AI-generated image

UCSD Health AI: 17% Sepsis Drop, Flow Revolutionized

UC San Diego Health deploys predictive AI to cut sepsis mortality 17%, save 50 lives yearly, optimize patient flow via Mission Control, and pilot safe generative AI integrations amid capacity strains.
Industry HealthcareAcademic Medical Center
Use Cases Sepsis PredictionPatient Flow OptimizationHospital Capacity ManagementClinical Decision Support
AI Tools Generative AIPredictive AnalyticsDeep LearningCOMPOSEREpic GPT-4Mission Control

Inhalt

Key Facts

  • Company: UC San Diego Health
  • Company Size: 30,000+ employees, 1M+ annual patient visits
  • Location: San Diego, California
  • AI Tool Used: COMPOSER deep learning model, Mission Control ops center, Epic GPT-4
  • Outcome Achieved: 17% sepsis mortality reduction, 50 lives saved/year, enhanced patient throughput

Want to achieve similar results with AI?

Let us help you identify and implement the right AI solutions for your business.

The Challenge

Sepsis, a life-threatening condition, poses a major threat in emergency departments, with delayed detection contributing to high mortality rates—up to 20-30% in severe cases. At UC San Diego Health, an academic medical center handling over 1 million patient visits annually, nonspecific early symptoms made timely intervention challenging, exacerbating outcomes in busy ERs [1][2]. A randomized study highlighted the need for proactive tools beyond traditional scoring systems like qSOFA.

Hospital capacity management and patient flow were further strained post-COVID, with bed shortages leading to prolonged admission wait times and transfer delays. Balancing elective surgeries, emergencies, and discharges required real-time visibility [3][4]. Safely integrating generative AI, such as GPT-4 in Epic, risked data privacy breaches and inaccurate clinical advice [4].

These issues demanded scalable AI solutions to predict risks, streamline operations, and responsibly adopt emerging tech without compromising care quality.

The Solution

UC San Diego Health implemented COMPOSER, a deep learning model trained on electronic health records to predict sepsis risk up to 6-12 hours early, triggering Epic Best Practice Advisory (BPA) alerts for nurses [2][5]. This quasi-experimental approach across two ERs integrated seamlessly with workflows [5].

Mission Control, an AI-powered operations command center funded by $22M, uses predictive analytics for real-time bed assignments, patient transfers, and capacity forecasting, reducing bottlenecks [3][4]. Led by Chief Health AI Officer Karandeep Singh, it leverages data from Epic for holistic visibility.

For generative AI, pilots with Epic's GPT-4 enable NLP queries and automated patient replies, governed by strict safety protocols to mitigate hallucinations and ensure HIPAA compliance [4]. This multi-faceted strategy addressed detection, flow, and innovation challenges.

Quantitative Results

  • Sepsis in-hospital mortality: 17% reduction
  • Lives saved annually: 50 across two ERs
  • Sepsis bundle compliance: Significant improvement
  • 72-hour SOFA score change: Reduced deterioration
  • ICU encounters: Decreased post-implementation
  • Patient throughput: Improved via Mission Control

Ready to transform your business with AI?

Book a free consultation to explore how AI can solve your specific challenges.

Implementation Details

Overview of AI Ecosystem

UC San Diego Health, under Chief Health AI Officer Karandeep Singh, built a comprehensive AI framework integrating predictive and generative technologies. Launched in phases from 2021-2024, it started with sepsis prediction and expanded to operations via Mission Control and gen AI pilots, all anchored in Epic EHR[4].

Sepsis Prediction: COMPOSER Model

The cornerstone is COMPOSER, a deep learning algorithm analyzing 50+ variables like vitals and labs. Implemented January 2021-April 2023 in two EDs, it triggers nurse-facing BPAs ~6 hours pre-sepsis, outperforming baselines. A before-after quasi-experimental study on 6,217 patients validated impact via propensity matching[2][5]. Training used de-identified data, with ongoing retraining for accuracy.

Mission Control Command Center

Funded by $22M investment, Mission Control—opened in 2023—centralizes AI-driven ops. Dashboards predict bed turnover, discharge readiness, and OR scheduling using predictive analytics. Real-time ML models forecast capacity peaks, automating transfers and reducing admission delays by hours. Integrated with Epic, it supports 1,200+ beds across facilities[3][4].

Generative AI Integration

Piloting Epic GPT-4, UCSD enables ambient documentation, patient summaries, and query responses. Safeguards include human oversight, custom fine-tuning, and audits. Early tests show time savings for clinicians, with expansion planned[4]. Ethical governance via AI council ensures equity and safety.

Timeline and Approach

2021: COMPOSER pilot. 2022: RCT validation. 2023: Mission Control launch, gen AI tests. Iterative via agile sprints, clinician feedback, and partnerships (e.g., Epic). Challenges like data silos overcome by FHIR standards; regulatory hurdles via IRB approvals[1][5]. Scalability now system-wide, with ROI tracking.

Interested in AI for your industry?

Discover how we can help you implement similar solutions.

Results

The deployment yielded transformative outcomes, with COMPOSER reducing sepsis in-hospital mortality by 17% in a rigorous study, directly saving an estimated 50 lives annually across two ERs alone[1][4]. Sepsis bundle compliance surged, minimizing 72-hour Sequential Organ Failure Assessment (SOFA) deteriorations and cutting ICU encounters, boosting ICU-free days[5]. Extrapolated system-wide, impacts amplify amid 1M+ visits. Mission Control revolutionized patient flow, slashing admission wait times and optimizing hospital capacity during surges—key post-COVID. Predictive models enable proactive staffing and transfers, enhancing throughput without added beds[3]. Funded by $22M, ROI emerges via efficiency gains, positioning UCSD as a leader[7]. Generative AI pilots with Epic GPT-4 freed clinician time for care, with safe NLP improving summaries and replies. Ongoing expansions under Singh's vision yield outcomes-based AI, earning accolades as top innovator. Future scales to readmissions, equity analytics[4][6].

Contact Us!

0/10 min.

Contact Directly

Your Contact

Philipp M. W. Hoffmann

Founder & Partner

Address

Reruption GmbH

Falkertstraße 2

70176 Stuttgart

Contact

Phone

+49 175 5190660

Email

p.hoffmann@reruption.com

Social Media

Sources