enkh-Urel Enkhjargal","doi":"10.1093/jamiaopen/ooag144","DOIUrl":"10.1093/jamiaopen/ooag144","url":null,"abstract":"<p><strong>Objectives: </strong>Mental disorders are very common among children and adolescents around the world. In Mongolia, the international Strengths and Difficulties Questionnaire (SDQ) is used to detect mental disorders in adolescents. In this article, different types of classification methods were compared to determine adolescent emotional and behavioral problems using the SDQ.</p><p><strong>Materials and methods: </strong>Data were collected from teenagers, teachers, and parents in Govi-Altai Province, and the databases were created for each group. The teenager database was divided into 10 folds using cross-validation, and the models were developed using classification methods and evaluated using performance measures. The results were mainly analyzed using the Bayes model.</p><p><strong>Results: </strong>The teenagers have emotional and behavioral problems due to emotional and peer interactions, but they were at risk of developing disorders due to hyperactivity and behavioral changes.</p><p><strong>Conclusion: </strong>Future work will examine the difficulties encountered in the creation of emotional and behavioral problems among the teenagers involved in the study using a progressive classification method.</p>","PeriodicalId":36278,"journal":{"name":"JAMIA Open","volume":"9 4","pages":"ooag144"},"PeriodicalIF":4.7,"publicationDate":"2026-07-30","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://www.ncbi.nlm.nih.gov/pmc/articles/PMC13424436/pdf/","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"148654267","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"Large language model-based extraction of rheumatoid arthritis clinical disease activity index from clinical notes.","authors":"Reid Weisberg, Ruoqi Yang, Iram Kamdar, Justin Chestang, Joan Bathon, Jon T Giles, Chunhua Weng, Elizabeth Park","doi":"10.1093/jamiaopen/ooag152","DOIUrl":"10.1093/jamiaopen/ooag152","url":null,"abstract":"<p><strong>Objectives: </strong>We extracted a validated disease activity measure in rheumatoid arthritis (RA), the Clinical Disease Activity Index (CDAI), from a large tertiary academic medical center electronic health record (EHR) using an automated large language model (LLM)-based approach without requiring model pretraining.</p><p><strong>Materials and methods: </strong>The New York Presbyterian/Columbia University Medical Center Clinical Data Warehouse contains EHR data for over 4.5 million patients. RA patients were identified using International Classification of Disease-9 (ICD-9) and ICD-10 codes. Expert-curated CDAI keywords were extracted from unstructured notes using an automated natural language processing (NLP) pipeline leveraging GPT-4o API, a HIPAA-compliant, institutionally approved LLM platform. Performance was evaluated against expert chart review.</p><p><strong>Results: </strong>Among 2756 RA patients with notes, 1038 (37.7%) were seropositive, 796 (28.9%) were seronegative, and 922 (33.4%) had unknown serostatus. Clinical Disease Activity Index and its components were extracted in 15.4% (160/1038) of seropositive patients indicating remission or low disease activity. Clinical Disease Activity Index documentation was more frequent among patients with multiple notes and among faculty, with high extraction accuracy (precision/recall/F1 = 0.97).</p><p><strong>Discussion: </strong>This represents the first attempt to employ a zero-shot, ChatGPT-powered LLM platform to extract RA disease activity measures from real-world EHR data. Although a low prevalence of documentation was noted, important distinctions were observed when patients were subgrouped by serostatus, level of training, and number of visits.</p><p><strong>Conclusion: </strong>An LLM-based pipeline accurately extracted CDAI from a single large academic EHR, revealing infrequent real-world documentation.