The tech lounge opened on Monday morning with a bang – loads of giants in the field chatting and presenting in an informal zone. Super busy, people everywhere – standing room only!
Leo Celi from MIT gave an inspirational talk about data – its impressive what he has set up.
He recommends this (free online) book that he and others wrote if you are interested in the field (http://www.springer.com/gp/book/9783319437408) and he is clearly passionate about getting people involved and doing “big data” properly – hackathons and datathons galore in europe and abroad (plug alert – if you are interested in them see this one in London in December with Mervyn Singer et al.)
Another good editorial that Leo recommended to read is this one in the NEJM (not paywalled!) http://www.nejm.org/doi/full/10.1056/NEJMra1614394
He made the point that AI has come a long way, but we need to be careful to do it properly so as not to over sell the field, which is still in its infancy.
Successes he quoted include Articial intelligence diagnosing diabetic retinopathy but he was scathing about IBMs Watson – calling it a “digital canary” and the AI equivalent of a mechanical turk (a sham chess “computer” with a human inside) His point was that it is not truly harnessing data on its own, and so should not be claimed that it is – although important on the way to doing that.
Then Derek Angus gave a really easy to understand run through of how big data and RCTs can coalesce and we can really harness the power of data to help us do what we do.
Derek talked about two trials that are currently running “within” the electronic health records systems in the USA, which cost 10% of the equivalent traditional RCT so you can see why these techniques are attractive.
His concept is to remove the bias from randomisation alteration by clinicians (based on initial results of studies) and handing that over to computers in a concept known as “response adaptive randomisation”. I have tried to capture it in this thread:
An exploration of what questions big data might answer followed from Theodoros Kyprianou. What is big data?:
He took the gathered crowds through the different ways machine learning can occur on the ICU (and in healthcare generally); to recap there is
Supervised learning – data given to computer and outcomes known, computer tries to sort data to predict outcomes
Unsupervised learning – data given to computer and outcomes not known, computer tries to sort data into its own groups
Reinforcement learning – computer given reward structure if certain outcomes met – aims to maximise reward and discover easiest route to outcome
He talked about the sort of things where this might be useful to clinicians – for example in making simple choices and decisions, or creating healthcare and illness classifications or even making diagnoses. An interesting possibility was letting machine learning do the “physiological fine tuning” on a unit.
Overall this session was really fun and it was great to hear from all the speakers on how machine learning is being used currently but also “future gaze”; inspiring to think what our units and hospitals might look like one day!