Criteria for Running a Successful Challenge
Each Challenge requires substantial time, effort, and expertise to run. A well run Challenge provides valuable research and training opportunities for the Challenge organizers and participants alike, while a poorly run Challenge can reflect poorly on our community and even cause harm.
We appreciate when we receive suggestions that we run a particular Challenge, and we want them to be aware of the requirements or running a Challenge. This list is not exhaustive, and we note that the some of the tasks in this list often take several months or even years to prepare:
- Challenge data
- Sufficient, high-qualify, real-world data
- Clinically meaningful labels
- Curated in collaboration with domain experts
- Public training set and hidden validation and test sets
- Clearly documented
- Appropriate and standard file formats
- IRB approved
- Fully deidentified
- Ethically sound
- Challenge objective
- Meaningful, clinically relevant, and relevant to PhysioNet
- Clearly defined and stated in collaboration with domain experts
- Challenge evaluation metrics
- Clinically relevant
- Capture the Challenge objective
- Clearly defined and stated in collaboration with clinical domain experts
- Challenge algorithms
- Publicly available example implementations available in common languages with clear documentation and instructions
- Flexible to encourage a diversity of approaches
- Restrictive to ensure that algorithms address the defined problem
- Challenge timeline
- Challenge data, objective, evaluation metrics, and algorithms must be available from the beginning of the Challenge
- Run time must be at least 3 months, with at least 6 months preferred
- Challenge deliverables
- Clear definition of scientific outputs
- Generate publicly available software
- Archive both public training data and hidden validation and test data at PhysioNet
- Publications and code links provided to PhysioNet after the Challenge
- Code for Challenge evaluation metrics must be submitted to PhysioNet
- Affiliation with a public scientific event recommended
- Challenge organizers
- PI(s) explicitly assume responsibility of Challenge, including the throughout and after the Challenge
- Sufficient and responsive staff
- Sufficient computational resources for at least 100 teams and 1000 code submissions
- Successful completion of a previous data science competition
Supported by the National Institute of Biomedical Imaging and Bioengineering (NIBIB) under NIH grant number R01EB030362.
© PhysioNet Challenges. Website content licensed under the Creative Commons Attribution 4.0 International Public License.
Back