Bio
Dong-han Yao, M.D., is a physician informaticist and emergency physician at Stanford University. Dr. Yao holds a B.A. in Molecular & Cell Biology and Immunology from University of California, Berkeley, and an M.D. from Mount Sinai School of Medicine. He completed his Emergency Medicine Residency training at University of California, Los Angeles, and his fellowship training in Clinical Informatics at Stanford University.
Dr. Yao is an invited speaker at grand rounds, national conferences, and workshops on the topic of prompt engineering and generative AI for both healthcare and non-clinical applications around the country. He collaborates with the Stanford Health Care Data Science Team (DSatSHC) on both enterprise-level AI education and research, as well as co-development and evaluation of novel generative AI platforms and technologies for healthcare.
His scholarly and operational work include expanding patient access to acute care via virtual care, responsible integration of AI into medical education and the clinical continuum, and leveraging technology to streamline physician workflow and improve patient outcomes in the emergency department. His past informatics work includes award-winning usage of mobile devices to improve the efficiency and accessibility of medical documentation during the height of the COVID-19 pandemic, creation of novel patient discharge mechanisms for academic hospital centers, and development and implementation of new interdisciplinary clinical pathways for the emergency department. Dr. Yao's clinical interests include critical care, cardiac emergencies, telemedicine, and novel care delivery models in emergency medicine.
Clinical Focus
- Emergency Medicine
- Clinical Informatics
- Artificial Intelligence
- Telemedicine
- Large Language Models
- Machine Learning
Honors & Awards
-
Overall Winner, Stanford Quality Improvement and Patient Safety Symposium, Stanford University (2024)
-
Pediatrics Fellow Scholarship Award, Stanford Department of Pediatrics (2024)
-
Grand Prize Winner, UCLA Resident Informaticist Project Symposium, University of California, Los Angeles (2021)
-
Distinction in Research, Mount Sinai School of Medicine, Icahn School of Medicine at Mount Sinai (2019)
-
Grand Prize Winner, SINAInnovation MedMaker Challenge, Icahn School of Medicine at Mount Sinai (2016)
-
Patricia S. Levinson Research Award, Icahn School of Medicine at Mount Sinai (2016)
-
IBM Grand Prize Winner, DeveloperWeek SF (2015)
-
People’s Choice Winner, NASA SpaceApps Challenge (2015)
Boards, Advisory Committees, Professional Organizations
-
Executive Board, AMIA Clinical Informatics Fellows (ACIF) (2024 - Present)
-
Co-Chair, Stanford Resident Safety Council (2024 - Present)
Professional Education
-
Fellowship: Stanford University Clinical Informatics Fellowship (2025) CA
-
Board Certification: American Board of Emergency Medicine, Emergency Medicine (2024)
-
Residency: UCLA Emergency Medicine Residency (2023)
-
Medical Education: Icahn School of Medicine at Mount Sinai (2019) NY
-
Fellowship, Stanford University, Clinical Informatics (2025)
-
Residency, University of California, Los Angeles, Emergency Medicine (2023)
-
MD, Icahn School of Medicine at Mount Sinai (2019)
-
BA, University of California, Berkeley, Molecular & Cell Biology, Immunology (2014)
All Publications
-
Answering real-world clinical questions using large language model, retrieval-augmented generation, and agentic systems.
