Artificial Intelligence Emerging as a Part of Medical Practice
Promises to relieve physicians of burdensome tasks and improve patient diagnosis and treatment
By Jim Braibish, St. Louis Metropolitan Medicine
From St. Louis Metropolitan Medicine Second Quarter 2024
Artificial intelligence—the next wave of technology— is emerging in many ways in medical practices and health systems in the St. Louis area.
From drafting exam notes, to helping detect cancers, to identifying patients at greatest risk for readmission, applications of AI are becoming more prevalent. AI promises to not only improve diagnosis and treatment, but also free physicians from burdensome tasks and allow them to focus on patient care.
Artificial intelligence processes massive amounts of data and uses algorithms, modeled after the decision-making processes of the human brain, to “learn” from this data and generate increasingly accurate results over time.
“AI has been around for a long time in medicine—as far back as the late 1950s. What has changed is the technology has finally caught up with the promise of what could be done,” said Philip Payne, PhD, the Janet and Bernard Becker Professor, chief data scientist, and director of the Institute for Informatics, Data Science and Biostatistics at Washington University School of Medicine.
Advances in the volume of available data and computing speed make AI more effective today.
“Artificial intelligence opens up a fourth dimension of care,” added Gautum Agarwal, MD, director of precision medicine for the Mercy system. “First there was just the physical exam, then basic laboratory values, then X-rays, and now we have genomics and artificial intelligence. Because these technologies are relatively inexpensive to deploy, they will make care more accessible to more people.”
AI-Based Scribe Function
With the advent of the AI-based writing tool ChatGPT, attention has focused on assisting physicians with one of their most burdensome tasks, clinical documentation. Each of the four area health systems is developing an AI scribe function.
BJC HealthCare and its academic partner, Washington University School of Medicine, are conducting pilots of several AI products to help generate clinical notes. Thomas Maddox, MD, SM, who leads the Healthcare Innovation Lab at BJC and Washington University, described the effort:
“We installed a product in the exam room that records the clinical encounter. At the conclusion of the visit, it feeds the audio transcription into a large language model like ChatGPT. Then that model transforms the recording into a clinical note following the format that physicians often use. The physician then reviews and edits the draft instead of starting with a blank page.”
Mercy is implementing its note generation project in partnership with Microsoft. SSM Health has an AI-based scribe function available to about 50 clinicians now, but hopes to expand the number.
“Ambient documentation is a game changer,” said Ann Cappellari, MD, chief medical information officer for SSM Health. “The physician is focused on the patient instead of the computer because the app is recording the encounter.”
The technology also captures greater detail about the patient than the physician may be able to recall later in the day when writing exam notes, she added. This could include personal tidbits. She also noted the importance of obtaining patient consent to record the exam.
At St. Luke’s, Darren Haskell, MD, chief medical officer, summed up the benefit of their AI-powered transcription service: “Our hope for this technology is that it will allow our providers to get back to having real conversations with their patients, rather than looking at a computer screen. We feel these tools will improve patient satisfaction and reduce provider burnout.”
BJC’s pilot has proven popular with patients and physicians, Dr. Maddox said, with 97% of patients expressing a positive response to the technology. One area that still needs improvement is reporting physical exam findings which are done by observation and not conversation. “This underscores how we still need to work on aligning technology and workflows,” he added.
A related project in development is auto-generating drafts of responses to inquiries through the patient portal. Dr. Payne explained, “Answering these questions can take a lot of time that our providers don’t always have. The provider edits the AI-generated response, so they don’t have to start with a blank screen.”
Identifying and Prioritizing High-Risk Patients
AI has the ability to process large volumes of data and quickly compare this to established clinical guidelines, noted Dr. Haskell. “These clinical decision support tools have been deployed in our primary care population health efforts, and in some of our specialty practices. AI systems can quickly process a patient chart and identify potential opportunities to improve patient care. These suggestions can then be brought to the attention of the physician at the time of the encounter, or even between encounters.”
