Shamim Nemati
University of California-San Diego, CA·Biomedical Engineering
In hospitals, the difference between life and death often depends on spotting a patient's decline hours early. Two examples are sepsis, the body's runaway response to infection, and respiratory failure.
- Fields
Biomedical Engineering
Computer Science
Medicine & Clinical Research
- Topics
sepsis
septic shock
organ injury
- Grants
- 3 known grants, through Jul 2030
What they work on
- Compare a large language model–supported differential diagnosis system against the binary sepsis model COMPOSER for accuracy, with expert clinician review of the diagnosis lists, using retrospective hospital data.
- Run a prospective randomized trial testing whether showing multiple ranked diagnoses improves diagnostic timeliness, patient outcomes and clinician trust compared with yes-or-no AI alerts.
- Deploy and iteratively improve VentNet, an EHR-based machine learning model predicting respiratory failure in COVID-19 patients 24 hours ahead, by adding imaging data and actionable explanations.
- Discover sepsis, septic shock and organ injury phenotypes that generalize across five health systems using domain adaptation and multi-task learning on hourly EHR data.
- Integrate bedside device and wearable sensor data, such as continuous blood pressure and lactate, into sepsis analytics and extend FHIR and OMOP interoperability standards for these sources.
More about the research
Electronic health records (EHRs) and bedside monitors produce large amounts of data that machine learning can use to forecast these events. Turning those forecasts into tools that doctors trust and use safely is a separate challenge. Three projects build, deploy and test such prediction systems for critical illness.
Why it matters
Sepsis kills more than 270,000 people a year in the United States. Respiratory failure strained ventilator supplies during COVID-19. Earlier, more transparent warnings could help clinicians give antibiotics, fluids or ventilation at the right time and allocate ventilators and staff. Showing whether multi-diagnosis AI reduces clinician over-reliance could guide safer adoption of AI decision support in hospitals.
Known grants
From NIH and NSF research grants. Labs often have other funding we can't see.
RAAB-AI: Reducing Automation and Anchoring Bias in AI Systems
NIH · May 2022 – Jul 2030
VentNet: A Real-Time Multimodal Data Integration Model for Prediction of Respiratory Failure in Patients with COVID-19
NIH · Feb 2022 – Jan 2027
GeneRAlizable Sepsis Phenotyping (GRASP) using Electronic Health Records and Continuous Monitoring Sensors
NIH · Aug 2021 – Nov 2026
Closest labs
Jacqueline Stocking
University of California-Davis, CA·Medicine & Clinical Research·Early-career
Optimize and validate an automated, EHR-based model predicting postoperative respiratory failure, using surgical patient data from 2014 to 2021 and pre- and intra-operative risk factors.
Selen Bozkurt
Emory University, GA·Medicine & Clinical Research
Translate clinical trial protocol eligibility text into computable phenotypes and query packages tied to standard vocabularies, and validate them against expert references and prescreening logs.
Romain Pirracchio
University of California-San Francisco, CA·Medicine & Clinical Research
Train machine learning models on electronic health record data from three health systems to estimate each patient's individual treatment effect from antibiotics, fluid resuscitation and vasopressors in suspected community-onset lung sepsis.
The school
University of California-San Diego: 84 labs on PI Pages, $1.9B a year on research, 637 research doctorates a year.
Reach out
A start for your email. Fill in the brackets.