Let us help you identify and implement the right AI solutions for your business.
Sepsis is a leading cause of hospital mortality, affecting over 1.7 million Americans annually with a 20-30% mortality rate when recognized late. At Duke Health, clinicians faced the challenge of early detection amid subtle, non-specific symptoms mimicking other conditions, leading to delayed interventions like antibiotics and fluids. Traditional scoring systems like qSOFA or NEWS suffered from low sensitivity (around 50-60%) and high false alarms, causing alert fatigue in busy wards and EDs.[1]
Additionally, integrating AI into real-time clinical workflows posed risks: ensuring model accuracy on diverse patient data, gaining clinician trust, and complying with regulations without disrupting care. Duke needed a custom, explainable model trained on its own EHR data to avoid vendor biases and enable seamless adoption across its three hospitals.[2]
Duke's Sepsis Watch is a deep learning model leveraging real-time EHR data (vitals, labs, demographics) to continuously monitor hospitalized patients and predict sepsis onset 6 hours in advance with high precision. Developed by the Duke Institute for Health Innovation (DIHI), it triggers nurse-facing alerts (Best Practice Advisories) only when risk exceeds thresholds, minimizing fatigue.[3]
The model was trained on Duke-specific data from 250,000+ encounters, achieving AUROC of 0.935 at 3 hours prior and 88% sensitivity at low false positive rates. Integration via Epic EHR used a human-centered design, involving clinicians in iterations to refine alerts and workflows, ensuring safe deployment without overriding clinical judgment.[4]
Book a free consultation to explore how AI can solve your specific challenges.
Duke Health initiated Sepsis Watch in 2016 through DIHI, collaborating with data scientists, clinicians, and ethicists. The team curated a dataset from over 250,000 patient encounters (2014-2017), labeling sepsis per Sepsis-3 criteria. Using deep neural networks (LSTM-based for time-series data), the model processed 200+ features from EHRs, outperforming traditional ML like XGBoost with an AUROC of 0.935 for 3-hour predictions and 0.927 at 6 hours. [1] Explainability features, like feature importance visualizations, built trust.
Launched in November 2018 on inpatient units, Sepsis Watch integrated as an Epic Best Practice Advisory (BPA), alerting nurses via banner when risk >5%. Phased rollout: pilot on one unit, then system-wide by 2020 including all three EDs (Duke University Hospital, Duke Regional, Duke Raleigh). Customization ensured alerts fired only for high-confidence cases, with low volume (1-2/day per unit) to combat fatigue. Clinician feedback loops refined thresholds quarterly. [2]
Key hurdles included data quality (missing EHR values) addressed via imputation and robust training; regulatory hurdles navigated with IRB approval and FDA discussions (deemed non-device); and change management. DIHI's human-centered AI approach involved 100+ clinician interviews, resulting in 95% satisfaction and <10% override rate. Scalability tested via A/B pilots showed no unintended consequences. [3]
Post-deployment, Duke tracked metrics via dashboards: time-to-antibiotics dropped 1.2 hours, bundle compliance rose 20%. A 2020 JMIR study confirmed feasibility, with plans for expansion to other conditions like AKI. By 2023, monitoring thousands daily, it reshaped Duke's AI governance, emphasizing reproducibility and equity. [4] Ongoing updates incorporate new data, maintaining performance.
Discover how we can help you implement similar solutions.
Duke Health's Sepsis Watch has transformed sepsis management, enabling earlier interventions that correlate with improved survival rates. Post-implementation, time to broad-spectrum antibiotics decreased by an average of 1.2 hours, and sepsis bundle compliance surged by 20%, directly linking to better outcomes per internal audits. Clinicians reported high utility, with override rates under 10%, indicating strong trust in the AI. [2]
A landmark 2020 implementation study in JMIR documented successful real-world adoption, rare for ML in healthcare, with no increase in workload and positive feedback from nurses (88% found alerts actionable). System-wide deployment by 2020 across EDs prevented sepsis progression in high-risk patients, contributing to a 12% reduction in sepsis mortality at Duke facilities compared to benchmarks. [3]
Long-term impact includes cost savings from fewer ICU days and establishing DIHI as an AI leader. As of 2025, Sepsis Watch monitors 1,000+ patients daily, inspiring expansions to other predictions. Challenges like bias mitigation were overcome via diverse training data, ensuring equitable performance across demographics. This case exemplifies scalable, ethical AI in academia. [5]
Founder & Partner
Reruption GmbH
Falkertstraße 2
70176 Stuttgart
Phone