At SSM Health Saint Louis University Hospital, AI is being used to help radiologists and residents prioritize which cases to read first.
“We have an AI algorithm that reviews our imaging studies and looks for about 10 very important diagnoses that if you catch early, you can do something to change the outcome,” Dr. Cappellari said. Examples include stroke and blood clots in the lungs.
Washington University and BJC are using AI to calculate risk assessment scores for readmission, mortality, disease progression and other concerns, said Dr. Payne.
He described a program for preventing sepsis. “We have sepsis risk scores today that allow us to assess a variety of factors including clinical, demographics, how they traverse the health system, and how they were admitted. We put these together using AI and provide feedback to our providers, so they can pay more attention to those patients who are at risk.”
Another Washington University Medicine and BJC effort is in the area of palliative care, where they are working to build tools to anticipate patients with an acute disease burden and high risk of mortality. “Before the patient is admitted to the hospital or has an invasive procedure, there can be a conversation around their goals for care, thus optimizing clinical decisions and patient preferences,” Dr. Payne added.
For the past three years, St. Louis Children’s Hospital has used AI to identify recent patients at highest risk of readmission or visit to the emergency department.
Dr. Maddox explained: “The model uses AI to identify kids at high risk for medical complications. Our team of nurses reaches out to them to see if we can help the kids and their families troubleshoot any potential risks. For example, in children with asthma, it might be making sure they have inhalers and know how to use them.”
Every day, nurses receive a report that shows patients with moderate to high risk scores, indicating patients who are at most need or might benefit from care management or coordination. Over 2,200 patients were served in this program in 2023.
Precision Medicine
AI is the foundation for precision medicine, the ability to tailor treatments to the individual patient based on their predicted response or disease. Both Mercy and Washington University have precision medicine programs.
“Through 20 years of electronic health records at Mercy, we have been able to generate massive amounts of really organized de-identified data on our patients that we can then utilize to make earlier and more precise diagnoses,” Dr. Agarwal said. “Early diagnosis is the key to preventing the spread of disease and subsequent hospitalization, major surgeries and intense therapies.”
To help build a strong foundation of data, Mercy has joined with Mayo Clinic to collaborate to use the most current data science and years of de-identified patient outcomes to develop high-value solutions and algorithms leading to more optimal care for patients.
One element of precision medicine is germline testing that checks for genetic mutations in one’s DNA that could lead to cancer.
“We know that 10% of people who have cancer will have some hereditary basis for it, meaning they were predisposed to it from genetics,” Dr. Agarwal said. Patients complete an online questionnaire—developed using AI—to see if they might benefit from germline testing.
Another aspect of precision medicine offered by Mercy is pharmacogenomics—an AI-based technology that helps predict how the patient will respond to a particular medication, so the most effective medication can be chosen.
“The genomic profile will show whether the patient is a rapid metabolizer or a poor metabolizer of a particular drug,” he said.
Mercy also offers a multi-cancer early detection test that screens for some 52 types of cancers from a blood sample. This Galleri test was developed by a company named GRAIL. Not covered by most insurance programs, the cost is $949.
“We have done more than 2,000 Galleri tests so far and people have tested positive for cancers they otherwise would not detected,” Dr. Agarwal said. The test is particularly useful for catching such cancers as pancreatic, ovarian and esophageal, which are not part of standard screenings.
Mercy was involved in the early trials for multi-cancer early detection testing and continues to participate in national trials looking at their use. Washington University is participating in a new clinical trials network launched by the National Cancer Institute to evaluate the effectiveness of the multicancer screening.
Educating Medical Students in AI
The newly updated Gateway Curriculum at Washington University School of Medicine focuses on giving future clinicians exposure to fundamental technologies, methods and foundations that will improve quality, safety, outcomes and value of care, Dr. Payne described.
