Journal Of The American Medical Informatics Association[JOURNAL]
Sun
200 papers
Characterizing emergency clinician engagement with social drivers of health data among patients with opioid use disorder.
Molina MF, Pimentel SD, Fenton C … +2 more , Adler-Milstein J, Gottlieb LM
J Am Med Inform Assoc · 2026 Jun · PMID 42310717 · Publisher ↗
OBJECTIVES: To characterize emergency department (ED) clinician engagement with electronic health record (EHR)-based social drivers of health (SDOH) data; test whether engagement differs in encounters with opioid use dis...
OBJECTIVES: To characterize emergency department (ED) clinician engagement with electronic health record (EHR)-based social drivers of health (SDOH) data; test whether engagement differs in encounters with opioid use disorder (OUD); and, among OUD encounters, assess whether engagement is associated with medications for OUD (MOUD). MATERIALS AND METHODS: We conducted a cross-sectional study of adult ED encounters (January 2023-October 2024). OUD encounters, identified with a structured phenotype, were matched (1:2) to non-OUD encounters. Audit logs captured clinician engagement with structured SDOH questions ("SDOH Wheel"), ICD-10 Z codes in the Problem List, Social History free text, and social work notes. Engagement was any SDOH documentation or review of preexisting SDOH data during the encounter. Logistic regression estimated associations. RESULTS: Among 17 103 encounters (5701 OUD; 11 402 non-OUD), clinician SDOH documentation was rare (<1%). Clinicians most often reviewed the SDOH Wheel (1103/3953; 27.9%), followed by social work notes (1711/10 670; 16.0%), Z codes (29/620; 4.7%), and Social History free text (232/6942; 3.3%). Engagement occurred in 17.1% of encounters and was higher with OUD (23.1% vs 14.1%; OR 1.84, 95% CI 1.70-2.00). Among OUD encounters, engagement was not associated with MOUD (OR 1.27, 95% CI 0.95-1.70); however, MOUD treatment varied by race and ethnicity, reflecting persistent disparities. DISCUSSION: ED clinicians infrequently documented and favored review of structured, accessible SDOH data. Engagement increased in OUD encounters yet neither predicted MOUD nor mitigated racial and ethnic treatment disparities. EHR interfaces that surface SDOH data, coupled with targeted decision supports, might influence equitable, time-sensitive ED care.
Large language models for full-text methods assessment: a case study on mediation analysis.
Zhang W, Nguyen T, Stuart EA … +1 more , Chen YT
J Am Med Inform Assoc · 2026 Jun · PMID 42308515 · Publisher ↗
OBJECTIVE: Systematic reviews remain labor-intensive, particularly when extracting methodological details from full texts. Using mediation analysis as a case study, we evaluated whether large language models (LLMs) can m...
OBJECTIVE: Systematic reviews remain labor-intensive, particularly when extracting methodological details from full texts. Using mediation analysis as a case study, we evaluated whether large language models (LLMs) can match human-expert-level full-text methodological review on key causal assumptions (eg, no unmeasured confounding, temporal ordering) and best practices (eg, sensitivity analyses, interaction assessments, covariate adjustment) for psychiatry and psychology studies. MATERIALS AND METHODS: We evaluated 6 LLMs from 3 major families (ChatGPT-4o-mini/4o/o3/5, Claude Sonnet 4, Gemini 2.5 Flash) on 180 full-text mediation analysis articles from 2013 to 2018 previously reviewed by expert methodologists. LLMs assessed 14 binary methodological criteria ranging from straightforward checks (eg, whether the exposure was randomized) to nuanced assessments (eg, whether the temporal ordering between mediator and outcome was established). Performance was benchmarked against expert consensus labels and individual reviewers using accuracy, precision, recall, F1, AUC, and PR-AUC. RESULTS: LLM performance strongly correlated with human reviewers across methodological criteria (accuracy correlation 0.71; F1 correlation 0.95), indicating tasks difficult for humans were likewise challenging for models. Advanced LLMs achieved near-human accuracy on explicit methodological features but lagged behind top reviewers by up to 15% on inference-intensive tasks. Longer documents reduced model accuracy. Common model errors include overinterpreting on linguistic cues and colloquial use of technical terms. DISCUSSION AND CONCLUSION: Our findings support a criterion-specific human-AI collaboration strategy for full-text methodological assessment and provide a reproducible framework for future testing in other evidence-synthesis settings.
AI, humanity, and the open world of health care: enduring imperatives for the next century.