Digital health
2025; 11: 20552076251348850
Abstract
The practice of evidence-based medicine can be challenging when relevant data are lacking or difficult to contextualize for a specific patient. Large language models (LLMs) could potentially address both challenges by summarizing published literature or generating new studies using real-world data.We submitted 50 clinical questions to five LLM-based systems: OpenEvidence, which uses an LLM for retrieval-augmented generation (RAG); ChatRWD, which uses an LLM as an interface to a data extraction and analysis pipeline; and three general-purpose LLMs (ChatGPT-4, Claude 3 Opus, Gemini 1.5 Pro). Nine independent physicians evaluated the answers for relevance, quality of supporting evidence, and actionability (i.e., sufficient to justify or change clinical practice).General-purpose LLMs rarely produced relevant, evidence-based answers (2-10% of questions). In contrast, RAG-based and agentic LLM systems, respectively, produced relevant, evidence-based answers for 24% (OpenEvidence) to 58% (ChatRWD) of questions. OpenEvidence produced actionable results for 48% of questions with existing evidence, compared to 37% for ChatRWD and <5% for the general-purpose LLMs. ChatRWD provided actionable results for 52% of questions that lacked existing literature compared to <10% for other LLMs.Special-purpose LLM systems greatly outperformed general-purpose LLMs in producing answers to clinical questions. Retrieval-augmented generation-based LLM (OpenEvidence) performed well when existing data were available, while only the agentic ChatRWD was able to provide actionable answers when preexisting studies were lacking.Synergistic systems combining RAG-based evidence summarization and agentic generation of novel evidence could improve the availability of pertinent evidence for patient care.
View details for DOI 10.1177/20552076251348850
View details for PubMedID 40510193
View details for PubMedCentralID PMC12159471
-
LLMonFHIR: A Physician-Validated, Large Language Model-Based Mobile Application for Querying Patient Electronic Health Data.
JACC. Advances
2025; 4 (6 Pt 1): 101780
Abstract
To improve healthcare quality and empower patients, federal legislation requires nationwide interoperability of electronic health records (EHRs) through Fast Healthcare Interoperability Resources (FHIR) application programming interfaces. Nevertheless, key barriers to patient EHR access-limited functionality, English, and health literacy-persist, impeding equitable access to these benefits.This study aimed to develop and evaluate a digital health solution to address barriers preventing patient engagement with personal health information, focusing on individuals managing chronic cardiovascular conditions.We present LLMonFHIR, an open-source mobile application that uses large language models (LLMs) to allow users to "interact" with their health records at any degree of complexity, in various languages, and with bidirectional text-to-speech functionality. In a pilot evaluation, physicians assessed LLMonFHIR responses to queries on 6 SyntheticMass FHIR patient datasets, rating accuracy, understandability, and relevance on a 5-point Likert scale.A total of 210 LLMonFHIR responses were evaluated by physicians, receiving high median scores for accuracy (5/5), understandability (5/5), and relevance (5/5). Challenges summarizing health conditions and retrieving lab results were noted, with variability in responses and occasional omissions underscoring the need for precise preprocessing of data.LLMonFHIR's ability to generate responses in multiple languages and at varying levels of complexity, along with its bidirectional text-to-speech functionality, give it the potential to empower individuals with limited functionality, English, and health literacy to access the benefits of patient-accessible EHRs.
View details for DOI 10.1016/j.jacadv.2025.101780
View details for PubMedID 40373519
-
Using a Large Language Model to Identify Adolescent Patient Portal Account Access by Guardians.
JAMA network open
2024; 7 (6): e2418454
View details for DOI 10.1001/jamanetworkopen.2024.18454
View details for PubMedID 38916895
-
GPT-4 VS. NLP: ENHANCING CONFIDENTIALITY IN ADOLESCENT HEALTH PORTALS BY AUTOMATED DETECTION OF INAPPROPRIATE GUARDIAN ACCESS
SPRINGER. 2024: S607-S608
View details for Web of Science ID 001433572701481
-
Therapeutic Anticoagulation with Heparin in Critically Ill Patients with Covid-19.
