Which patients will crash? How AI is powering early-warning programs and other healthcare innovations

At St. Michael’s Hospital in Toronto, the healthcare teams that look after hundreds of patients each day get an extra helping hand from a relatively new workplace entity: an artificial intelligence-driven program that flags patients at risk of unexpectedly “crashing.” “Physicians, nurses and other professionals in the hospital don’t always recognize that the patient in front of us is getting sicker,” Dr. Amol Verma, an internal medicine specialist at St. Michael’s Hospital, explained during an Ontario Medical Association media briefing on AI in healthcare. “So the question we asked was: Are there some subtle clues or signs that artificial intelligence might be able to detect earlier, to alert us to those patients who might be deteriorating?’” The answer led to the hospital developing CHARTWATCH, an early-warning program that looks at 100 variables in a patient’s chart—including vital signs and lab results—and determines if a patient is at a low, moderate or high risk of needing intensive care within the next 24 hours. “It takes all that data in the electronic medical record here at the hospital and every hour generates a new prediction of which patients might become so sick that they need intensive care or die,” said Dr. Verma, a professor in AI research and education in medicine at the University of Toronto and co-lead of the GEMINI data collection and analysis project at St. Michael’s Hospital. “We implemented our CHARTWATCH tool around October 2020 and, compared to the time before we had the tool in our unit, we’ve seen a 26% reduction in unexpected deaths.” Read: Your medical notes delegated to generative AI CHARTWATCH was just one of several existing and future AI applications in medicine highlighted during the Sept. 20 OMA briefing. Dr. Ibukun Abejirinde, a scientist at Women’s College Hospital Institute for Health System Solutions and Virtual Care in Toronto, talked about an in-the-works project that uses AI to predict and prevent diabetes. “Our team members have been … developing models that can predict up to five years in advance the likelihood that somebody will be diagnosed with Type 2 diabetes, or that they will suffer a complication from it,” said Dr. Abejirinde. “Those models have been validated. Their performance has been established.’ With the development phase of the project now finished, the project team must now address two important issues: social acceptability of the use of AI in healthcare—an issue that takes into account questions regarding ethics and privacy—and the challenge of scaling the benefits of AI to a large population. “So our plan is we're going to develop a framework using a very participatory, stakeholder-driven approach working with people within the community,” said Dr. Abejirinde. “The framework is going to feed into a live output whereby the makers will have a dashboard that will allow them to see: who has the highest risk for patients, where do they live, what are the chances that patients in some specific communities, or with this particular history are at risk of becoming pre-diabetic, or have been diagnosed with Type 2 diabetes?” Read: A therapeutic revolution: What will the treatments of the future look like? The goal, she added, is for policymakers to be able to use the dashboard to design healthcare systems with a preventive approach. “And if we're not able to prevent (diabetes) from happening, how do we ensure that we're improving the quality of life of those who get diagnosed with diabetes?” she said. For Canadians to take full advantage of what AI can offer in healthcare, policymakers need to ensure that while they have measures in place to protect patients’ health data, these safety measures don’t become roadblocks to innovation, said Dr. Fahad Razak, an internist at St Michael’s Hospital and Canada Research Chair in Healthcare Data and Analytics at the University of Toronto. “The protections themselves shouldn't be a barrier to using this data to improve people's lives at the end of the day,” he said. “All of this is a social contract—it’s a trade off of risks . . .. There is no 100% risk-free way of having this kind of data stored . . .. But we have to come to a social contract where we don't have protections absolutely to the extent that they reduce or eliminate the ability to generate benefits for patients for the healthcare system.” Dr. Verma said the right AI application can also make a big difference in a physician’s practice. Instead of spending hours each day reviewing and updating patient records, what if doctors could leverage the latest generation of language models in AI to retrieve specific patient information, capture appointment notes and advance actions such as specialist referrals? AI has also shown great potential as a diagnostic tool, he added. “We have already seen diagnostic technologies enabled with artificial intelligence that improve our ability to do things like detect cancer by 15%-to-20% over current methods with just humans alone,” said Dr. Verma. “So I think it’s incumbent upon us to incorporate those technologies into things like colonoscopies for screening colon cancer and breast cancer screening programs to allow us to prevent or detect diseases earlier and have better outcomes.” Read: Top 5 takeaways for med students to be AI-ready

Which patients will crash? How AI is powering early-warning programs and other healthcare innovations

Register for free access or login.

Related Topics