</p>","PeriodicalId":36278,"journal":{"name":"JAMIA Open","volume":"9 4","pages":"ooag152"},"PeriodicalIF":4.7,"publicationDate":"2026-07-30","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://www.ncbi.nlm.nih.gov/pmc/articles/PMC13424428/pdf/","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"148654256","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"From fragmented records to living evidence: health system-governed, artificial intelligence-driven, continuously updated real-world clinical data from Truveta.","authors":"Hannah A Burkhardt, Sarah J Blach, Srinivasa R Burugapalli, Brianna M G Cartwright, Ryley Martin, Michael Simonov, Sarah Stewart, Grace Turner, Angela L Winegar, Nicholas Stucky","doi":"10.1093/jamiaopen/ooag142","DOIUrl":"10.1093/jamiaopen/ooag142","url":null,"abstract":"<p><strong>Objectives: </strong>Real-world data (RWD) have historically suffered from fragmentation, delayed availability, variable data quality, and limited analytic utility. Truveta developed an artificial intelligence (AI)-enabled data platform to address these longstanding challenges in using RWD for clinical research. This paper describes Truveta's partnership model, platform design, data scale, and research applications.</p><p><strong>Materials and methods: </strong>Truveta de-identifies, aggregates, and harmonizes electronic health record (EHR) data for 130 million patients-1 in 3 Americans-from US health systems. The platform links structured and unstructured EHR content with closed claims, mortality, and social determinants of health. Data undergo daily ingestion, normalization to standard ontologies, and de-identification. Advanced AI, including NLP, extracts key clinical concepts from free-text notes, such as physician notes, imaging reports, and pathology narratives, transforming them into standardized variables suitable for large-scale observational research.</p><p><strong>Results: </strong>Truveta Data comprise over 130 million de-identified patient records that are updated daily and represent diverse geographic regions, care settings, and patient populations in the United States. The data have supported over 100 scientific publications to date; additionally, they support health system participants' own research interests and patient care insights. Published studies have addressed treatment effectiveness, post-market device surveillance, COVID-19 vaccine safety, and health equity.</p><p><strong>Discussion: </strong>Truveta addresses critical barriers that have hindered the realization of a learning health system. Unlike prior RWD initiatives limited by scope, latency, and vendor dependence, Truveta enables near-real-time, population-scale research grounded in rich clinical data. Its governance model ensures alignment with ethical, privacy, and regulatory standards. By rethinking the role of health systems from passive suppliers into active, incentivized partners, Truveta creates an unprecedented virtuous cycle of data quality and continuous improvement.</p><p><strong>Conclusion: </strong>Truveta represents a paradigm shift in real-world evidence generation. By aligning incentives and AI-driven harmonization, it provides a scalable, sustainable infrastructure for continuously updated clinical data. Providing large-scale, high-fidelity EHR data and daily updates, the platform accelerates clinical discovery, policy and public health decision-making, and improved patient outcomes.</p>","PeriodicalId":36278,"journal":{"name":"JAMIA Open","volume":"9 4","pages":"ooag142"},"PeriodicalIF":4.7,"publicationDate":"2026-07-29","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://www.ncbi.nlm.nih.gov/pmc/articles/PMC13418207/pdf/","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"148621644","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"Correction to: Measurement, drivers, and outcomes of patient-initiated secure messaging use and intensity: a scoping review.","authors":"","doi":"10.1093/jamiaopen/ooag153","DOIUrl":"https://doi.org/10.1093/jamiaopen/ooag153","url":null,"abstract":"<p><p>[This corrects the article DOI: 10.1093/jamiaopen/ooaf087.].</p>","PeriodicalId":36278,"journal":{"name":"JAMIA Open","volume":"9 4","pages":"ooag153"},"PeriodicalIF":4.7,"publicationDate":"2026-07-28","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://www.ncbi.nlm.nih.gov/pmc/articles/PMC13509951/pdf/","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"148833214","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"Understanding end-user contexts and identifying design preferences of an artificial intelligence-based clinical decision support tool for early autism detection.","authors":"Adesuwa Emovon, Lauren Driggers-Jones, Matthew Engelhard, Gary Maslow, Geraldine Dawson, Benjamin A Goldstein, Lauren Franz","doi":"10.1093/jamiaopen/ooag145","DOIUrl":"10.1093/jamiaopen/ooag145","url":null,"abstract":"<p><strong>Objectives: </strong>Building on innovations for autism detection-where artificial intelligence (AI)-based models monitor clinical data within electronic health records-this study evaluates the context for clinical decision support (CDS) deployment and identifies design preferences.