When National University Health System (NUHS) switched on DISCOVERY AI in 2018, they never thought the journey would see them create a new benchmark for patient care.
DISCOVER AI had a simple goal. “It was essentially our R&D platform to design and aggregate large amounts of clinical data and share with a group of researchers and clinicians,” says professor Ngiam Kee Yuan, who is the group chief technology officer of NUHS.
Beyond just a data destination, it also allowed researchers to build and test AI and ML tools. “On DISCOVERY AI, you could test these tools to near-production,” says Ngiam.
Why near-production? That’s because the tools still need to integrate with the various electronic medical record (EMR) systems. And to some extent, DISCOVERY AI acted as a sandbox to research AI and ML models.
When Singapore deployed a new EMR system on Epic Systems, called National Electronic Health Record (NEHR), Ngiam’s team decided to think broader.
“We did not want to run one tool; we wanted to run hundreds of AI tools in production. So we decided on an automation and AI platform so that it can run in real-time and do backend calculations and deliver the next generation of care,” says Ngiam.
ENDEAVOUR AI was born.
Taking the microservices route
The platform’s foremost goal is to reduce data wastage. NUHS wanted to make the same data available for different AI and ML projects.
“Every project will need different sources of data that may overlap. So if you do it individually [for each project], it can be extremely expensive,” says Ngiam.
So, Ngiam’s team approached the project from the end users’ point of view. In AI, this requires a fair bit of complex feature engineering.
“Healthcare data is pretty complex. So how do you reduce that complexity? You need to do it with clinicians; you cannot do it with engineers. It requires specialist attention,” says Ngiam.
Together with clinicians, NUHS plans to deploy as many as 150 distinct AI and automation tools as microservices on ENDEAVOUR AI. These AI tools incorporate multi-domain patient information, such as demographics, text, images, lab data, and medications prescribed.
Ngiam believes the new deployment is expected to translate into significant cost savings, from a patient’s care at admission to predicting a patient’s length of stay and optimizing scarce bed resources.
One area where the new platform shows its mettle is breast cancer detection. The AI tool instantly identifies risk factors for breast cancer, and, if needed, the patients concerned are referred for a mammogram and specialist treatment.
Re-architecting for efficiency
To ensure ENDEAVOUR AI can meet its high goals, Ngiam needed the data architecture to be highly available and modular.
TIBCO Software’s Business Works offered NUHS a zero-code model-driven environment to reduce development complexity and accelerate time to market. “It is a very capable software in doing a number of data transformations. We did not want ETL. It converts the incoming data into a Kafka stream to broadcast to other AI tools,” says Ngiam.
TIBCO Streambase helped NUHS automate workflows. “It provides the actual logic,” says Ngiam. Meanwhile, TIBCO Spotfire helped the clinicians visualize the data analysis.
Ngiam noted that TIBCO Software’s choice was based on its ability to offer an enterprise-grade integration solution to design the workflows. The ability to integrate and support models written in other codes made the vendor’s solution attractive.
"Healthcare institutions aggregate vast quantities of data, but most of the data collected is only analyzed retrospectively. TIBCO technology enables the NUHS ENDEAVOUR AI platform to stream data in real-time, feeding live data into AI models that produce actionable insights on the fly,” Ngiam adds.
For Ngiam, ENDEAVOUR AI’s journey has only just begun.
“One of the challenges we have right now is to have the entire system work well with different models. Working with hundreds of different models on a streaming basis creates a lot of complexity. So you need to design the system in a way that aggregates the outputs using a certain logic,” he explains.
Another is transferability. “It’s not simply plug and play [for healthcare systems across geographies],” says Ngiam. For example, chest X-ray images may be transferable, but text data may not be because “the way we write can be completely different,” he points out.
“This requires re-training of the models, and such requirements need to be addressed in the system architecture design,” he further adds.
Whatever the challenge, NUHS is confident that its ENDEAVOUR AI platform will shape the future of AI-based healthcare.
Winston Thomas is the editor-in-chief of CDOTrends and HR&DigitalTrends. He is always curious about all things digital, including new digital business models, the widening impact of AI/ML, unproven singularity theories, proven data science success stories, lurking cybersecurity dangers, and reimagining the digital experience. You can reach him at [email protected].
Image credit: iStockphoto/Blue Planet Studio