Our Critical Care Datathon, in partnership with MIT Critical Data and Google Cloud, took place at Imperial College London last weekend (08/09/18 – 10/09/18).

This datathon was a learning forum where multi-disciplinary teams explored large critical care datasets to answer specific clinical questions.

The day began with several clinicians pitching problems in healthcare which could be explored by using data.

These pitches ranged from looking at oxygen levels to determine whether oxygen is medicine or poison, trying to understand the obesity paradox in intensive care, and examining the relationship between lactic acid and sepsis.

After the pitches, the 50+ datathon competitors were arranged into eight teams: 
  • Harmony
  • Chloride
  • Circadian
  • Boarding
  • Sepsis
  • End of Life
  • Oxygen
  • Blood
  • Obesity
  • ATB

Throughout the datathon they used MIMIC III, a freely accessible critical care database, to try and answer their research questions.

Dr Leo Anthony Celi, one of the creators of MIMIC, was on hand to guide the datathon teams.

Reflecting on the event, Dr Celi said: “It is always invigorating to watch AI research in action and witness data scientists and clinicians learning from each other.

“The deliberate focus on reproducibility during this datathon is worth mentioning, because it is crucial if this field were to truly advance.

“We are already drowning in claims of superior algorithms and sensational discoveries as machine learning becomes mainstream.

“But if there is no mechanism in place to replicate and validate the models, it is possible that the AI successes that we celebrate are just fluff.”

In addition to running the AIMed datathon, Dr Celi has contributed an article on data regulation and ethics committees to AIMed Magazine Issue 04, which you can read here.

Explaining the structure of the Datathon, Dr Celi said: “We try to make sure that the teams are balanced in terms of the different expertise being represented in each team.”

He explained that datathon teams spend their first hour “fleshing out the question, coming up with the design of the study, trying to look for what variables represent the outcomes that they’re interested in, the exposure that they’re interested in.

“So that would take about an hour and then they dive in and start extracting the variables, looking at the distribution of the variables – Does it make sense? Does it look real to you or not? – before they can proceed with the study.”

Our datathon teams were also guided by additional experienced data analysts:

Will Knottenbelt – Professor of Applied Quant. Analysis, Imperial College London

Melek Somai- Physician & Data Scientist, Faculty of Medicine, Imperial College London

Matthieu Komorowski – Physician & Data Scientist, Faculty of Medicine, Imperial College London

Tom Pollard – MIT, LCP

The winning teams were:
  • Team End of Life
  • Team Obesity
  • Team Chloride

They have been invited to give presentations at AIMed Europe this week. It’s not too late to sign up!