AI in Healthcare: From evaluating strokes to predicting patient downturns, AI is transforming health care

At the Yale New Haven Comprehensive Stroke Center, physicians like Dr. Joseph Schindler can use artificial intelligence-enhanced technology to evaluate patients who are suffering from strokes. In real-time, the technology helps physicians see what’s happening inside a patient’s brain, locate blockages, evaluate damage and determine if more brain tissue is at risk. It’s all key information for deciding on a treatment plan, according to Schindler. Artificial intelligence is being used throughout Connecticut’s healthcare industry — to evaluate strokes, detect tumors, develop drugs, and help medical providers with mundane tasks so they can focus on critical patient care. And that’s just a fraction of how the healthcare industry is using AI now — the potential is growing. Industry leaders statewide are investigating new ways to leverage the technology, such as using generative AI to draft initial responses to patient medical queries. AI is expected to help physicians predict patients’ future health problems, so they can recommend preventive measures. Dr. Lee Schwamm, senior vice president and chief digital health officer for the Yale New Haven Health system, said AI is frequently used at Yale for imaging and diagnostics, helping doctors evaluate images and identify and quantify abnormalities on tests, ranging from mammograms and colonoscopies to electrocardiograms. Lee Schwamm Global medical technology manufacturer Medtronic’s “GI Genius” is used to help physicians detect colorectal polyps during colonoscopies. Medtronic, which has a Connecticut presence, reports the AI-enabled technology boosts benign tumor detection by up to 14.4%. “(AI) does a very good job of highlighting areas of abnormality — though it has to be reviewed carefully by a radiologist or specialist because (artificial intelligence is) not perfect,” Schwamm said. Dr. Barry Stein, chief clinical innovation officer at Hartford HealthCare, said AI has many clinical uses. In stroke victims, he said algorithms can help predict patients who are most likely to develop secondary strokes, which allows medical professionals to intervene with preventive measures. For oncology patients, Hartford HealthCare has developed an algorithm to help predict which patients are at highest risk of becoming severely ill following chemotherapy. Hartford HealthCare has also developed a partnership with the Massachusetts Institute of Technology to help translate AI into clinical practice and improve patient outcomes. The AI market in health care was valued at $15 billion in 2022, and is expected to reach $188 billion by 2030, according to Statista. The growth could mean cost savings for the industry. According to a January report by McKinsey & Co. and Harvard University, wider adoption of AI could lead to savings of 5% to 10% in U.S. healthcare spending, which translates up to about $360 billion annually. AI into the future OpenAI’s ChatGPT, launched in 2022, processes large amounts of data to produce human-like responses to questions. Yale, in collaboration with Wisconsin-based electronic health record vendor Epic, will be experimenting with AI’s use to create a first-draft response to a patient message in a patient portal, according to Schwamm. If a patient emails or messages their doctor with a question — such as whether their potassium level was normal on a blood test — generative AI can draft an initial response. The provider can then review, edit and send it. This saves time for caregivers to focus on other duties. “We are about to start piloting that project in the very near future,” Schwamm said. AI will also potentially play a role in care navigation and patient interaction. Instead of calling to request a doctor’s appointment, a patient might first interact with conversational AI. The patient would describe their symptoms, like eye pain, to an AI platform. The AI would ask relevant questions, like whether the eye appears pink, or if the patient has experienced eye trauma. It would then suggest a possible diagnosis, such as conjunctivitis, and connect the patient to virtual urgent care services, according to Schwamm. Yale is exploring multiple vendors to bring AI into use during medical visits, too. The goal, Schwamm said, is for artificial intelligence to capture medical interview and office visit information and craft it into an organized, structured note that is ready for human editing. “Those are just a few examples of where I think we might start to see really powerful use of these tools,” Schwamm said. “The other is to prompt physicians into thinking about various possible diagnoses based on a list of the symptoms the patient is complaining about.” AI and machine-learning tools can also help identify deteriorating patients, leading to faster analysis and intervention, Schwamm said. “What we haven’t seen yet are situations where AI detects a problem, triggers an intervention and then resolves the issue without human intervention,” Schwamm said. He foresees AI adjusting insulin pumps when someone’s sugar is too low, or alerting people about their blood pressure. Stein anticipates that AI will help predict what patients’ future health trajectories look like, taking into account a person’s health history and demographics. “It can help us predict what is going to happen to our patients, and help us prescribe what is best for our patients,” Stein said. Administrative tasks, payment integrity AI is also helpful with routine, clerical administrative tasks, which can free up employees to do other duties, Schwamm noted. Artificial intelligence is expected to take over many administrative tasks at Yale, such as accurately coding visits based on the care given and submitting paperwork for reimbursement. “We have a terrible workforce shortage problem in health care, and it’s only going to get worse,” Schwamm said. “We don’t have enough doctors, nurses, pharmacists, etc. So, we have to figure out ways to support our providers being more effective and more productive with their time and to focus their attention on more complicated tasks.” At Hartford HealthCare, AI is helping with scheduling nurses, to match their areas of expertise to where they are needed, while factoring in preferred shifts and vacations. AI can finish a scheduling task within minutes, while it would have taken a human several hours, Stein said. Bristol Health CEO and President Kurt Barwis said many insurance companies now use AI to trigger payment denials for healthcare claims. Bristol Health is investigating how it can use artificial intelligence to combat this, as AI can write appeal letters in seconds. The technology could potentially mean more insurance money for the health system. It would also free up time for staff to spend on other tasks, he said. Goodroot is a Canton-based parent company to numerous healthcare technology firms, including Penstock. Insurance companies employ Penstock to help evaluate healthcare claims and ensure billing accuracy. Markus Waite Markus Waite, Goodroot’s chief information officer, said as more medical providers use AI to capture their notes, Goodroot’s family of companies are seeing higher-quality medical records. The company is exploring AI’s use to comb through records and highlight areas of concern for further review, such as if there are improper codes on forms. “We are being very cautious. We want to make sure that our companies don’t rely on the AI results as a crutch, or as the record of truth,” Waite said. “We’re making sure it’s just to use and highlight areas that they may need to look at, but not to make a decision.” Challenges, words of caution Several healthcare industry leaders said human interpretation of whatever artificial intelligence presents is important. Computers may be able to answer questions, but they are still basing their conclusions on information they’ve been given. If false information has been provided, AI may respond with answers that seem true, but aren’t, cautions Stein. “If people start acting on (bad information), quality and safety can be compromised,” Stein said. Education about AI is key for clinicians, as most don’t learn about it in medical school or during early career training, notes Stein. Hartford HealthCare is focused on ensuring its physicians are well-trained on AI, and know how to recognize problems with the technology. When AI identifies a potential problem area, such as during an examination, human involvement is still key, according to Yale’s Schwamm. Michael Urban, senior lecturer and director of the University of New Haven’s doctor of occupational therapy program, said he believes AI will lead to better quality of care, but also urges caution. “AI doesn’t know everything, and a trained provider needs to be making sure the best decision is made,” Urban said.