Healthcare must become better at bridging the gap between those in clinical care and those who are designing artificial intelligence (AI) systems if AI's full transformative potential is to be reached.
That was the verdict from an illustrious panel of global digital health leaders at the sixth Asia Pacific HIMSS-Elsevier Digital Healthcare Awards, held earlier this month in Brisbane, as part of HIMSS AsiaPac18.
HIMSS Analytics Global Vice-President John Daniels said living in the age of AI promises limitless potential benefits if we are able to transform vast pools of data into useful knowledge and insights.
Elsevier Product and Partnership Director Tim Morris said collecting that data was no longer a challenge – instead, the two key challenges are now structuring and analysing that data to provide valuable information, and creating systems that provide information in useful ways.
"It's not just about the digital interaction; it's about providing information at the right time and the right place for someone within their workflow and augmenting what they are doing, rather than trying to take it away with a ‘black box’,” he said.
Royal Children's Hospital Chief Medical Information Officer, Professor Mike South, said he was not the slightest bit anxious about being replaced by an AI-driven machine.
"There are plenty of things for us to do, but there are plenty of things that machines or algorithms can do very effectively," he said.
"If you just think about diabetes monitoring in hospitals, we have nurses measuring blood-sugar levels. If we had those nurses ringing the doctor with every result, we would need teams of doctors just answering the phones and saying, 'Give five units of insulin, etc.' in response to that.
"But we don't, because we worked out an algorithm that we put up on a wall and that the nurse refers to. That sort of high-volume, low-value stuff doesn't need a doctor's input. Let's automate all of these things and let's preserve doctors' skill for things that we really need."
However, South said for this to happen, there needs to be more dialogue between patients and clinicians on the ground and those designing AI systems.
"These things are currently operating in two worlds. I hear people who are not connected to clinical care talking about AI, and people in clinical completely confused about what it is. We need people who can bridge that translational gap,” he said.
Morris said one of the most promising imminent applications of AI in clinical care is around early diagnosis, a point echoed by South.
"People sit on a waiting list for months and then, when they have a hospital appointment, we ask them questions that could've been asked right at the beginning, that might have contained red flags for some serious problem," he said.
"We need much smarter ways to manage the huge demand."
Morris said a partnership Elsevier has in Germany holds exactly this sort of promise.
"We have access to around five million anonymised records in a database, which goes from early childhood to elderly. We have blood values, we have their diagnoses, their outcomes and information," he said.
"We're using machine learning to understand the disease trajectory within the population base and starting to understand the early identification and the differential diagnoses on what can be achieved.”
But there are gaps that still exists between the data that is available and the production of AI tools.
“It starts with data, with really good machine learning and understanding disease trajectory within that dataset, and then building out the tools later,” he said.
“I go to countries where they want to implement AI, but when we ask them where they are on their journeys in implementing hospital systems, they say they haven’t even got an EMR. So, there is that ‘we need to run before we can walk’ scenario.
“[At Elsevier], we believe in the basics, delivering systems that can actually make a difference. [For example], if you discover that you need to change the dosage of a drug for certain people, how do you implement that as a procedure within your hospital?
“The most simple way would be to implement an order set, which is evidence-based. You don’t need massive AI to do that, you just need basic EMRs,” he said.
To find out more on where you are on your Clinical Decision Support adoption journey please click here.