The New England journal of medicine
2021
Abstract
BACKGROUND: Thrombosis and inflammation may contribute to morbidity and mortality among patients with coronavirus disease 2019 (Covid-19). We hypothesized that therapeutic-dose anticoagulation would improve outcomes in critically ill patients with Covid-19.METHODS: In an open-label, adaptive, multiplatform, randomized clinical trial, critically ill patients with severe Covid-19 were randomly assigned to a pragmatically defined regimen of either therapeutic-dose anticoagulation with heparin or pharmacologic thromboprophylaxis in accordance with local usual care. The primary outcome was organ support-free days, evaluated on an ordinal scale that combined in-hospital death (assigned a value of -1) and the number of days free of cardiovascular or respiratory organ support up to day 21 among patients who survived to hospital discharge.RESULTS: The trial was stopped when the prespecified criterion for futility was met for therapeutic-dose anticoagulation. Data on the primary outcome were available for 1098 patients (534 assigned to therapeutic-dose anticoagulation and 564 assigned to usual-care thromboprophylaxis). The median value for organ support-free days was 1 (interquartile range, -1 to 16) among the patients assigned to therapeutic-dose anticoagulation and was 4 (interquartile range, -1 to 16) among the patients assigned to usual-care thromboprophylaxis (adjusted proportional odds ratio, 0.83; 95% credible interval, 0.67 to 1.03; posterior probability of futility [defined as an odds ratio <1.2], 99.9%). The percentage of patients who survived to hospital discharge was similar in the two groups (62.7% and 64.5%, respectively; adjusted odds ratio, 0.84; 95% credible interval, 0.64 to 1.11). Major bleeding occurred in 3.8% of the patients assigned to therapeutic-dose anticoagulation and in 2.3% of those assigned to usual-care pharmacologic thromboprophylaxis.CONCLUSIONS: In critically ill patients with Covid-19, an initial strategy of therapeutic-dose anticoagulation with heparin did not result in a greater probability of survival to hospital discharge or a greater number of days free of cardiovascular or respiratory organ support than did usual-care pharmacologic thromboprophylaxis. (REMAP-CAP, ACTIV-4a, and ATTACC ClinicalTrials.gov numbers, NCT02735707, NCT04505774, NCT04359277, and NCT04372589.).
View details for DOI 10.1056/NEJMoa2103417
View details for PubMedID 34351722
-
Therapeutic Anticoagulation with Heparin in Noncritically Ill Patients with Covid-19.
The New England journal of medicine
2021
Abstract
BACKGROUND: Thrombosis and inflammation may contribute to the risk of death and complications among patients with coronavirus disease 2019 (Covid-19). We hypothesized that therapeutic-dose anticoagulation may improve outcomes in noncritically ill patients who are hospitalized with Covid-19.METHODS: In this open-label, adaptive, multiplatform, controlled trial, we randomly assigned patients who were hospitalized with Covid-19 and who were not critically ill (which was defined as an absence of critical care-level organ support at enrollment) to receive pragmatically defined regimens of either therapeutic-dose anticoagulation with heparin or usual-care pharmacologic thromboprophylaxis. The primary outcome was organ support-free days, evaluated on an ordinal scale that combined in-hospital death (assigned a value of -1) and the number of days free of cardiovascular or respiratory organ support up to day 21 among patients who survived to hospital discharge. This outcome was evaluated with the use of a Bayesian statistical model for all patients and according to the baseline d-dimer level.RESULTS: The trial was stopped when prespecified criteria for the superiority of therapeutic-dose anticoagulation were met. Among 2219 patients in the final analysis, the probability that therapeutic-dose anticoagulation increased organ support-free days as compared with usual-care thromboprophylaxis was 98.6% (adjusted odds ratio, 1.27; 95% credible interval, 1.03 to 1.58). The adjusted absolute between-group difference in survival until hospital discharge without organ support favoring therapeutic-dose anticoagulation was 4.0 percentage points (95% credible interval, 0.5 to 7.2). The final probability of the superiority of therapeutic-dose anticoagulation over usual-care thromboprophylaxis was 97.3% in the high d-dimer cohort, 92.9% in the low d-dimer cohort, and 97.3% in the unknown d-dimer cohort. Major bleeding occurred in 1.9% of the patients receiving therapeutic-dose anticoagulation and in 0.9% of those receiving thromboprophylaxis.CONCLUSIONS: In noncritically ill patients with Covid-19, an initial strategy of therapeutic-dose anticoagulation with heparin increased the probability of survival to hospital discharge with reduced use of cardiovascular or respiratory organ support as compared with usual-care thromboprophylaxis. (ATTACC, ACTIV-4a, and REMAP-CAP ClinicalTrials.gov numbers, NCT04372589, NCT04505774, NCT02735707, and NCT04359277.).
