UvA PhD student: “Screening for colon cancer can be more efficient with algorithms”
It is impossible to imagine a society without artificial intelligence, from generating text and images to creating music. Algorithms are also slowly starting to play a role in medicine. PhD candidate Tim Kortlever researched the potential of algorithms to detect colon cancer.
Every year, some two million Dutch people between the ages of 55 and 75 find a purple envelope on their doormat. Inside is an invitation to the population screening for colon cancer with a stool test called FIT (fecal immunochemical test).
Based on the presence of blood in your poop, the risk of colon cancer can be estimated, and the need for possible follow-up testing determined. Although the number of people with colon cancer in the Netherlands is declining, thanks to population screening, the FIT could be even more effective in detecting colon cancer. The addition of an algorithm could play a role in this. We posed five questions to PhD student Tim Kortlever, who researched the use of algorithms in colorectal cancer screening.
Why is it important for people to participate in population-based colorectal cancer screening?
“Many patients do not experience symptoms until colon cancer is already at an advanced stage, and then they go to the doctor too late. Bowel cancer arises from tiny polyps in the lining of the colon. Most of these polyps are harmless, but sometimes, over the years, polyps can grow into advanced polyps, a preliminary stage of colon cancer, and eventually into colon cancer. Once colon cancer develops, it grows deeper into the bowel wall and can metastasize. The five-year survival rate then, unfortunately, is around 12 percent. That is the likelihood of being alive five years after the cancer diagnosis. By systematically screening people without symptoms with such a population-based test, colon cancer can be found at an early stage, giving a much higher chance of being cured.”
So what exactly is wrong with the stool test?
“Well, that might be a bit of an exaggeration. But some things can be done more efficiently and better. To do that, it is important to know exactly what the stool test, the FIT, does. The test measures the hemoglobin level in your stool. Hemoglobin is a protein that is part of your red blood cells. When your intestinal wall is damaged by cancer, for example, blood appears in your stool. That blood is not always visible to the naked eye, which is why we measure hemoglobin levels. If the level of hemoglobin in the stool exceeds a certain amount, the FIT’s alarm bells go off and you are referred for an exploratory colon examination. During a colonoscopy, colon cancer or an advanced polyp is found in about one-third of people.”
“However, we know from previous research that the likelihood of finding these abnormalities during the colonoscopy varies greatly from person to person. For an equal amount of hemoglobin in the stool, a 75-year-old man who smokes has a higher risk of relevant abnormalities than a 50-year-old woman who does not smoke. This is not currently taken into account in population studies. Under the guidance of my PhD supervisors Evelien Dekker and Patrick Bossuyt, I investigated whether adding other risk factors to the stool test using an algorithm could improve population screening.”
What would such a screening using artificial intelligence look like?
“It’s good to point out that this is not yet about artificial intelligence but about the mathematical models underlying it, namely algorithms. So the self-learning ability of computers, machine learning, is not involved here yet.”
“The idea is to combine the stool test with those other risk factors. Besides the hemoglobin level in the stool, we use age, gender, smoking, and whether there are family members with colon cancer. Based on those factors, the algorithm calculates a new probability for each individual that colorectal cancer, or a precursor to it, will be found during the screening test. Only people with a high risk are then referred. As a result, instead of “one-size-fits-all,” the test becomes a more personalized test, which in theory would allow us to detect more people with colorectal cancer in a timely manner, and fewer people would need to be unnecessarily referred for an exploratory examination.”
What else does such personalized screening provide?
“First of all, a colonoscopy costs the healthcare system a lot of time and money. By screening more accurately, we can reduce the number of people who undergo an exploratory examination unnecessarily. That creates some efficiency. In addition, people have to make time for such an examination and come to the hospital. Often this also causes tension because people are afraid that colon cancer will be found. Finally, there is also a very small chance (1 in 250) that a complication will occur during the examination, such as bleeding.”
“So by having even better screening, you not only relieve the healthcare system but also a group of people who do not have colorectal cancer and who for some other reason have elevated hemoglobin levels in their stool.”
It sounds very promising. Did the algorithm work in your study?
“Anticlimax: My results were disappointing, ha-ha. On paper, algorithms have a lot of potential, but testing them in practice still proved very difficult. One reason was the circumstances under which the study was conducted. After all, the population screening for colon cancer started in the Netherlands in 2014. I couldn’t start my research until 2020 because the study design had to be checked by the Health Council first. That is a long procedure with very strict requirements. Most people between the ages of 55 and 75 had long since been screened by then. As a result, there was a relatively small group of people available for the study, which was also not very diverse. For that reason, I could not properly compare the effectiveness of my algorithm with that of the current stool test.”
“Still, I think algorithms in combination with a stool test are the way forward. For example, my fellow researcher Willemijn de Klaver, together with colleagues from Rotterdam, demonstrated this in a recent publication in the Lancet Oncology. In her research, she combined hemoglobin with several proteins in the stool that have also been linked to colon cancer. This combination led to better detection of advanced polyps compared with the current stool test, without referring more people for an exploratory examination. It still has a long way to go before it is implemented, but her research shows that algorithms can certainly play a role in colon cancer detection in the future.”