</p><p><strong>Materials and methods: </strong>This observational study utilized contextual inquiry to elicit perspectives from 8 clinicians and twenty caregivers during 18- to 24-month well-child visits at Duke-affiliated clinics. Data were analyzed using rapid qualitative analysis techniques.</p><p><strong>Results: </strong>Workflow analysis identified 6 user tasks, 3 technology-user interactions, and 5 clinical decision points. Technologies that streamlined screening included patient portals, digital tablets, and note templates. Clinicians identified 2 major barriers-limited screening tool accuracy and challenges in implementing follow-up steps-and 3 facilitators: electronic screening, early intervention provider input, and staff referral coordination support. For design, CDS should include clear, actionable outputs, with explanations of prediction data, visual summaries linked to next steps, and educational resources. Embedding CDS within the EHR, with outputs delivered at key points during the clinical encounter, along with caregiver-facing materials, would improve workflow efficiency.</p><p><strong>Discussion: </strong>Findings highlight key integration points for an autism detection AI-based CDS tool and stress the need for clinical utility and caregiver-centered communication. Effective design requires alignment with clinical workflow, including the timing of outputs, meaningful explanations, and integration with caregiver communication.</p><p><strong>Conclusion: </strong>Findings will inform the design of an AI-based CDS tool for autism detection, providing workflow-informed integration points and user preferences. Future work should refine explainability and optimize delivery of outputs within clinical encounters to support decision-making and caregiver engagement.</p>","PeriodicalId":36278,"journal":{"name":"JAMIA Open","volume":"9 4","pages":"ooag145"},"PeriodicalIF":4.7,"publicationDate":"2026-07-23","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://www.ncbi.nlm.nih.gov/pmc/articles/PMC13394496/pdf/","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"148580544","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"Development of a clinical trial knowledge management application for community oncology.","authors":"Elizabeth O'Brien, Aanshi Thumar, Melanie Manning, Krishna Gunturu, Peter Yu, Tony K W Hung","doi":"10.1093/jamiaopen/ooag140","DOIUrl":"10.1093/jamiaopen/ooag140","url":null,"abstract":"<p><strong>Introduction: </strong>Participation in cancer clinical trials is low in community oncology settings, partly because institution-specific trial information is fragmented. We evaluated feasibility of embedding curated trial content in an AI-enabled knowledge management application.</p><p><strong>Materials and methods: </strong>At a regional community oncology network, coordinators and disease teams compiled actively recruiting trials. Core elements (title, conditions, biomarkers, stage/line, and recruiting status) were structured for point-of-care display and uploaded. AI-assisted extraction generated protocol summaries and eligibility elements, which underwent systematic human validation.</p><p><strong>Results: </strong>Fifty-three trials across 10 disease groups were embedded and validated; 91% were recruiting. Trials covered 28 cancer types; 30% were biomarker-specific and most enrolled advanced/metastatic disease. Initial configuration took 2-4 weeks per disease group using existing personnel, without added staffing or electronic health record (EHR) build.</p><p><strong>Discussion: </strong>Embedding institution-specific trial content within an AI-enabled knowledge application is feasible in community oncology using existing clinical and research infrastructure, establishing a prerequisite for future usability and implementation studies.</p>","PeriodicalId":36278,"journal":{"name":"JAMIA Open","volume":"9 4","pages":"ooag140"},"PeriodicalIF":4.7,"publicationDate":"2026-07-22","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://www.ncbi.nlm.nih.gov/pmc/articles/PMC13390100/pdf/","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"148562855","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"From clinics to discoveries: harnessing national EHR interoperability for research.","authors":"Josh Lemieux, Emily R Pfaff, Richard A Moffitt, Jay Nakashima, Raymond Duncan, Erica Galvez, John P Giannini, Julie A McMurry, Anita Walden, Melissa Haendel","doi":"10.1093/jamiaopen/ooag137","DOIUrl":"10.1093/jamiaopen/ooag137","url":null,"abstract":"<p><strong>Objectives: </strong>To assess whether existing nationwide Health Information Networks (HINs) and Health Information Exchanges (HIEs) can be leveraged for electronic health record (EHR) data acquisition for research-specifically, for the NIH's <i>All of Us</i> Research Program.