Horvitz E
J Am Med Inform Assoc · 2026 Jun · PMID 42308013 · Publisher ↗
This address was delivered by Eric Horvitz, MD, PhD, at the 2026 graduation ceremony of Columbia University School of Nursing on May 19, 2026, where he received the Second Century Award for Excellence in Health Care. The...
This address was delivered by Eric Horvitz, MD, PhD, at the 2026 graduation ceremony of Columbia University School of Nursing on May 19, 2026, where he received the Second Century Award for Excellence in Health Care. The address considers the responsibilities of clinicians in shaping the future of artificial intelligence in medicine. It frames health care as an "open world," where information is incomplete, time is limited, and decisions are made under uncertainty. As AI transforms biomedicine and clinical care, the address emphasizes the importance of clinician engagement in guiding how these technologies are developed and used, and calls for systems that strengthen clinical judgment, support care teams, and advance human health, dignity, connection, and trust.
Privacy-enhancing sequential learning under heterogeneous selection bias in multi-site electronic health records data.
Kundu R, Salvatore M, Patel KK … +4 more , Ohno-Machado L, Cho H, Shi X, Mukherjee B
J Am Med Inform Assoc · 2026 Jun · PMID 42298300 · Publisher ↗
OBJECTIVES: To develop privacy-enhancing statistical methods for estimating disease risk parameters across multiple electronic health record (EHR) sites with heterogeneous selection mechanisms, avoiding individual-level...
OBJECTIVES: To develop privacy-enhancing statistical methods for estimating disease risk parameters across multiple electronic health record (EHR) sites with heterogeneous selection mechanisms, avoiding individual-level data sharing. We illustrate their utility via a cross-biobank analysis of smoking and 97 cancer subtypes using NIH All of Us (AOU) and Michigan Genomics Initiative (MGI) data sites. MATERIALS AND METHODS: Distributed health platforms often render centralized algorithms infeasible due to patient privacy protection. We propose Sequential Pseudo-Likelihood (SPL) and Sequential Augmented Inverse Probability Weighting (SAIPW) to adjust for selection bias using summary statistics shared across sites and external population information. SAIPW employs flexible auxiliary models for multiple robustness. We compared SPL and SAIPW against unweighted and centralized/meta-learning benchmarks in simulations, applying them to harmonized MGI (n = 50 935) and AOU (n = 241 563) data. RESULTS: Unweighted estimators exhibited substantial bias. SPL and SAIPW yielded unbiased estimates with valid coverage, with SAIPW remaining robust to selection model misspecification. Both approaches showed negligible efficiency loss relative to centralized methods. Meta-learning methods proved unstable for rare outcomes. Real-data analyses consistently identified strong associations between smoking and lung, bladder, and larynx cancers. DISCUSSION: These findings highlight the necessity of adjusting for site-specific selection biases in distributed health networks. SPL and SAIPW offer practical, scalable solutions that bypass the instability of meta-analysis for rare events, successfully harmonizing diverse biobanks while strictly enhancing patient privacy. CONCLUSION: Our framework enables valid, privacy-enhancing inference across EHR sites subject to heterogeneous selection, facilitating scalable, distributed research using real-world data.
Augmenting clinical decision-making with large language models: evaluation across general and specialty tasks.
Tao Y, Lin X, Sun H … +16 more , Bao W, Zhu X, Wu S, Liu Y, Li Z, Wu J, Zhao Z, Qiu L, Ni Q, Zhang S, Tang Z, Ying Z, An Y, Tian J, Liu Z, Lu J
J Am Med Inform Assoc · 2026 Jun · PMID 42289821 · Publisher ↗
OBJECTIVE: Large language models (LLMs) have shown promise in clinical applications, yet prior studies mainly evaluated their standalone performance on benchmarks or examinations. This study assesses how LLMs support cli...
