About

Jason Hom is a Clinical Professor in the Department of Medicine at Stanford University and is affiliated with the Center for Artificial Intelligence in Medicine & Imaging (AIMI). His work focuses on the application of artificial intelligence in the field of medicine and medical imaging. As a faculty member at Stanford, he contributes to advancing research and education in healthcare AI, supporting initiatives that integrate cutting-edge AI technologies into clinical practice. His role involves leadership within AIMI, fostering collaborations that aim to improve patient outcomes through innovative AI solutions.

Research topics

• Computer Science
• Medicine
• Natural Language Processing
• Biology
• Genetics
• Internal medicine
• Information Retrieval
• Artificial Intelligence
• Endocrinology
• Bioinformatics
• Medical education
• Computational biology
• Database
• World Wide Web
• Operating system
• Medical physics
• Biochemistry
• Psychology

Selected publications

• From iPatient to Ai-Patient: a responsibility to medical education

  BMJ Digital Health & AI · 2026-03-01

  articleOpen accessSenior author

  Our institution was an early implementer of a major electronic health record (EHR) in 2008.

  PublisherOA PDFDOI

• Deployment and Evaluation of an EHR-integrated, Large Language Model-Powered Tool to Triage Surgical Patients

  ArXiv.org · 2026-03-18

  articleOpen access

  Surgical co-management (SCM) is an evidence-based model in which hospitalists jointly manage medically complex perioperative patients alongside surgical teams. Despite its clinical and financial value, SCM is limited by the need to manually identify eligible patients. To determine whether SCM triage can be automated, we conducted a prospective, unblinded study at Stanford Health Care in which an LLM-based, electronic health record (EHR)-integrated triage tool (SCM Navigator) provided SCM recommendations followed by physician review. Using pre-operative documentation, structured data, and clinical criteria for perioperative morbidity, SCM Navigator categorized patients as appropriate, not appropriate, or possibly appropriate for SCM. Faculty indicated their clinical judgment and provided free-text feedback when they disagreed. Sensitivity, specificity, positive predictive value, and negative predictive value were measured using physician determinations as a reference. Free-text reasons were thematically categorized, and manual chart review was conducted on all false-negative cases and 30 randomly selected cases from the largest false-positive category. Since deployment, 6,193 cases have been triaged, of which 1,582 (23%) were recommended for hospitalist consultation. SCM Navigator displayed high sensitivity (0.94, 95% CI 0.91-0.96) and moderate specificity (0.74, 95% CI 0.71-0.77). Post-hoc chart review suggested most discrepancies reflect modifiable gaps in clinical criteria, institutional workflow, or physician practice variability rather than LLM misclassification, which accounted for 2 of 19 (11%) false-negative cases. These findings demonstrate that an LLM-powered, EHR-integrated, human-in-the-loop AI system can accurately and safely triage surgical patients for SCM, and that AI-enabled screening tools can augment and potentially automate time-intensive clinical workflows.

  PublisherOA PDF

• Deployment and Evaluation of an EHR-integrated, Large Language Model-Powered Tool to Triage Surgical Patients

  arXiv (Cornell University) · 2026-03-18

  preprintOpen access

  Surgical co-management (SCM) is an evidence-based model in which hospitalists jointly manage medically complex perioperative patients alongside surgical teams. Despite its clinical and financial value, SCM is limited by the need to manually identify eligible patients. To determine whether SCM triage can be automated, we conducted a prospective, unblinded study at Stanford Health Care in which an LLM-based, electronic health record (EHR)-integrated triage tool (SCM Navigator) provided SCM recommendations followed by physician review. Using pre-operative documentation, structured data, and clinical criteria for perioperative morbidity, SCM Navigator categorized patients as appropriate, not appropriate, or possibly appropriate for SCM. Faculty indicated their clinical judgment and provided free-text feedback when they disagreed. Sensitivity, specificity, positive predictive value, and negative predictive value were measured using physician determinations as a reference. Free-text reasons were thematically categorized, and manual chart review was conducted on all false-negative cases and 30 randomly selected cases from the largest false-positive category. Since deployment, 6,193 cases have been triaged, of which 1,582 (23%) were recommended for hospitalist consultation. SCM Navigator displayed high sensitivity (0.94, 95% CI 0.91-0.96) and moderate specificity (0.74, 95% CI 0.71-0.77). Post-hoc chart review suggested most discrepancies reflect modifiable gaps in clinical criteria, institutional workflow, or physician practice variability rather than LLM misclassification, which accounted for 2 of 19 (11%) false-negative cases. These findings demonstrate that an LLM-powered, EHR-integrated, human-in-the-loop AI system can accurately and safely triage surgical patients for SCM, and that AI-enabled screening tools can augment and potentially automate time-intensive clinical workflows.