View details for DOI 10.1056/NEJMoa2105911
View details for PubMedID 34351721
-
The Outcome of Patients With Localized Undifferentiated Pleomorphic Sarcoma of the Lower Extremity Treated at Stanford University
AMERICAN JOURNAL OF CLINICAL ONCOLOGY-CANCER CLINICAL TRIALS
2019; 42 (2): 166–71
View details for DOI 10.1097/COC.0000000000000496
View details for Web of Science ID 000465420000010
-
The Outcome of Patients With Localized Undifferentiated Pleomorphic Sarcoma of the Lower Extremity Treated at Stanford University.
American journal of clinical oncology
2018
Abstract
BACKGROUND: As a diagnosis of exclusion, Undifferentiated Pleomorphic Sarcoma (UPS) has unclear clinical characteristics. The objective of this retrospective cohort study is to investigate which clinical and prognostic factors of primary lower-extremity UPS will determine failure.METHODS: We retrospectively reviewed 55 primary lower-extremity UPS cases treated at Stanford between 1998 and 2015. Overall Survival (OS) and Disease-Free Survival (DFS) curves were calculated. Univariate Fisher's Exact Tests were used to examine relationships between disease recurrence, treatment, patient factors, tumor characteristics, and surgical margins.RESULTS: 5-year DFS and OS rates were 60% (95% CI, 45%-72%) and 68% (95% CI, 53%-79%), respectively. The 5-year DFS rate for patients with positive margins was 33.3% (95% CI, 5%-68%) compared with 63% (95% CI, 47%-76%) for patients with negative margins. (Log-rank, P=0.03). The OS rate for those with disease recurrence was 42% % (95% CI, 16%-67%) compared with 76% (95% CI, 59%-87%) for patients who did not have disease recurrence (log-rank, P=0.021). Local failure occurred more frequently with omission of radiation therapy (Fisher's exact test, P=0.009).CONCLUSIONS: Positive surgical margins are an important prognostic factor for predicting relapse in UPS. Relapse of any kind led to worse OS. Radiation therapy improved local control of disease but had no statistically significant effect on DFS, highlighting the need for improved diagnostics to identify those at highest risk for hematogenous metastasis and for selection of patients for adjuvant systemic treatment.
View details for PubMedID 30557163
-
Predictors of Same-Day Discharge in Primary Total Joint Arthroplasty Patients and Risk Factors for Post-Discharge Complications.
The Journal of arthroplasty
2017; 32 (9S): S150-S156.e1
Abstract
Same-day (<24 h) discharge total joint arthroplasty (TJA) may be a safe and effective option for certain patients with end-stage osteoarthritis. Given the growing pressure to improve quality and lower TJA episode costs, surgeons must identify which TJA patients can be appropriately discharged home quickly and safely. This study identifies characteristics associated with same-day discharge post-TJA as well as assesses risk factors for complications in this select patient population.Bivariate and multivariate analyses were performed using perioperative variables from the 2011 to 2014 National Surgical Quality Improvement Program database.In total, 7474 primary TJAs among 120,847 TJA patients were discharged within 24 h post-surgery. These patients were more likely to be younger (<50 years), male sex, American Society of Anesthesiologists class 1 or 2, and less likely to be obese or taking steroids (P < .05 for all). They were also less likely to have co-morbidities. Rates of severe adverse event (SAE) or unplanned readmission post-discharge were 1.3% and 1.9%, respectively. Multivariate analysis identified age >80 (odds ratio [OR] 4.16, P = .001), smoking (OR 1.61, P = .03), bleeding-causing disorders (OR 2.56, P = .01), American Society of Anesthesiologists class 3 or 4 (OR 1.42, P < .05), and SAE pre-discharge (OR 13.13, P < .0001) as independent predictors for adverse events or readmission in this population.Patient characteristics, co-morbidities, and SAEs pre-discharge can be used to assess potential for discharge within 24 h. The results of our analysis may be used to develop risk stratification tools for identification of patients that are truly appropriate for same-day discharge TJA.
View details for DOI 10.1016/j.arth.2016.12.017
View details for PubMedID 28089186
-
Home Discharge After Primary Elective Total Joint Arthroplasty: Postdischarge Complication Timing and Risk Factor Analysis.
