Faced with a 2021 U.S. Preventive Services Task Force recommendation to begin colorectal cancer screenings at 45, Sanford Health’s clinical teams encountered a conundrum: How to manage screenings for 100,000 newly eligible patients with a limited supply of gastroenterologists in the rural Dakotas.
The solution: A new tool that could be embedded in the EHR aided by augmented intelligence (AI)—also called artificial intelligence—and designed to help physicians develop preventive-care plans for patients who may be at risk of developing colon cancer. Sanford Health’s algorithm identifies nontraditional risk factors for colon cancer. In an initial study, this approach yielded a fivefold increase in detection rates over the traditional risk factors considered and screening questions employed. The next step is for this tool to be rolled out in primary care offices across Sanford.
For rural population, the AI model is a game changer, said Jeremy Cauwels, MD, an internist and chief medical officer at South Dakota-based Sanford Health, which is a member of the AMA Health System Program that provides enterprise solutions to equip leadership, physicians and care teams with resources to help drive the future of medicine.
Since the pilot’s start, Sanford Health—the largest rural health system in the country—has also revamped its scheduling and testing procedures for patients who might otherwise have faced delays due to backlogs. This has led to better adherence to screenings and earlier detection. For average-risk patients, the health system has increased its offerings of Food and Drug Administration (FDA)-approved at-home tests that detect abnormal blood and DNA in stool samples.
If a patient drives 200 miles or more for their colonoscopy, “we want to make sure we’re prioritizing the right person so that when we bring them in … we have a high likelihood of diagnosing a cancer early and potentially saving years of their life,” he said.
In an interview with the AMA, Dr. Cauwels discussed the origins of the AI pilot at Sanford Health, how the health system plans to roll this out to physicians and his vision for using AI to detect a variety of diseases earlier to improve outcomes.
AMA: What are challenges faced by rural communities in getting screened for colorectal cancer?
Dr. Cauwels: We already know that colorectal cancer is the second leading cause of death in the United States. North Dakota, one of our states that we serve, has one of the highest rates of colon cancer in the country. A recent Centers for Disease Control and Prevention report shows that rural Americans are more likely to die of preventable causes of death like cancer.
We’re trying to push forward to say: No matter where you live, it's extremely important for us to be able to bring the full suite of medical knowledge and medical ability to you, whether you live in a city of 200,000 people or a town of 200.
AMA: How is Sanford Health using AI to help address those challenges in screening for colorectal cancer?
Dr. Cauwels: The most promising is an algorithm that we're working to embed in our electronic medical record. In 2023, a group of GI physicians approached our enterprise data analytics team wanting to better understand factors driving an increase in colon cancer diagnoses.
Our team developed a comprehensive AI-based model which identifies genetic, environmental and lifestyle factors that could put patients at an increased risk for colon cancer. It’s designed to support our primary care doctors to try to do a better job of predicting who may be more likely to be diagnosed with colorectal cancer so they can be more informed when talking to the patient about the importance of colorectal screenings and how to understand who might be at higher risk than traditional risk models.
Our data analytics team put this together and did the regression analysis a couple of years ago. They discovered 85 different independent variables that can increase your risk of colon cancer. These variables aren't standard things like family history of colon cancer or other things that require medical tests to determine they are truly variables you can find in anyone's electronic record—age, smoking history, all those little tidbits.
If I wanted to sit down and put them all together, I could. Or I could let a machine do the mathematics for me in the background and help me come to the same conclusion much faster and much more easily. This allows us to spend more time caring for our patients, which is what we all want to do.
The standard of care in physician well-being recognition
Institutions in this article have been honored by the Joy in Medicine™ Health System Recognition Program.
AMA: How does the AI tool work in assisting physicians behind the scenes?
Dr. Cauwels: We were able to train an AI model on the entire library of patients at Sanford Health, all the way back to when we went live with an electronic medical record 20 years ago. Next was a forward-looking scientific study where we took patients coming into our hospital to get screened, and asked: What would our risk factor calculator show us before they got screened? Could we prospectively say these people would be more likely to be find cancer or other abnormalities? And sure enough, the model proved itself.
The next step is publishing the study and making sure that the medical associations and the American College of Gastroenterology would be willing to roll it out in primary care offices.
AMA: What metrics or outcomes are you tracking to assess the effectiveness of the AI pilot?
Dr. Cauwels: Prospectively, we found that nontraditional risk factors were about five times better than traditional risk factors in detecting risk of colon cancer. In an initial six-month study, which included a dataset of more than 450,000 individuals 45–80 years old, traditional risk factors led to a 2.4% colon cancer detection rate. Applying additional risk factors to the AI model yielded a five-fold increase in the colon cancer detection rate (12%), potentially predicting 2,604 future cases of colon cancer.
Then, 91% of the people we identified to be at higher risk were not identified by traditional risk factors. Only 9% of those folks in our model have the traditional risk factors that would put them at higher risk. Traditional models utilize a median of five risk factors for colon cancer, whereas the AI-based model developed by Sanford Health looks at 85 variables including things such as neurological disease and psychiatric conditions.
AMA: How is the AI model affecting scheduling and testing procedures, particularly for patients who might otherwise face delays in getting appointments?
Dr. Cauwels: We used this AI model in an early pilot in one of our hospitals in North Dakota. We had a backlog of over a year from when a doctor would put in the order for a colonoscopy and when a patient would finally get one. We used this to help us reconsider that queue and prioritize the patients that this evidence-based AI model showed were the likeliest to benefit from earlier screening.
We also sent DNA-based stool testing to the people in the line and said: If you have a positive stool test, we'll move you up in the line faster so you can get taken care of as well. With that stool-based testing we managed to move up literally hundreds of patients sooner in the queue.
AMA: Has the AI model improved collaboration between clinical care teams?
Dr. Cauwels: It has. We have some GI doctors who are much more likely to use AI and to understand it. In fact, that same group is also using a product called GI Genius, which is an FDA-cleared, AI-assisted colonoscopy.
Early detection is essential to preventing colorectal cancer, but certain polyps can be difficult to see with the human eye due to location and other factors. The module has 13 million images it uses to compare to the patient’s colon in real-time, and AI software highlights polyps to assist the doctor performing the colonoscopy.
In a randomized trial, the GI Genius technology increased the detection of pre-cancerous polyps by 13%. Each 1% increase in the detection rate decreases patients’ risk of colorectal cancer by 3%.
AMA: What are the next steps for the AI pilot in colorectal cancer screening and your study?
Dr. Cauwels: The next step is to roll it out to our entire primary care footprint and teach them that all of this can help them in every single visit they have during a day.
AMA: How do you see AI continuing to transform preventive care and early detection beyond colorectal cancer?
Dr. Cauwels: We can identify lots of illnesses and diseases where, if we had a clearer picture of patient risk, we could intervene at a different time. We'll be able to predict diabetes before it happens and help move things along the right way.
I'm very interested in things such as breast cancer and cervical cancer—are there pieces in the electronic medical record that we're just not putting together fast enough for our 15-minute visits? How can the institution of medicine help our doctors not spend so much time digging through the chart, but rather bring them the information and eliminate that burnout because they don't have to spend time dragging through page after page of electronic data. Rather, the data is brought to them in a way that's more useful to them when the patient's sitting in front of them.
From AI implementation to EHR adoption and usability, the AMA is fighting to make technology work for physicians, ensuring that it is an asset to doctors—not a burden.