  PublisherDOI

• MedFactEval and MedAgentBrief: A Framework and Workflow for Generating and Evaluating Factual Clinical Summaries

  2025-12-01 · 1 citations

  articleOpen access

  Evaluating factual accuracy in Large Language Model (LLM)-generated clinical text is a critical barrier to adoption, as expert review is unscalable for the continuous quality assurance these systems require. We address this challenge with two complementary contributions. First, we introduce MedFactEval, a framework for scalable, fact-grounded evaluation where clinicians define high-salience key facts and an "LLM Jury"-a multi-LLM majority vote-assesses their inclusion in generated summaries. Second, we present MedAgentBrief, a model-agnostic, multi-step workflow designed to generate high-quality, factual discharge summaries. To validate our evaluation framework, we established a gold-standard reference using a seven-physician majority vote on clinician-defined key facts from inpatient cases. The MedFactEval LLM Jury achieved almost perfect agreement with this panel (Cohen's κ = 81%), a performance statistically non-inferior to that of a single human expert (κ = 67%, P < 0.001). Our work provides both a robust evaluation framework (MedFactEval) and a high-performing generation workflow (MedAgentBrief), offering a comprehensive approach to advance the responsible deployment of generative AI in clinical workflows.

  PublisherDOI

• Enriched phenotypes in rare variant carriers suggest pathogenic mechanisms in rare disease patients

  BioData Mining · 2025-01-17 · 1 citations

  articleOpen access

  BACKGROUND: The mechanistic pathways that give rise to the extreme symptoms exhibited by rare disease patients are complex, heterogeneous, and difficult to discern. Understanding these mechanisms is critical for developing treatments that address the underlying causes of diseases rather than merely the presenting symptoms. Moreover, the same dysfunctional series of interrelated symptoms implicated in rare recessive diseases may also lead to milder and potentially preventable symptoms in carriers in the general population. Seizures are a common and extreme phenotype that can result from diverse and often elusive pathways in patients with ultrarare or undiagnosed disorders. METHODS: In this pilot study, we present an approach to understand the underlying pathways leading to seizures in patients from the Undiagnosed Diseases Network (UDN) by analyzing aggregated genotype and phenotype data from the UK Biobank (UKB). Specifically, we look for enriched phenotypes across UKB participants who harbor rare variants in the same gene known or suspected to be causally implicated in a UDN patient's recessively manifesting disorder. Analyzing these milder but related associated phenotypes in UKB participants can provide insight into the disease-causing mechanisms at play in rare disease UDN patients. RESULTS: We present six vignettes of undiagnosed patients experiencing seizures as part of their recessive genetic condition. For each patient, we analyze a gene of interest: MPO, P2RX7, SQSTM1, COL27A1, PIGQ, or CACNA2D2, and find relevant symptoms associated with UKB participants. We discuss the potential mechanisms by which the digestive, skeletal, circulatory, and immune system abnormalities found in the UKB patients may contribute to the severe presentations exhibited by UDN patients. We find that in our set of rare disease patients, seizures may result from diverse, multi-step pathways that involve multiple body systems. CONCLUSIONS: Analyses of large-scale population cohorts such as the UKB can be a critical tool to further our understanding of rare diseases in general. Continued research in this area could lead to more precise diagnostics and personalized treatment strategies for patients with rare and undiagnosed conditions.