OBJECTIVE: Large language models (LLMs) have shown promise in clinical applications, yet prior studies mainly evaluated their standalone performance on benchmarks or examinations. This study assesses how LLMs support clinicians across specialties, disease contexts, experience levels and decision-making stages. METHODS: We evaluated 3 LLMs (Deepseek-R1, GPT-4o-mini and LLaMA-4) using 2 task types: (1) general-disease tasks spanning specialties and disease incidence levels, and (2) prostate disease scenario covering the clinical workflow, including diagnosis, treatment planning, postoperative rehabilitation and prognosis. Clinicians with different seniority completed tasks independently and repeated them after reviewing LLM-generated responses. Performance was rated by experts using a 5-point Likert scale, and differences were analyzed with the Wilcoxon signed-rank test. RESULTS: LLM assistance significantly improved clinician performance across general disease tasks (P < .05). In the specialty scenario, junior clinicians' scores increased by 0.579-0.723 (15.9%-20.8%) across 4 clinical stages, whereas senior clinicians' improvements ranged from 0.053-0.303 (1.3%-7.5%). With LLM support, junior clinicians surpassed senior clinicians' unaided performance in postoperative rehabilitation and prognosis stages (P < .05) and achieved comparable performance in diagnosis and treatment stages (P > .05). Performance varied among models, with Deepseek-R1 performing best in diagnostic tasks. DISCUSSION: These findings suggest that LLM assistance may provide greater benefit to less-experienced clinicians and are particularly effective in downstream decision-making stages, indicating a potential role in mitigating experience-related performance disparities. CONCLUSIONS: LLMs may serve as supportive clinical-decision support tools across diseases, specialties, clinician experience levels, and workflow stages, with observed improvements in decision quality and reduced performance disparities.
A national survey of clinical informatics educational opportunities in United States medical schools.
Lee GY, Li K, Patil A … +5 more , Jiang S, Zheng C, Baral J, You J, Kim E
J Am Med Inform Assoc · 2026 Jun · PMID 42289820 · Publisher ↗
OBJECTIVE: To assess prevalence, characteristics, and institutional predictors of clinical informatics (CI) education and student organizations in US allopathic (MD) and osteopathic (DO) medical schools. MATERIALS AND ME...
OBJECTIVE: To assess prevalence, characteristics, and institutional predictors of clinical informatics (CI) education and student organizations in US allopathic (MD) and osteopathic (DO) medical schools. MATERIALS AND METHODS: We reviewed 222 US medical schools from the 2024 Medical School Admission Requirements (MSAR) and American Association of Colleges of Osteopathic Medicine (AACOM) databases. Using predefined criteria, we identified CI-related courses and student groups and abstracted institutional characteristics including affiliated CI fellowships. Bivariate and multivariable logistic regression identified predictors. RESULTS: Of 222 schools, 30.2% offered at least one CI course and 23.0% had a student group. In bivariate analyses, MD programs and institutions with CI fellowships were significantly more likely to offer courses (both P < 0.001). In multivariable analyses, MD program type was the strongest predictor (adjusted odds ratio [aOR]=6.58, 95% confidence interval [CI] 2.22-22.41), followed by CI fellowship presence (aOR = 3.36), private school status (aOR = 2.08), and class size (aOR = 1.01). All 51 student groups were at MD institutions, and urban setting was associated with group presence (P = 0.034). DISCUSSION: The association with CI fellowships suggests a relationship between graduate and undergraduate medical education institutions. The association with MD programs could be influenced by differing curricular demands and flexibility. The association with urban settings may reflect the role of local innovation ecosystems. CONCLUSION: CI educational opportunities vary across US medical schools, concentrated at MD programs and institutions with GME-level infrastructure. These findings establish a baseline for CI opportunities and highlight the need to understand whether institutional differences translate to measurable competency gaps.
Let large language models judge each other: multi-agent peer-reviewed reasoning for medical question answering.
Zhan Z, Zhou S, Zhang R
J Am Med Inform Assoc · 2026 Jun · PMID 42289818 · Publisher ↗
OBJECTIVE: To enhance the accuracy, interpretability, and robustness of large language models (LLMs) in medical question answering (MedQA). MATERIALS AND METHODS: We designed a multi-agent peer-reviewed reasoning method...
OBJECTIVE: To enhance the accuracy, interpretability, and robustness of large language models (LLMs) in medical question answering (MedQA). MATERIALS AND METHODS: We designed a multi-agent peer-reviewed reasoning method in which multiple LLM agents independently generate chain-of-thought (CoT) reasoning with candidate answers, then act as peer reviewers to evaluate each other's reasoning for factual correctness and logical soundness. The highest-rated reasoning chain is selected to produce the final answer. Experiments were conducted with 5 state-of-the-art LLMs (Llama-3.1-8B, Qwen2.5-7B, Phi-4, DeepSeek-LLM-7B, and GPT-oss-20B) on 3 benchmark datasets: HeadQA, MedQA-USMLE, and PubMedQA. Performance was compared against single-model CoT reasoning and CoT-based majority voting. RESULTS: Peer-reviewed reasoning consistently outperformed both baselines. The best model combination achieved an average accuracy of 0.820 across datasets, exceeding the strongest single model (0.777) and majority voting ensembles (up to 0.789). The method also scaled effectively with more participating models, while peer assessments reliably distinguished high- from low-quality reasoning chains. CONCLUSION: The proposed multi-agent peer-reviewed reasoning method enables LLMs to act as both solvers and evaluators, yielding superior performance in MedQA. By emphasizing reasoning quality rather than answer agreement alone, this approach improves accuracy, interpretability, and robustness, offering a promising direction for trustworthy biomedical AI systems.
