There is still something slightly futuristic about machine learning. After all, it’s less than a decade since Hollywood thought it was worth boosting the sci-fi credentials of a movie by calling it A.I. Artificial Intelligence.  

However, the use of AI or machine learning products is becoming more and more common in healthcare, not least as a result of the Covid-19 pandemic. Governments and planners have turned to computer models to project the progress of the disease, calculate risk scores for patient cohorts and individuals, and support clinicians in triage and treatment decisions.  

NHS context  

Bruce Horne, product specialist at Orion Health, told the last of its 2020 series of autumn webinars that these trends would continue. “The NHS Long Term Plan talks about the importance of technology, and especially AI, to improving outcomes and efficiency,” he said.  

“National bodies have been making awards in this area, and NHSX has created an NHS AI Lab. We are also seeing customers who are interested in AI. For example, one of our customers Connecting Care in Bristol used AI to score Covid-19 risk and made the tool available through its portal”

Orion Health has a range of products and services to support these developments. The webinar focused on Machine Learning Manager and Algorithm Library; although Orion Health also offers data de-identification tools and document tagging that makes it much easier to turn free text into searchable, machine readable data that can be mined for further insights.  

Orion Health Machine Manager and Algorithm Library  

Machine Learning Manager is important because, as Horne went on to point out: “An algorithm has no value without effective deployment.”  

As things stand, organisations that want to roll-out machine learning products need to come up with their own model for working out how to get them into the right hands, find out what is happening to them, and keep them up to date. But MLM can do the job for them.  

MLM can handle the algorithms and calculators in the Algorithm Library or those developed by third parties. In all cases, it will look after their administration and deployment, whether that is through a dashboard, or a portal, or as a data extract.  

The Algorithm Library, meanwhile, gives users access to a long list of locally configurable algorithms and calculators that provide ‘out of the box’ predictive insights into areas such as risk of readmission and surgical outcomes. But it can also be used to store and manage third-party tools.  

Think first, and don’t forget change management  

Even with this kind of support, machine learning and AI is a complex business. “All of the above looks great, but there are some logistical considerations,” Horne said, in a section headed ‘important considerations’.  

“The first thing is to have a strategy, driven by local needs,” he said, “because then you can consider how AI can help you to meet challenges instead of [coming at things] the other way around. The second thing is to start simple: deploy something well and understand it fully before you move onto something more complex.  

“And the third thing is to understand that there is a change management task: you need end-users to understand these tools and how they will impact their workload and working practices.” As in so many areas of health tech: “Clinical engagement is critical.”  

Finally, users need good local governance practices for the storage and control of data, they need good infrastructure, and they need good data. “Data is key,” Horne emphasised. “You need to understand your data, and when it is good enough, and how you can continuously improve it.”  

The New Zealand Algorithm Hub  

To help customers that want to move into the world of machine learning and AI, Orion Health has an Intelligence Services team. Pieta Brown, senior data science consultant, explained that it can provide data science consultancy and support on strategy as well as specific development skills.  

Her colleague, Genevieve Dawick, then explained how it had helped the New Zealand government to develop a cloud-based, MLM-based, Algorithm Hub to support the fight against Covid-19.  

“In New Zealand, we have had a fairly successful Covid-19 elimination strategy,” she said [the country implemented an early travel ban, lockdown and test and trace strategy, and at the time of the webinar had seen just 2,078 cases of the disease and 25 deaths].  

“Because of our unique situation, our focus has been on three areas: Covid disease spread, models to help clear [elective care] backlogs, and preparedness if outbreaks were to occur. We also created some simple calculators in case they were required.”  

Learning lessons, tracking concept drift  

One of the challenges that the development team has faced is that New Zealand has had too few cases of the disease to fully validate the tools.  

However, the team has focused on making sure that they should work in a New Zealand context and not adversely impact on any of its population groups, including the Maori population that has traditionally experienced worse health outcomes.  

Dawick said it has been working with a governance group that “really helps keeps us on our toes about what is deployed and how it should be used.” The hub is now being updated so it can be used on a ‘business as usual’ basis. New context is being added to the tools that are in it and additional calculators are being added. 

Experience tracking is also being added, so the team can assess the performance of the different modules and keep on top of ‘concept drift’. “This is important,” Dawick said as the webinar drew to its close. “You need to track the performance of the model because it can change over time.  

“Changes in lockdown regimes is a good example. They change, so you need to refine the model to make sure you are keeping up with these different factors over time.”  

Improving patient care, saving lives  

Because the Orion Health Machine Learning Manager makes this kind of administration achievable, it supports the very non-sci-fi aim of using AI and machine learning in healthcare to improve clinical decision making and, with it, efficiency and quality. Or, as Horne summarised, to “improve patient care and save even more lives.”