…A national health data science challenge established to advance human health through machine learning
The 2019 DII National Health Data Science Challenge Workshop
On October 18, 2019, the DII Challenge Workshop was hosted by McWilliams School of Biomedical Informatics at UTHealth Houston, formerly UTHealth Houston School of Biomedical Informatics (SBMI) as the culminating event of the 2019 DII Challenge, SBMI’s first national data science competition. For more information on the DII Challenge Workshop, click to view the SBMI's 2019 DII Challenge news story. Below, see some of the highlights of this event; for more photos, click to view the DII Challenge Album .
2019 DII Challenge - Winning Teams
& Workshop Presenters*
Task 1, Sepsis Onset Prediction: GuanLab Team
Yuanfang Guan and Xianghao Chen* – University of Michigan
Task 2, Sepsis Mortality Prediction: NCH Team
Simon Lin*, Xianlong Zeng*, Steve Rust, and Sven Bambach – Nationwide Children’s Hospital
Task 3, Innovation Track: LCP Team
Tom Pollard* and Alistair Johnson* – MIT
Honorable Mention: Buckeye AI Team
Ping Zhang, Changchang Yin*, and Dongdong Zhang* – The Ohio State University
…And Congratulations to All of our Participating DII Challenge Teams!!
Developing advanced AI/machine learning techniques to leverage large, diverse sets of health data (e.g., derived from electronic health records, claims, social determinants of health, and devices) represents an extraordinary means for potentially improving patient safety and quality of care through groundbreaking solutions to fundamental questions that can only be derived through the effective management and analysis of big data.
To this end, UTHealth School of Biomedical Informatics is hosting The 2019 DII National Data Science Challenge to enable participating teams to leverage a subset of the de-identified, EHR derived Cerner Health Facts® database for the purpose of solving a clinically relevant problem to advance human health.
The specific use case and related tasks leveraging Cerner Health Facts® have been determined by SBMI, in collaboration with its sponsor, Cerner Corporation. AWS is providing cloud support for the Challenge. In 2019 the Challenge use case is focused on predictive accuracy for two tasks (tracks): sepsis onset and sepsis mortality. The most innovative solution (not necessarily the one with best performance) to the above referenced tasks constitutes a third track for the Challenge.
Specifics regarding the use case and related tasks can be found here.
RULES (Abbreviated Version)
(A complete copy of the Rules and Conditions is incorporated into the Challenge material provided electronically as part of registration. A hard copy may also be requested via the Challenge e-mail address: DII.NDSC@uth.tmc.edu.)
- Participants must be 18 years of age or older to enter the Challenge.
- Participants must reside within the US and hold a valid, personal US tax identification number. Participants who are not US citizens or permanent residents are solely responsible for determining whether the receipt of prize funds through this Challenge will affect their visa status.
- Participants must not be residents of states where the Challenge is restricted by filing requirements or is otherwise prohibited.
- Participants may not be employees, officers, faculty, researchers, trainees, or students (or members of their immediate families and those living in the same household) of UTHealth (Host) or Cerner (Sponsor).
- Participants may not be Challenge judges—or be related to a Judge.
- Participants are responsible for ensuring that their participation in the Challenge complies with any educational, employment, or contractual obligations they may have, including but not limited to, restrictions related to participation, publicity, and the acceptance of prizes.
- Participants must compete in teams of between 2-8 members. Each Participant in the team must meet all eligibility requirements, and a team leader must be designated as the official representative for correspondence and other purposes related to the Challenge. If any member of a team fails to meet the eligibility requirements or fails to comply with the Challenge Rules, the team will be disqualified.
- If a Participant attempts to compromise the integrity or the legitimate operation of this Challenge by hacking, cheating, or committing fraud in any way, in addition to disqualification, the Host may pursue legal actions.
Entry Submission Requirements
- Participant and/or Participant’s team members own or otherwise have all rights necessary to provide the submission and grant the rights;
- The submission does not contain material that violates or infringes another’s rights, including without limitation, privacy, publicity, or intellectual property rights infringement;
- The submission is the original work of the Participant and/or Participant’s team members, and must not have been previously published or won previous awards.
- The submission does not contain any Host (UTHealth), Sponsor (Cerner), infrastructure provider (e.g., AWS), or third party brand names, logos, trademarks, or any copyrighted components (other than those owned by the Participant or those for which Participant has obtained permission, including credits required by any licensing agreements);
- The submission does not contain any viruses, worms, spy ware, or other components or instructions that are malicious, deceptive, or designed to limit or harm the functionality of a computer;
- The submission does not in any way violate any applicable federal, state, or local laws or regulations;
- The submission is not subject to any confidentiality agreements.
Data Use Agreement
For the purposes of the Challenge, the Participant will be granted access to a Cerner Health Facts® dataset extract (the “Challenge Dataset”) made available through a secure infrastructure. As a condition of participation in the Challenge and prior to receiving access to the Challenge Dataset, Participant must to sign and comply with the terms of the Challenge Participant Data Use Agreement with Cerner (provided electronically, as part of the registration process).
The Host will select a panel of Judges to validate and score the submissions, and their decisions are final and binding. The Judges will review all eligible submissions received and select three winners based on the following aspects of the Challenge:
- For the first two tracks of the sepsis use case (onset prediction and mortality), algorithms that have the highest discrimination power (e.g., Area Under ROC curve), as determined through test data by the Judges.
- For the third track, the most innovative solution, as determined by the Judges, will be selected.
The following prizes will be awarded by the Host/Sponsor to the winning Challenge Participants:
- Track 1, Sepsis Onset Prediction: $7,500
- Track 2, Sepsis Mortality Prediction: $7,500
- Track 3, Innovation (the most innovative solution derived from either Track 1 or Track 2): $7,500
Winning Participants are responsible for all applicable taxes, required reporting related to any prize received as part of the Challenge, and the distribution of prize money among team members.
Privacy and Participant Data
Participant agrees that the Host may collect, store, share, and otherwise use personally identifiable information provided during the registration process and throughout the Challenge for the purposes of general communication and promotion, as well as to convey additional opportunities to the Participant.
The Challenge Team Member Specifics and Institutional Support Form can be downloaded here.
For each registered team entering the Challenge, a secure link containing credential information will be provided, allowing team members to log into our reserved AWS instances. This credential may be shared among team members, but cannot be disclosed to any person who is not a member of the team. The login will be a two-step process: You will log into the prescribed server and make a subsequent connection to the assigned AWS virtual machine.
Note that each team is allocated a budget of $1,000 in AWS credits for the purposes of the Challenge. Initially, each team will have a small GPU instance p2.xlarge ($0.9/hour) to become familiar with the data. The team can build an initial solution on this instance and then request a one-time switch to p3.2xlarge ($3.06/hour) to speed up their training and scale-up their model for the final stage of the competition. Please send an email to DII.NDSC@uth.tmc.edu in order to make the request to switch. If there are any questions, we recommend that you to refer to the FAQs accessible through the Challenge website; if your question has not been addressed, please send us your query via the Challenge e-mail address, DII.NDSC@uth.tmc.edu, so that we can provide you a response and update the FAQs section.