Large language model biases in health care: a scoping review and call for an integrated assessment framework.
He L, Do DP, Shet VG … +5 more , Farghaly O, Deshpande P, Madiraju P, Ye J, Beestrum M
J Am Med Inform Assoc · 2026 Jun · PMID 42287648 · Publisher ↗
OBJECTIVES: To conduct a scoping review of bias assessment in studies applying large language models (LLMs) to health data and to synthesize their prevailing conceptualization of bias. MATERIALS AND METHODS: Following PR...
OBJECTIVES: To conduct a scoping review of bias assessment in studies applying large language models (LLMs) to health data and to synthesize their prevailing conceptualization of bias. MATERIALS AND METHODS: Following PRISMA guidelines, we queried PubMed and Scopus. Two annotators screened titles, abstracts, and full texts for eligibility, calibrating their assessments throughout the process. For included studies, we extracted and summarized data on LLMs (name and version, development domain, open- or closed-sourced status, and commercial or academic origin), natural language processing tasks (task formulation, gold-standard dataset, evaluation metrics, prompting or fine-tuning strategies), and biases (type, assessment, and bias summary). RESULTS: Of the 1585 records retrieved, 76 papers met the eligibility criteria for full review. Among these, 59 reported identifying bias. Three major conceptualizations of bias emerged: behavioral output bias (nonstereotyping and stereotyping), predictive outcome bias, and representational bias. Studies generally adopted an observational approach (measuring bias using the existing dataset) or an experimental approach (altering prompts, eg, with different demographic information, and comparing outputs). DISCUSSION AND CONCLUSION: Behavioral output bias and predictive outcome bias, both of which emphasize parity, dominate existing studies. Whether evaluated against external accuracy or internal equality benchmarks, these approaches often assume that equal performance across groups is inherently desirable. Treating all disparities as bias risks conflating poor model behavior with real-world disparities, and researchers should remain aware of potential tradeoffs between parity and accuracy objectives. We introduce an integrated framework that combines parity and accuracy benchmarks and encourages transparent, context-aware interpretation of group differences.
Reasoning or reciting? A temporal contamination audit of large language models in clinical medicine.
Sheppert AP, Adams B, Sheppert AD … +1 more , Riley S
J Am Med Inform Assoc · 2026 Jun · PMID 42281378 · Publisher ↗
OBJECTIVE: Evaluate whether large language models reason or simply regurgitate training data in clinical diagnosis. MATERIALS AND METHODS: We audited 2000 clinical case reports from PubMed Central: 1000 from 2021 to 2022...
OBJECTIVE: Evaluate whether large language models reason or simply regurgitate training data in clinical diagnosis. MATERIALS AND METHODS: We audited 2000 clinical case reports from PubMed Central: 1000 from 2021 to 2022 (within training data) and 1000 from 2025 (after training cutoffs). Five frontier LLMs generated diagnoses evaluated by an independent AI judge validated against physician consensus (n = 10 000 evaluations). RESULTS: Diagnostic accuracy was virtually identical across temporal cohorts (66.8% contaminated vs 66.9% clean), directly contradicting the memorization hypothesis. Lexical similarity was uniformly low (mean ROUGE-L 0.057), and semantic similarity measured by BERTScore showed no memorization signal (F1 0.8182 contaminated vs 0.8195 clean, Δ = +0.0013), confirming that models generate novel reasoning rather than regurgitating training data. DISCUSSION: This large-scale audit, using both lexical and semantic similarity metrics, provides compelling evidence that LLMs engage in genuine clinical reasoning rather than regurgitating memorized training data. CONCLUSION: Models demonstrated equivalent accuracy on cases they could not have seen during training, suggesting they have internalized generalizable medical knowledge rather than memorizing specific cases.
The GUIDE-CDSS framework: a comprehensive framework to GUIDE and evaluate the implementation of Clinical Decision Support Systems in healthcare, based on an overview of reviews.
Janssen ERC, Knoop J, Farabolini G … +6 more , Baldini N, Hoogeboom T, Ceravolo MG, Negrini S, Kiekens C, PREPARE Project Group
J Am Med Inform Assoc · 2026 Jun · PMID 42281367 · Publisher ↗
OBJECTIVES: Despite the significant potential for Clinical Decision Support Systems (CDSSs) to improve care processes and health outcomes, several barriers hinder their widespread implementation in healthcare. While nume...
