Alex MacLeod, manager, HealthShare commercial initiatives, InterSystems, followed Yossi Cohen and his ideas for a ICS prevention cookbook with a presentation on machine learning. But she started by showing a video focusing on Manifest Medex, a California organisation that is “quite similar to a local health and care record exemplar” that is running a health information exchange, built on InterSystems technology.
She explained this was because Manifest Medex has gone from simply sharing information to using it for predictive analytics. And that is the direction of travel. “For the past few decades, we have been focusing on collecting and integrating data that you can surface through a UI or a notification or you can push it to the patient and their care givers,” she said.
“It was beautifully demonstrated by the Lincolnshire Care Record. But what we need to do now is take that information and start to understand trends and opportunities and also use it for machine learning and the like to create a more accurate presentation of the future.”
What is innovation in healthcare
After all, MacLeod argued: “Innovation is supposed to look at the big problems.” The big problems in the NHS are satisfying demand, and administration cost. And AI is one of the biggest innovations in healthcare. “So we have been applying it to these big problems.”
Specifically, she said, “we have been looking at appointment no-shows; which cause problems because they leave clinicians “idle”. Data will show how many patients are likely to DNA for any one clinic; but it won’t tell an administrator how to do what airlines do, which is ‘overbook’ intelligently to take account of that. Whereas data science can take many, many variables and come up with a program to do that.
Also, she said, “we have been working on readmission”, and how to stop patients from being readmitted by focusing on interventions during the discharge process. Traditionally, she said, US hospitals use a calculator called LACE, with four variables, to estimate this.
But a project that tried the intervention on all patients, that used the LACE score, and that used a machine learning approach found intervening on all patients is expensive and ineffective, that LACE prevented two readmissions, and that a machine learning approach prevented 2.4.
In money terms, she said that in comparison to LACE “you are saving $2,000 more. Which doesn’t sound that great. Until you apply it to one of our customers, which is discharging 785 patients per day. They save $15,000 a day, or $5.5 million a year. And that is worth doing.”
Of course, this is not easy. On paper, it would be more or less impossible. But InterSystems has created tools that make it much easier to capture and manipulate data, and it is now releasing specific tools to customers. “So, we have built a product called QuickML, that automates machine learning for you in the background, and that is in IRIS, it is in IRIS for Health, and it is available for you,” MacLeod said.
“We have also created a product, called Patient Index, that matches records. And for patient insight, we have started to build all sorts of solutions… including something called Clean Data as a Solution, that cleans the data and leaves you free to do work on top.”
To close, she showed another video, this time of Concerto Health AI, which uses the clean data solution to support its cancer research. In questions, Mike Fuller, discussed the issue of openness and bias. MacLeod said it was “essential” to be open about how algorithms were constructed and used: and to be very aware of the potential for biases. Otherwise, Fuller said, “there is a theme today, about people getting trapped into silos and biases, and that is a danger in this area.” MacLeod: “Absolutely.” And to make it easier to understand, she said, InterSystems is providing lots of courses to help people along the way.