The foundation of the Duke Learning Health System- Harnessing the power of data science using re-usable tools & technology

Poster
Harnessing the power of data science
Organization
Abstract

The modern era of health analytics is built on data science. Data science merges computer science with statistics, creating new potential for extracting clinical insight from the exponentially increasing “big” data relevant to our patients’ health. Realizing this potential requires new capabilities and innovative technologies. For effective implementation in a healthcare setting, an Accelerator is needed to foster the development of these capabilities and technologies.

By increasing Duke Health’s capability to execute data-intensive and value-generating projects on a wider scale, the Accelerator (called the Duke “Crucible”) consists of a focused team with unique authorities and resources can further the development of our learning health system by increasing Duke’s capacity for innovative health data science development. Modern machine learning and artificial intelligence rapidly saturate the computational capabilities of conventional health IT; they require specific expertise, robust design, and novel architectures for successful implementation. A group that appreciates these demands while understanding how they interface with healthcare workflows is imperative as Duke transforms into a true learning health system.

The Deep Care Management platform illustrates the kind of project such an accelerator would undertake, and has been undertaken by the Duke Crucible. This platform—developed by Duke Connected Care (Duke Health’s Accountable Care Organization) in conjunction with Duke informatics and technology experts—employs methods and technologies that, until now, have never been used at Duke for health care delivery. Deep Care Management employs a deep learning network to generate over 1.5 million predictions for unplanned admissions across 52,000 patients and 32 diagnostic categories every month. This population risk stratification is then triaged into direct action by PHMO care managers who use the Deep Care Management platform on a daily basis.   

The Duke Crucible provides technical guidance and development capabilities for the complex data science problems faced by the clinical and translational science communities, while establishing security and regulatory guidelines and technical standards with institutional Leadership.  The Duke Crucible is designed to work in small, agile, and cross-functional project teams that understand, document, and solve the “problem” presented by each project. In rapid succession, the team designs and fine tunes its solution through an iterative process with stakeholders. Throughout this design and development effort, the Crucible works with the broader IT organization to ensure an application’s transition to production, deployment, and support can occur seamlessly. 

Authors
Ebony
Boulware
Dianne
Oliver-Clapsaddle
Executive Director, Data Science Product Development
Erich
Huang
Assistant Dean
Rebbecca
Moen
Chief Administrative Officer, CTSI
Warren
Kibbe
Professor
Yasii
Mirdamadi
Operations Coordinator
Hannah
Campbell
Shelley
Rusincovitch
Ursula
Rogers
Sr. Informaticist
Robert
Overton
johanna
odell
anthony
leiro
mike
winters
Benjamin
Neely
Senior Biostatistician
Jonathan
Turner
Sr. Analyst
David
Lally