“How do we get our future clinicians to be critical consumers of these technologies? How do we get them to look beyond AI as a black box? They learn to use judgement to make decisions about when and how to apply these tools,” he added.
Washington University also offers an innovation track where students focus on evolving technologies. They attend school for an additional year and earn a master’s degree in biomedical informatics along with the M.D. degree.
Mercy is incorporating genomics and AI into their residency program, Dr. Agarwal said.
At Saint Louis University, education is focused on appropriate use of AI by students in academic and clinical settings. Johan Bester, MBChB, PhD, associate dean of preclerkship curriculum and professor of family and community medicine and health care ethics, said: “Students must know that not everything that is returned by an AI is necessarily correct, and that students must double-check information and sources. It is also important that students know they may never input protected patient information or identifiable information into an AI. Lastly, we tell students that they may not represent the work of an AI as their own work.”
What’s Next for AI in Medicine?
The best is yet to come for AI in medicine, according to Dr. Payne.
“The real win—the places where generative AI will become deeply embedded in medicine—might be areas we are not even working on right now. Everyone is focused on generative AI in the exam room so we can get the computer out of the way between the patient and the provider. The question is still very open as to where the very real win will exist,” he said.
Getting to those best uses of AI requires involving physicians, other providers and patients in the design process, he added. “If we treat AI like a widget and don’t involve providers in designing and evaluating technology, it is likely to fail.”
Will AI replace physicians? To the contrary, each of the experts interviewed for this article sees AI as becoming a commonly accepted part of medical practice.
“Clinicians who do not embrace these technologies and understand how to harness them will be left behind,” Dr. Cappellari said.
Dr. Maddox expressed: “The real power of AI is that it can be very, very assistive to physicians, nurses and other clinicians. We have already demonstrated this with our projects to generate exam notes and identify patients at-risk for readmission.”
Added Dr. Payne: “AI will move from front and center to something deeply embedded in all aspects of medicine. It can free providers from high-friction tasks such as documentation so we can get back to providing care as human beings.”
Dr. Agarwal is excited about how AI and genomics can augment the physician’s work. “I will be able to leverage these technologies to do things way more efficiently and accurately. Most importantly, it will enable me to focus on the basic tenet that I love most about medicine, which is interacting with the patient. I can understand who they are, what their ailments are, what they feel and what they want out of life. That aspect of medicine will never be replaced. But we will be augmented by technologies that will help me diagnose more precisely.”
——————————————
Study Shows AI May Improve Breast Cancer Screening
Using artificial intelligence (AI) to supplement radiologists’ evaluations of mammograms may improve breast-cancer screening by reducing false positives without missing cases of cancer, according to a study by researchers at Washington University School of Medicine and Whiterabbit.ai, a Silicon Valley-based technology startup. The study was published April 10 in the journal Radiology: Artificial Intelligence.
The researchers developed an algorithm that identified normal mammograms with very high sensitivity. They then ran a simulation on patient data to see what would have happened if all of the very low-risk mammograms had been taken off radiologists’ plates, freeing the doctors to concentrate on the more questionable scans. The simulation revealed that fewer people would have been called back for additional testing but that the same number of cancer cases would have been detected.
“This simulation study showed that very low-risk mammograms can be reliably identified by AI to reduce false positives and improve workflows,” said senior author Richard L. Wahl, MD, professor of radiology.
AI-Driven Texting Platform Prevents Chemo-Related Hospitalizations
Mercy is using an AI-driven texting platform to predict and flag chemotherapy patients who may be at risk for hospitalization. It is named the Chen Chemotherapy Model for lead data scientist Jiajing Chen, who developed the model but lost her own battle with cancer in 2023.
The Chen Chemotherapy Model creates a risk score for non-leukemia patients. The model predicts the likelihood of outpatient chemotherapy patients experiencing symptoms that may result in hospitalization within 30 days of their chemotherapy treatments. Patients receive a daily text with a list of symptoms.