The Journal of arthroplasty
2017; 32 (2): 375-380
Abstract
Bundled payment programs for primary total joint arthroplasty (TJA) have identified reducing nonhome discharge as a major area of cost savings. Health care providers must therefore identify, risk stratify, and appropriately care for home-discharged TJA patients. This study aimed to analyze risk factors and timing of postdischarge complications among home-discharged primary total hip arthroplasty (THA) and total knee arthroplasty (TKA) patients and risk stratify them to identify those who would benefit from higher level care.Patients discharged home after elective primary THA/TKA from 2011 to 2014 were identified in the National Surgical Quality Improvement Program database. Bivariate and multivariate analyses were performed using perioperative variables.A total of 50,376 and 71,293 home-discharged THA and TKA patients were included for analysis, of which, 1575 THA (3.1%) and 2490 TKA (3.5%) patients suffered postdischarge severe complications or unplanned readmissions. These patients were older, smokers, obese, and functionally dependent (P < .001 for all). In multivariate analysis, severe adverse event predischarge, age, male gender, functional status, and 10 other variables were all associated with ≥1.22 odds of postdischarge severe adverse event or readmission (P < .05). THA and TKA patients with 2, 3, or ≥4 risk factors had 1.43-5.06 times odds of complications within 14 days post discharge and 1.41-3.68 times odds of complications beyond 14 days compared to those with 0 risk factors (P < .001 for all).Risk factors can be used to predict which home-discharged TJA patients are at greatest risk of postdischarge complications. Given that this is a growing population, we recommend the development of formal risk-stratification protocols for home-discharged TJA patients.
View details for DOI 10.1016/j.arth.2016.08.004
View details for PubMedID 27865568
-
Myocardial Infarction Risk in Arthroplasty vs Arthroscopy: How Much Does Procedure Type Matter?
The Journal of arthroplasty
2017; 32 (1): 246-251
Abstract
This study aimed at assessing short-term risk of serious cardiac events after elective total joint arthroplasty (TJA) as compared to a less-invasive procedure, knee arthroscopy (KA).Patients who underwent elective primary total hip arthroplasty (THA), total knee arthroplasty (TKA), or KA from 2011 to 2014 were identified in the American College of Surgeons National Surgical Quality Improvement Program database. A 1:1 propensity matching was used to generate 2 control cohorts of KA patients with similar characteristics. Bivariate and multivariate analyses were assessed using perioperative variables.A total of 24,203 THA, 21,740 TKA, and 45,943 KA patients were included. Bivariate analysis revealed significantly higher rates of serious 30-day cardiac events (myocardial infarction or cardiac arrest) among THA (0.15% vs 0.05%, P < .001) and TKA patients (0.14% vs 0.05%, P < .03) vs KA controls. In multivariate analysis controlling for patient characteristics and comorbidities, THA and TKA were associated with a 2.61 and 1.98 times odds of serious 30-day cardiac events as compared to controls (P ≤ .03 for both). Additional independent predictors of serious 30-day cardiac events included age, smoking, cardiac disease, and American Society of Anesthesiologists class 3/4. In the THA and TKA cohorts, serious cardiac events occurred within the first 3 days postoperation compared to 4 days in controls.After controlling for patient characteristics and comorbidities, TJA increased the short-term risk of serious cardiac event compared to a less-invasive procedure. This information better quantifies the risk differential for patients considering surgery as they engage in shared decision making with their providers. In addition, our data may have an impact on perioperative management of antithrombotic medications used in patients with cardiac disease. The median time in days to serious cardiac event was 2 in THA and 3 in TKA vs 4 in KA, which may have implications in postoperative monitoring of patients after surgery.
View details for DOI 10.1016/j.arth.2016.06.033
View details for PubMedID 27480828
-
Novel Intracranial Xenografts Of CNS Lymphoma Implicate a Role For Cereblon As a Mediator Of Lenalidomide Efficacy
Blood
2013; 122 (21)
View details for DOI 10.1182/blood.V122.21.374.374
https://orcid.org/0000-0002-5468-807X