</p><p><strong>Materials and methods: </strong>The <i>All of Us</i> Center for Linkage and Acquisition of Data (CLAD) collaborated with eHealth Exchange, the nation's largest HIN, to transmit participant-authorized queries to an HIE and a hospital system. Returned FHIR and C-CDA records were mapped to the OMOP Common Data Model and compared with existing <i>All of Us</i> EHR data.</p><p><strong>Results: </strong>Newly obtained records provided complementary information, enhancing completeness and research utility.</p><p><strong>Discussion: </strong>This pilot demonstrates the feasibility of using HINs/HIEs for patient-authorized data exchange for research. Barriers remain-including inconsistent technical capacity to transact authorizations and data quality variations.</p><p><strong>Conclusion: </strong>HINs and HIEs can support secure, patient-authorized EHR exchange for research, offering a pathway to more comprehensive, participant-driven biomedical discovery.</p>","PeriodicalId":36278,"journal":{"name":"JAMIA Open","volume":"9 4","pages":"ooag137"},"PeriodicalIF":4.7,"publicationDate":"2026-07-22","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://www.ncbi.nlm.nih.gov/pmc/articles/PMC13390099/pdf/","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"148563093","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"Sequential multi-site fine-tuning for incremental deployment of large language models for mobility functional status extraction.","authors":"Xingyi Liu, Muskan Garg, Eunji Jeon, Heling Jia, Cynthia S Crowson, Jennifer St Sauver, Sandeep R Pagali, Sunghwan Sohn","doi":"10.1093/jamiaopen/ooag139","DOIUrl":"10.1093/jamiaopen/ooag139","url":null,"abstract":"<p><strong>Objective: </strong>This study evaluated sequential multi-site fine-tuning of large language models (LLMs), simulating incremental deployment across institutions for extracting mobility functional status from unstructured clinical notes.</p><p><strong>Materials and methods: </strong>We assembled a corpus of 600 clinical notes from 3 institutions, yielding 3810 annotated note sections (1200 training; 2610 testing). Building upon our previously established LLM mobility extraction pipeline, which utilized optimized prompt engineering, we applied parameter-efficient low-rank adaptation (LoRA) fine-tuning. We evaluated sequential multi-site fine-tuning, where the model is trained locally and only the updated LoRA weights are transferred to the next site. We compared this against joint fine-tuning, which requires pooling all multi-site patient data centrally before training. Performance was measured by micro F1-scores for 2 distinct subtasks: mobility extraction and impairment classification.</p><p><strong>Results: </strong>Sequential multi-site fine-tuning yielded comparable results to centralized joint fine-tuning for mobility extraction (F1, 0.861 vs 0.894) and impairment classification (F1, 0.912 vs 0.900). Both sequential and joint fine-tuning outperformed an untuned 70B model. Additionally, single-site LoRA models showed robust cross-site generalization.</p><p><strong>Discussion: </strong>Sequential multi-site fine-tuning offers a resource-friendly approach to adapt LLMs to clinical environments. By retaining previously acquired knowledge without severe interference, this strategy allows for local fine-tuning at each institution without the need to pool or share sensitive patient data required for joint fine-tuning.</p><p><strong>Conclusion: </strong>Sequential multi-site fine-tuning provides a viable solution for the scalable, incremental deployment of LLMs across institutions, showing promising generalizability for clinical information extraction from electronic health record narratives.</p>","PeriodicalId":36278,"journal":{"name":"JAMIA Open","volume":"9 4","pages":"ooag139"},"PeriodicalIF":4.7,"publicationDate":"2026-07-20","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://www.ncbi.nlm.nih.gov/pmc/articles/PMC13384054/pdf/","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"148608351","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}