  PublisherOA PDFDOI

• Expert-level validation of AI-generated medical text with scalable language models

  Research Square · 2025-07-08 · 1 citations

  preprintOpen access
  PublisherOA PDFDOI

• 12-Month Weight Loss and Adherence in a Real-World Tirzepatide-Supported Digital Obesity Service

  Preprints.org · 2025-11-27

  preprintOpen access

  Background: Obesity management is evolving with the integration of dual GIP/GLP-1 receptor agonists (Tirzepatide) into comprehensive Digital Weight-Loss Services (DWLSs). While clinical trials demonstrate high efficacy, real-world data are necessary to evaluate long-term adherence and identify predictive markers for patient persistence in these scalable care models. This study retrospectively assessed the 12-month effectiveness and adherence of a Tirzepatide-supported DWLS and identified demographic, clinical, and behavioral predictors of weight loss and program attrition. Methods: Data from 19,693 patients enrolled in the Juniper UK DWLS were analyzed. Adherence was defined by a minimum of 10 medication orders and 12-month weight submission. Weight loss in the full cohort was evaluated using the Last Observation Carried Forward (LOCF) method. Binary logistic and multiple linear regression models identified predictors of adherence and weight loss, respectively. Results: The 12-month adherence rate was 27%. The adherent sub-cohort (n=5,322) achieved a mean weight loss of 22.60 (±7.46) percent, compared to 13.62 (±10.85) percent in the full cohort (LOCF). Long-term adherence and weight loss were positively associated with weekly weight tracking consistency and health coach communication. Conversely, intensive tracking frequency and high weight loss velocity in the first month were sig-nificant inverse predictors of 12-month adherence. Reporting side effects was positively correlated with adherence, suggesting a reporting bias among engaged patients. Conclusion: The DWLS model facilitates maximum therapeutic efficacy for adherent patients. However, patient persistence remains the primary translational challenge. Clinical strategies should prioritize promoting sustainable, moderate behavioral pacing to mitigate attrition risk and optimize the public health effectiveness of medicated DWLSs.

  PublisherOA PDFDOI

• A typology of physician input approaches to using AI chatbots for clinical decision-making

  npj Digital Medicine · 2025-12-05 · 4 citations

  articleOpen access

  Recent studies have found that physicians with access to a large language model (LLM) chatbot during clinical reasoning tests may score no better to worse compared to the same chatbot performing alone with an input that included the entire clinical case. This study explores how physicians approach using LLM chatbots during clinical reasoning tasks and whether the amount of clinical case content included in the input affects performance. We conducted semi-structured interviews with U.S. physicians on experiences using an LLM chatbot and developed a typology based on input patterns. We then analyzed physician chat logs from two randomized controlled trials, coding each clinical case to an input approach type. Lastly, we used a linear mixed-effects model to compare the case scores of different input approach types. We identified four input approach types based on patterns of content amount: copy-paster (entire case), selective copy-paster (pieces of a case), summarizer (user-generated case summary), and searcher (short queries). Copy-pasting and searching were utilized most. No single type was associated with scoring higher on clinical cases. Other factors such as different prompting strategies, cognitive engagement, and interpretation of the outputs may have more impact and should be explored in future studies.

  PublisherOA PDFDOI

• Impact of Centers for Medicare &amp; Medicaid Services Screening Mandate on Inpatient Z-Code Documentation of Social Drivers of Health

  Journal of General Internal Medicine · 2025-11-21 · 2 citations

  articleOpen accessSenior author
  PublisherOA PDFDOI

• Publisher Correction: GPT-4 assistance for improvement of physician performance on patient care tasks: a randomized controlled trial

  Nature Medicine · 2025-02-17 · 8 citations

  erratumOpen access
  PublisherOA PDFDOI

Frequent coauthors

• Max Wintermark

  University of Maryland, Baltimore

  126 shared

• Hugo J. Bellen

  Baylor College of Medicine

  92 shared

• Michael F. Wangler

  Baylor College of Medicine

  86 shared

• May Christine V. Malicdan

  National Institutes of Health

  84 shared

• William A. Gahl

  83 shared

• David A. Sweetser

  82 shared

• Jonathan Chen

  Stanford University

  81 shared

• Jill A. Rosenfeld

  Harvard University

  79 shared