Sepsis detection and timely  care coordination using augmented data science

This Clinical Decision Support (CDS) analytic identifies onset of sepsis quicker and, more accurately using patient’s vitals, laboratory data, and medical administrations from the time patient is admitted to any given point.

This CDS, developed using machine learning, captures patient’s progression real-time while continually consuming reams of that patient’s data. 

• October 2015
  An inter-disciplinary team at Duke proposes innovation to improve early detection of Sepsis
  
  
• April 2016
  Pilot project launched in the academic flagship hospital, Duke University Hospital. The core is the development of this sepsis tool.
  
  • Duke undertakes a computationally intensive approach to study over 32 million data points of their highly curated in-patient data. The resources expended for multi-output Gaussian processes and recurrent neural networks produce results that are superior to other (less intensive) machine learning models.
  • The extensive, curated multi-health-records data together with the computational intensity produces an AI tool Duke is confident to install in its ED after substantial model verification
• November 2018 Prior to 2018, Duke uses CDS Algorithms built into EHR
  • Sepsis watch is launched
  • Extended to community hospitals over the next few months
    
• Implementation complete in FY 2020
  Improvements noticed across the health system:
  • SEP-1 Bundle Compliance increased significantly
  • Average Length of Stay reduced (due to early detection; timely administration of care)
  • Morality reduced
    

• Improved bundle compliance and, consequently reduced readmission & mortality that, among other factors, improve payor reimbursement rates
• Reduced ALOS through early-detection frees up beds for new patients directly increasing opportunity to increase revenue
• Confidence in using a CDS platform built exclusive by, and, implemented successfully, in a large hospital system using highly curated patient data and advanced data science