OBJECTIVES: Despite the significant potential for Clinical Decision Support Systems (CDSSs) to improve care processes and health outcomes, several barriers hinder their widespread implementation in healthcare. While numerous systematic reviews have summarized potential barriers and facilitators for CDSS implementation, a comprehensive framework to guide and evaluate the implementation of CDSSs in healthcare is lacking. This overview of reviews, aims to establish a framework-GUIDE-CDSS-aimed at guiding and evaluating implementation of CDSSs in healthcare. MATERIALS AND METHODS: An overview of systematic and scoping reviews was conducted by searching 6 databases. Systematic reviews or scoping reviews that used qualitative research methods to described implementation determinants for CDSSs in the healthcare domain were included. The AMSTAR 2 tool was used to assess the methodological quality. Results were collated into the GUIDE-CDSS framework. This framework describes implementation determinants and elements within those determinants found to impact implementation of CDSSs in healthcare. RESULTS: Twenty-three reviews were included in the analysis. All reviews had at least 2 critical weaknesses, showing a limited methodological quality of included reviews. Eight determinants and 38 elements for implementation of CDSSs in healthcare and were described in the GUIDE-CDSS framework: perceived relevance, perceived effect, trustworthiness, ease of use, workflow, training and skills, resources, and implementation strategy. DISCUSSION AND CONCLUSION: This overview provides a comprehensive synthesis of the determinants influencing the implementation of CDSSs in healthcare, collated in the GUIDE-CDSS framework. The findings underscore that for successful CDSS development, implementation and evaluation is multifactorial. PROTOCOL REGISTRATION: This study was registered in PROSPERO (No. CRD42024512455).
The Big Mo: staying on the wave in an age of artificial intelligence.
Johnson KB
J Am Med Inform Assoc · 2026 Jun · PMID 42277441 · Publisher ↗
Dr. Kevin B. Johnson delivered this address on May 16, 2026, at the Commencement Ceremony of The D. Bradley McWilliams School of Biomedical Informatics, UTHealth Houston, to the graduating class of 2026. The address uses...
Dr. Kevin B. Johnson delivered this address on May 16, 2026, at the Commencement Ceremony of The D. Bradley McWilliams School of Biomedical Informatics, UTHealth Houston, to the graduating class of 2026. The address uses the concept of "The Big Mo" (compounding momentum) as a frame for understanding the current inflection point in AI and medicine. Drawing on his own career arc from paper-based clinical practice at Johns Hopkins through early adoption of health informatics to the present era of AI in healthcare, Johnson argues that the fears graduates hold about technological obsolescence and institutional instability are real but misdirected. He reframes both: biomedical informatics professionals are not targets of AI but its essential architects, and the external environment has always been uncertain for those doing important work. His charge to graduates is singular: stay on the wave.
From exchange to action: how hospitals receive and use health-related social needs data from external sources.
Richwine C, Chang W, Iott B
J Am Med Inform Assoc · 2026 Jun · PMID 42276595 · Publisher ↗
OBJECTIVE: To understand how hospitals receive and use data on patients' health-related social needs (HRSN) from external sources, and how methods of receipt relate to use. MATERIALS AND METHODS: Using 2024 nationally re...
OBJECTIVE: To understand how hospitals receive and use data on patients' health-related social needs (HRSN) from external sources, and how methods of receipt relate to use. MATERIALS AND METHODS: Using 2024 nationally representative survey data on hospitals (N = 2146) we described external sources of receiving HRSN data and variation by hospital characteristics, and how methods of receipt relate to uses of data to support patient care. RESULTS: In 2024, 70% of hospitals received HRSN data from external sources-via health information exchanges (HIEs) (51%), electronic health record (EHR) vendor networks (40%), national networks (36%), referral platforms (28%), other healthcare organizations (24%) and community-based organizations (20%). Large, urban, system-affiliated, non-profit and government-owned hospitals, value-based care participants, and those with an Epic EHR had higher rates of receiving HRSN data from any source compared to their counterparts. About half of hospitals reported using data received for various purposes (e.g., referrals, discharge planning). Hospitals that received data via HIEs or EHR vendor networks or directly from other organizations were more likely to use data across measured uses. Hospitals that received data via referral platform were more likely to use data for referrals, informing community needs, and population health analytics. DISCUSSION: Many hospitals receive HRSN data from external sources, often through existing connections to health information networks, which was associated with greater internal and downstream use. CONCLUSION: Efforts to improve bi-directional exchange between health and social service organizations may enable hospitals to more effectively identify and address patients' HRSN.
