New Zealand’s first ever health conference looking at artificial intelligence was organised to give the health system “a shake-up”.

Dr Marise Stuart, an organiser of the Hack Aotearoa conference, said New Zealand’s heath system needed to “pick up its game” in the use of artificial intelligence (AI). Last week’s conference explored how predictive data, smart technologies and robotics could improve the health of New Zealanders.

“There’s a junior doctors' strike going on now and a lot of that could be prevented by having good IT because junior doctors spend so much of their time typing. That’s just one small snag in the health system that AI could hugely have an effect on.”

Over two days, the 200-plus attendees heard examples from overseas and local speakers of how AI could assist doctors, as well as of risks involved with collecting and using data.

Scripps Research Translational Institute Professor Eric Topol, described as a guru in digital medicine, said the human side of technology was helping us to understand it better than ever before.

“All these layers of information, the physiology from sensors, the anatomy from scans, the different biologic layers from DNA, protein, metabolites, the microbiomes, the epigenome. If we can have all these layers of information for any given person, we could deliver far better care and prevention.”

Computers being able to pick up information from eye scans when humans couldn't was a good example, Topol said. 

“If you give international retina specialists a photograph of the retina and you ask them 'Is this from a male or a female?', the response is 50-50. They can’t do any better than that. For deep learning algorithms the accuracy is 97 percent. That’s pretty astounding.”

Another example was detecting an eye disease related to diabetes. Diabetic retinopathy can lead to blindness and is often underdiagnosed. An AI system capable of recognising 90 percent of diabetic retinopathy cases has now been developed.

“Now the receptionist in a doctor’s office can obtain retinal images and through deep learning and cloud-based analytics the determination of what grade of retinopathy can be determined."

However, Topol feels the hype of AI has exceeded the science.

“We’re short on validation and long on promise. Ultimately we would like to see the implementation, which really hasn’t started except in a few pockets around the world.”

He stressed the need for peer-reviewed studies and real-world, clinical validation.

A challenge for AI is that healthcare's existing risks could be exacerbated depending on the data used to train an AI system. In a paper published in Nature, Topol gave the example of algorithms used to diagnose melanoma which lack the inclusion of skin colour.

Massachusetts Institute of Technology Laboratory clinical research director and principal research scientist Dr Leo Celi works with data and algorithms. The topic of algorithms working for all populations is close to his heart.

“If you feed a computer biased data, you’re going to have biased algorithms. Every time we create a model, we make sure the model is performing consistently across different populations.”

This testing needed to be repeated to ensure it kept its accuracy over time, and algorithms shouldn't be patented because they have a limited application, Celi said. Instead, revenue for algorithm creators should lie in continuously testing and recalibrating an algorithm, he said.

Another focus of the Hack Aoteroa conference was Māori health, and how big data could be used in an intelligent way to “hack” the health system to improve it. 

Andrew Sporle, a Māori researcher who reviews big data projects, shared concerns around the use of data.

“Māori death rates are still 1.8 times higher than non-Māori. Is that due to the healthcare system? Well, yes.”

He said more of the same would not solve inequity: “We need to focus on disparities, or we risk making them worse.”

Collecting data now could help prevent problems for Māori youth as they age. However, Sporle urged caution around this.

“We need to start thinking about what is the process of this engagement? Is it based on social licence? Or is it a basis of partnership or ownership and control?

“The top-down approach we’ve had so far hasn’t delivered.”

A Treaty model based on engaged partnership was what we needed. 

“Data collection without reciprocity begins to look a lot like surveillance.”

Without clear benefits, and results matching the hype, he worried it would be “game over” for funding, and for people’s willingness to share their data.

New Zealand’s biggest insurer - and holder of many years’ worth of injury data - was a major sponsor of the event.

Accident Compensation Corporation CEO Scott Pickering said the use of data had become more critical in ACC's day-to-day operations.

“We are part of the wider trend in insurance, shifting from detect and repair to predict and prevent. We’re spending millions more on injury prevention each year and we are increasing the use of our data analytics.”

Data about injuries that happened within hospitals while patients were receiving treatment could be used to stop future hospital-caused injuries, Dr Leo Celi told the conference.

“You could build a system where an algorithm could in real-time predict a treatment injury," said Celi.

“Nothing exists in the world like this.”

SOURCE : https://www.newsroom.co.nz/2019/01/21/407803/will-big-data-make-everyone-healthier