Michael Oberst

Michael Oberst

Auto-generated profile, based on publicly available information

Michael Oberst

Johns Hopkins University, Baltimore, USA

Research Focus & Metrics

Academic Career Length

7Years (2020-Present)

(Calculated once Scilit publication data is synced: current year − year of first publication + 1)

Representative Publications

View All Publications
4 November 2020 in Science Translational Medicine
Verified
Journal Article
Closed access
PDF file
SKSanjat Kanjilal
Sanjat Kanjilal
MOMichael Oberst
Michael Oberst
SBSooraj Boominathan
Sooraj Boominathan
HZHelen Zhou
Helen Zhou
DHDavid C. Hooper
David C. Hooper
DSDavid Sontag
David Sontag
Machine learning models for predicting antibiotic resistance could reduce the use of broad-spectrum antibiotics in urinary tract infection.

28 August 2025 in NEJM AI
Verified
Journal Article
Closed access
MAMajid Afshar
Majid Afshar
FRFelice Resnik
Felice Resnik
MBMary Ryan Baumann
Mary Ryan Baumann
JHJosie Hintzke
Josie Hintzke
KLKayla Lemmon
Kayla Lemmon
JGJoel E. Gordon
Joel E. Gordon
Ambient artificial intelligence (AI) offers the potential to reduce documentation burden and improve efficiency through clinical note generation. Widespread adoption, however, remains limited due to challenges in electronic health record (EHR) integration, coding compliance, and real-world evaluation. This study introduces a framework and protocols to design, monitor, and deploy ambient AI within routine care. We launched an implementation phase to build technical workflows, establish governance, and inform a pragmatic randomized trial. A bidirectional governance model linked operations and research through multidisciplinary workgroups that incorporated the Systems Engineering Initiative for Patient Safety (SEIPS) framework. Integration into the EHR used Fast Healthcare Interoperability Resources (FHIR), and a real-time dashboard tracked utilization and documentation accuracy. To monitor drift, a difference-in-differences analysis was applied to three process metrics: time in notes, work outside work, and utilization. Audits of International Statistical Classification of Diseases and Related Health Problems, Tenth Revision (ICD-10) compliance were performed using an internally developed large language model (LLM), the validity of which was assessed through correlation with certified professional coders. Ambient AI utilization, measured as the proportion of eligible clinical notes completed using the system, had a weighted median of 65.4% (interquartile range, 50.6 to 84.0%). Iterative improvement cycles targeted task-specific adoption. A brief workflow issue related to a note template change initially reduced ICD-10 documentation accuracy from 79% (95% confidence interval [CI], 72 to 86%) to 35% (95% CI, 28 to 42%); accuracy returned to baseline after note template redesign and user training. The internally developed LLM coder achieved a strong correlation with professional coders (Pearson’s r=0.97). The trial enrolled 66 providers across eight specialties, powered at 90% for the primary outcome of provider well-being. We provide a publicly available framework and protocols to help safely implement ambient AI in health care. Innovations include an embedded pragmatic trial design, human factors engineering, compliance-driven feedback loops, and real-time monitoring to support deployment, ensuring fidelity before initiation of the clinical trial. (Funded by the University of Wisconsin Hospital and Clinics and the National Institutes of Health Clinical and Translational Science Award; NIH/ NCATS UL1TR002737; ClinicalTrials.gov number, NCT06517082.)

28 April 2026 in Nature Medicine
Verified
Journal Article
Closed access
PDF file
WAWouter Anton Christiaan van Amsterdam
Wouter Anton Christiaan van Amsterdam
MOMichael Oberst
Michael Oberst
JFJean Feng
Jean Feng
JWJenna Wiens
Jenna Wiens
STShengpu Tang
Shengpu Tang
AGAnna Goldenberg
Anna Goldenberg

22 September 2024 in medRxiv
Verified
Preprint
Open access
PDF file
CJChristina X Ji
Christina X Ji
SBSaul Blecker
Saul Blecker
MOMichael Oberst
Michael Oberst
MSMing-Chieh Shih
Ming-Chieh Shih
LHLeora I Horwitz
Leora I Horwitz
DSDavid Sontag
David Sontag
Objective The purpose of this study is to assess variation in first-line type 2 diabetes treatment empirically using a large clinical dataset. Since metformin, the guideline-recommended first-line treatment, is contraindicated for severe chronic kidney disease, we examine variation in this treatment decision on two axes—across estimated glomerular filtration rate (eGFR) measurements from the patient and across preferences from the prescribing provider. Study Design and Setting Using a large insurance claims dataset, we conducted a retrospective cohort study of patients who were newly initiated on a type 2 diabetes treatment (metformin versus a DPP-4 inhibitor or sulfonylurea). Three years of observation prior to treatment were required, and patients with type 1 or gestational diabetes or without eGFR results were excluded. To test whether the choice of treatment is significantly dependent on eGFR level, we performed a chi-squared test for association between eGFR level and treatment decision. 2) To test whether practice variation exists among providers that cannot be explained by treatment guidelines, we fitted restricted cubic spline models to predict treatment from patient age, eGFR, sex, history of heart failure, and treatment date. Then, we performed a generalized likelihood ratio test (GLRT) to assess whether a model that included provider-specific random effects is a better fit than a model without these random effects. Results Among 10,643 eligible patients, the choice of metformin versus a DPP-4 inhibitor or a sulfonylurea was significantly associated with eGFR level (p < 0.0001). Among the 2,271 patients seen by 173 providers with at least 10 patients in the cohort, a GLRT found significant variation exists across providers even after accounting for age, eGFR, sex, history of heart failure, and treatment date (p < 0.0001). Conclusion Our study found significant variation in first-line type 2 diabetes treatments—some that can be explained by treatment guidelines and some that may be due to provider preferences. Further studies can help elucidate whether such variation across providers is appropriate. The data-driven approaches in our study can also be applied to other disease areas to characterize variation in real-world clinical practice and potential opportunities for improvement. What is New?Key Findings We used a large health insurance claims dataset to show that first-line type 2 diabetes treatment is significantly associated with estimated glomerular filtration rate (eGFR). We established empirically that significant variation exists in how providers choose between metformin and DPP-4 inhibitors or sulfonylureas for first-line type 2 diabetes treatment, even after accounting for eGFR, age, sex, history of heart failure, and treatment date. What this adds to what is known? We proposed a new statistical approach to test for variation in treatment decisions across all providers while accounting for patient characteristics. We applied this approach to establish that providers may consider eGFR levels differently when prescribing first-line type 2 diabetes treatments. What is the implication and what should change now? Further work is needed to understand whether the provider variation we discovered for initial type 2 diabetes treatment is appropriate, and if not, how to remove this variation by reaching out to providers or improving treatment guidelines Our proposed statistical approach can be applied to determine whether provider variation exists for other diseases after accounting for patient characteristics relevant to the specific disease.