Use of SNOMED CT, 2020-2025: literature review.
Chang E, Mostafa J
J Am Med Inform Assoc · 2026 Jun · PMID 42275568 · Publisher ↗
OBJECTIVE: To examine the evolving role and application of SNOMED CT (SCT) during 2020-2025, a period marked by the COVID-19 pandemic and accelerated adoption of artificial intelligence (AI) in healthcare. MATERIALS AND...
OBJECTIVE: To examine the evolving role and application of SNOMED CT (SCT) during 2020-2025, a period marked by the COVID-19 pandemic and accelerated adoption of artificial intelligence (AI) in healthcare. MATERIALS AND METHODS: We searched PubMed and Embase for articles published from October 2020 to June 2025. Included articles were classified into focus categories by implementation maturity (Theoretical, Predevelopment, Implementation, Evaluation/Commodity, and Non-operational) and usage categories. We compared trends with our previous review covering January 2015-September 2020. RESULTS: Following exclusion criteria, 651 articles were included for final review. The United States (n = 188) and United Kingdom (n = 92) were the largest contributors. COVID-19 emerged as the second most-investigated domain. The 2020-2025 period witnessed a dramatic shift toward mature implementation stages: Implementation (26.4%) and Evaluation/Commodity (32.0%) categories expanded, while Theoretical and Predevelopment categories decreased. SCT was increasingly used for knowledge graph construction, machine learning model validation, and patient data retrieval from registries. The most prominent use case involved retrieving patient data from national and commercial registries (n = 186). DISCUSSION: SCT's role evolved from a clinical terminology to a machine-interpretable biomedical knowledge base, supporting explainable AI. However, coding consistency remains challenging, and evidence demonstrating improved patient outcomes is lacking. CONCLUSION: SCT use has matured significantly, with widespread implementation in data repositories and emerging applications in AI. Future research must demonstrate clinical and operational benefits to ensure continued adoption.
Theory and practice in biomedical informatics: a framework for discovery.
Stead WW, Aliferis CF, Bastarache L … +2 more , Lorenzi NM, Ed Hammond W
J Am Med Inform Assoc · 2026 Jun · PMID 42246620 · Publisher ↗
OBJECTIVE: Clarify disciplinary foundations and internal structure of biomedical informatics. METHODS: We analyze BMI's emergence at disciplinary intersections and map its internal structure across 4 domains: theory and...
OBJECTIVE: Clarify disciplinary foundations and internal structure of biomedical informatics. METHODS: We analyze BMI's emergence at disciplinary intersections and map its internal structure across 4 domains: theory and practice of knowledge discovery, knowledge representation and reasoning, knowledge architecture, and knowledge-driven transformation. We compare BMI with mathematics, computer science, biostatistics, and biomedical engineering, and illustrate emergent characteristics through a precision medicine example. RESULTS: BMI's distinctive contribution-elucidating the structure of biomedical knowledge and developing methods to discover, preserve, and make knowledge actionable-requires strength across all 4 domains. BMI developed these domains pragmatically: building systems, extracting principles, and formalizing theories. The discipline must now complement empirical approaches with rigorous theoretical work: assessing adequacy of existing theories, identifying gaps, and orchestrating collaborative development. CONCLUSIONS: BMI creates emergent capabilities across disciplines. As biomedicine becomes increasingly complex, BMI must strengthen its theoretical foundations while demonstrating transformative potential of knowledge spanning biological scales and time.
SeqBoard: a genomics-based data dashboard for comprehensive wastewater virome monitoring.
Bauer C, Reger N, Rustem HAL … +13 more , Tisza M, Triosi CL, Javornik Cregeen S, Ghobrial L, Gitter A, Wu F, Surathu A, Deegan J, Mena KD, Petrosino J, Boerwinkle E, Hanson BM, Maresso AW
J Am Med Inform Assoc · 2026 Jun · PMID 42243631 · Publisher ↗
OBJECTIVES: To develop the first public-facing dashboard that translates genomic sequencing data from wastewater into accessible and actionable community information concerning human pathogenic viruses, representing a sh...
