March 13, 2017

DIHI Team members often travel to present at and participate in conferences.  DIHI's 2015 RFA Chronic Kidney Disease Project was selected as a finalist for Boston Scientific Big Data Challenge. Mark Sendak and Faraz Yashar went to Boston to present their project ideas on March 9, 2017.  Please read about the experience below.

Mark Sendak, Clinical Informatics DIHI

On March 9, Faraz and I presented our vision for bringing machine learning and big data into healthcare at scale. It was the unveiling of the concept and, even though we didn’t win, we received invaluable feedback to help us shape the vision.

Over the last few years, DIHI has learned a lot about extracting, cleaning, and combining data sets from electronic health records and claims to support innovation projects. These efforts typically bring together experts in medicine and quantitative sciences who have little or no direct experience working with raw health care data. We’ve learned how to bridge these worlds.

Back in 2015, we started with two machine learning projects – chronic kidney disease and surgical complications. Both of these projects are now spinning out of Duke University Health System as analytics products. To learn more, check out StopCKD (https://www.indiegogo.com/projects/stopckd-fight-kidney-failure-and-more#/) and KelaHealth (https://www.indiegogo.com/projects/kelahealth-precision-medicine-for-surgery#/backers). In 2016, we grew our analytics team and brought on medical student scholars to support three projects:

- hospital admissions 

- sepsis

- congestive heart failure  

In 2017, we’ll be growing even more.

We’re building the technical infrastructure to rapidly generate high quality data sets for machine learning models. We work closely with physicians to review records to validate important outcomes and predictors and we work closely with model developers to maximize the clinical relevance of the prediction. Our goal is always to improve health system operations and clinical care.

The infrastructure we’re building to scale machine learning in health care doesn’t work just at Duke. We’re exploring avenues to open the infrastructure to bring model developers anywhere closer to relevant clinical questions to improve healthcare delivery. If you’re a health system interested in using one of the models we’ve developed or if you’re a model developer interested in testing performance on real clinical data, please reach out.

To watch the event, visit https://www.youtube.com/watch?v=dFj-CFvTdaE. Our pitch starts at the 1 hour mark.

For further information, please contact Mark Sendak or Faraz Yashar.