OBJECTIVES: To develop the first public-facing dashboard that translates genomic sequencing data from wastewater into accessible and actionable community information concerning human pathogenic viruses, representing a shift to sequencing-based public health wastewater monitoring. MATERIALS AND METHODS: We developed SeqBoard, a user-friendly dashboard that displays sequencing information from the total wastewater virome. The dashboard integrates diverse expertise and components, including data processing and analysis, visualization and management, security, and stakeholder engagement and feedback. We implemented a 3-tiered system for user interactions, customized to the general public, public health officials, and genomics experts. RESULTS: SeqBoard provides an intuitive interface for presenting genomic information as species-specific trend lines, level indicators, and all-site aggregates. It translates complex sequencing data into public health insights, including reporting on dozens of viruses of concern with modules for detections, variant information, and genomic context. DISCUSSION: The prevention of the next pandemic will require comprehensive pan-monitoring of deadly viruses and their evolution. Genomics-based dashboards will be essential for early detection of viral activity before significant clinical manifestation, thereby allowing public health systems to provide warnings, ready actions, and develop vaccines. CONCLUSION: SeqBoard shows that sequencing data can be translated into useful public health information, serving as a model for future sequencing-based pathogen dashboards. The dashboard is publicly available at https://tephi-ww.uth.edu/public-dashboard and represents the first publicly available dashboard providing pan viral genomic detection data for wastewater monitoring.
Optimizing Retrieval-Augmented Generation (RAG) in clinical medicine: methods and performance evaluation.
Li P, Patel A, Vallamchetla SK … +4 more , Heninger H, Contractor H, Tao C, Cheung J
J Am Med Inform Assoc · 2026 Jun · PMID 42243630 · Publisher ↗
OBJECTIVE: Evaluate how RAG architecture, including corpus structure, retrieval strategy, and pipeline complexity, affects LLM-based medical problem solving and knowledge retrieval in sleep medicine. MATERIALS AND METHOD...
OBJECTIVE: Evaluate how RAG architecture, including corpus structure, retrieval strategy, and pipeline complexity, affects LLM-based medical problem solving and knowledge retrieval in sleep medicine. MATERIALS AND METHODS: We benchmarked four open-source LLMs (Llama-3-8B, Llama -3 -70B, Qwen 2.5-14B, and Qwen 2.5-235B) using a knowledge base of five sleep medicine textbooks. We compared performance across three dimensions: corpus structure (raw text vs table-of-contents aligned.), retrieval strategy (dense embedding vs hybrid sparse-dense), and pipeline complexity (baseline vs augmented). Evaluation metrics included board-style multiple choice question (MCQ) accuracy and clinical case vignette diagnostic ranking. RESULTS: RAG improved MCQ accuracy for all models. Llama-8B saw the largest gain of 10.6% (61.8% to 72.4%), while Qwen-235B reached 87.3%. In clinical cases, Llama-8B accuracy dropped by 7.1% when using raw text and dense retrieval due to context noise. This was corrected by using structured hybrid configurations. Hybrid retrieval consistently outperformed dense-only methods. Overall, structured corpora improved primary diagnosis accuracy by 6.1% on average, with Qwen-235B reaching a peak 10.2% increase. DISCUSSION: RAG effectiveness depends on the balance between model size and data structure. Large models handle uncurated text well, but smaller models are easily distracted by irrelevant data. Hybrid retrieval is necessary to maintain precision with specialized medical terms. A structured corpus paired with a baseline hybrid pipeline offers the best stability and speed for clinical use. CONCLUSION: Rigorous data curation and hybrid retrieval are as essential as model scale for deploying safe, guideline-compliant AI in sleep medicine.
Characterizing surgeon workload with electronic health record data to predict time interval between surgeries and postoperative care delivery.
Akhagbosu J, Capan M, Balasubramanian H … +1 more , Kamine TH
J Am Med Inform Assoc · 2026 Jun · PMID 42228788 · Publisher ↗
OBJECTIVES: This study positions surgeon gap time, defined as the interval between consecutive surgeries performed by the same surgeon, as a surgeon-level metric of efficiency. Understanding gap time requires accounting...
OBJECTIVES: This study positions surgeon gap time, defined as the interval between consecutive surgeries performed by the same surgeon, as a surgeon-level metric of efficiency. Understanding gap time requires accounting for a surgeon's operative workload, yet no objective electronic health record (EHR)-derived measure exists. We conceptualize surgical case demand as an EHR-derived surrogate for operative workload and examine its association with surgeon gap time. MATERIALS AND METHODS: We analyzed 86 480 surgeries in 14 specialties performed between 2020 and 2023 at a US Medical Center. Surgical case demand was operationalized using patient and surgery features and clustered into demand types. Clinical implications were assessed by comparing postoperative care location and length of stay (LOS) across demand types. A classification tree was developed to assign demand types for future cases. Regression models were used to identify predictors of gap time. RESULTS: Clustering identified 3 demand types with distinct postoperative profiles. High-level recovery was required for 5.9%, 6.5%, and 17.2% of demand types 1, 2, and 3 cases, respectively (P < .001), with median LOS of 0.09, 0.97, and 1.80 days (P < .001). Demand type, case priority, surgical specialty, and surgical care location were significant predictors of gap time. DISCUSSION: Surgical case demand serves as an EHR-derived surrogate for operative workload, enabling structured analysis of surgeon gap time and the factors associated with it. CONCLUSION: Surgeon gap time is an objective metric that can be operationalized from EHR data, providing insights for scheduling, resource allocation, and overall health system efficiency.
Explainable machine learning in healthcare: methods, interpretation, and applications for clinical research.
Padmanabhan K, Lu M, Feng D … +4 more , Kan-Dobrosky N, Konduri S, Litman HJ, Livieratos A
J Am Med Inform Assoc · 2026 Jun · PMID 42224454 · Publisher ↗
OBJECTIVES: To provide a practical and methodologically grounded overview of explainable machine learning (XML) approaches in healthcare, with emphasis on their interpretation and application in clinical research and dec...
OBJECTIVES: To provide a practical and methodologically grounded overview of explainable machine learning (XML) approaches in healthcare, with emphasis on their interpretation and application in clinical research and decision support. By moving beyond traditional predictive models, this primer aims to foster trust, transparency, and informed clinical decision-making, ultimately bridging the gap between data science and medical practice. MATERIALS AND METHODS: We present a structured review of commonly used XML methodologies, including global and local interpretability tools such as SHapley Additive exPlanations (SHAP), Local Interpretable Model-Agnostic Explanations (LIME), Partial Dependence Plots (PDP), and Individual Conditional Expectation (ICE) plots. For each method, we explain the underlying mechanism at a high level, visualize representative outputs, and provide structured guidance on interpretation, appropriate use, and limitations, illustrated using the publicly available Heart Disease dataset. RESULTS: XML techniques provided intuitive visual and quantitative insights into how predictors influence model predictions. Global methods characterized population-level feature effects, whereas local methods revealed patient-level contributions useful for individualized interpretation. Our worked examples demonstrate how XML outputs can identify nonlinear relationships, detect interaction effects, and reveal heterogeneity in predicted risk across patients, addressing key challenges in translating ML predictions into interpretable outputs for clinical research. DISCUSSION AND CONCLUSION: XML tools offer valuable interpretability for ML models and support more transparent and accountable ML applications in clinical research. By providing a methodologically grounded overview alongside practical implementation examples and structured guidance on each method's strengths and limitations, this primer helps bridge the gap between advanced ML methodology and clinical applicability. Thoughtful adoption of XML approaches may facilitate better understanding, communication, and critical evaluation of ML predictions in healthcare research, ultimately supporting evidence-based clinical decision-making.
Explainability in context: calibrating appropriate trust and reliance in artificial intelligence.
Davis SE, Salwei ME
J Am Med Inform Assoc · 2026 May · PMID 42191222 · Publisher ↗
BACKGROUND AND SIGNIFICANCE: Predictive artificial intelligence (AI) promises to transform care delivery, enhance patient safety, and improve health outcomes. Realizing these benefits will require careful design, impleme...
BACKGROUND AND SIGNIFICANCE: Predictive artificial intelligence (AI) promises to transform care delivery, enhance patient safety, and improve health outcomes. Realizing these benefits will require careful design, implementation, and monitoring strategies to avoid unintended consequences, including automation bias (i.e., erroneously favoring recommendations from automated systems). Automation bias is particularly concerning due to the variability of AI performance across time and populations, leading to predictions that may be variably incorrect, uncertain, or unfair. APPROACH: We advocate for an expanded view of explainable AI that uses contextual information to help end users calibrate appropriate levels of trust and reliance. We propose multiple levels of contextualization-model, setting, subpopulation, and patient-that together provide insight for clinicians to evaluate the reliability of individual predictions. This includes information about historical and in-the-moment AI performance, algorithmic fairness, and prediction uncertainty. CONCLUSION: We outline an approach to integrate context-based explanations into decision support workflows to aid clinician interpretation without adding cognitive burden.
Correction to: Evaluating the effectiveness of biomedical fine-tuning for large language models on clinical tasks.
J Am Med Inform Assoc · 2026 Jun · PMID 42173163 · Full text
Abstract loading — click title to view on PubMed.
About
- Frequency
- Sun
- Papers found
- 200
- RSS feed
- Subscribe