Gastrointestinal surgeon and scientist at Amsterdam UMC Freek Daams explains a long-standing issue: “Many colon cancer patients undergo surgery whereby we remove a section of the intestine and reconnect the two ends. In about 90% of cases, this connection heals without problems; but in the remaining 10% of cases, complications arise – such as an anastomotic leak that causes infection or excrement to enter the abdominal cavity. This can be so dangerous that patients may even die from it.”
Predicting instead of treating
In recent years, Daams and his colleagues have researched how to detect and treat intestinal leaks. To determine who might develop complications and when, patients undergo extensive testing before their operation. Despite these efforts, unfortunately the leakage rate has not significantly decreased in the past ten years. However, this research has resulted in a database of around 5,000 colon cancer surgery patients in the Netherlands – data that could help predict what will lead to complications.
And prediction is precisely where AI tools excel. Daams explains: “We built a machine-learning model that can predict the likelihood of an anastomotic leak during surgery. This model proved to make predictions better than surgeons can, because it compares dozens of parameters from the database with an individual patient. It turned out to be particularly good at predicting when a complication would definitely not occur. Our data scientist even joked, ‘if this were a profit-prediction algorithm, we’d be making a lot of money with it on the Zuidas!’”
Making better decisions
So what does this mean in practice? “Over the past two years, we’ve run the AI model behind the scenes during operations in 18 hospitals. We then held that prediction close to our chest, in order to check whether the model performs well in real-world settings. We’re now awaiting those results. Sometimes I lie awake at night wondering: what if the model turns out to be completely useless?” says Daams. “But if the results are positive, the next step will be to really start using the knowledge gained via the AI model during operations.”
During surgery, there are ways to prevent leakage, such as by placing a stoma – temporarily or permanently. Daams explains: “That has a significant impact on a patient’s quality of life, and we can only intervene when we know a complication is likely. If a healthy patient has a high risk of leakage, you might still create a suture because they can recover well from an eventual leakage. But for an 80-year-old woman, you might decide to place a stoma instead. You make a decision based on all the knowledge and information that you have available. And we hope that this model can help us to make the best possible decision.”
“The use of AI also has consequences for the process that follows such an operation,” Daams continues. “Right now, we thoroughly examine all patients and conduct numerous tests, even though a leak only occurs in 8–10% of patients. With AI, we could reduce unnecessary screenings and even discharge low-risk patients sooner.”
The future of AI
Daams concludes: “I truly see a future for predictive models. “Not just for predicting intestinal leaks, but also for other common conditions like delirium (severe confusion) or heart attacks. If we can identify high- and low-risk patients early on, it will have a huge impact on the capacity and quality of healthcare.”
While AI offers hope for the future, it also raises ethical questions. How do surgeons and patients feel about it? “Ultimately, the surgeon bears the responsibility,” says Daams. “As a surgeon, you use all available knowledge – including AI – but you make the final decision yourself. These are also matters to discuss with the patient in advance: ‘What do you want me to do with the results?’ So far, reactions from both sides have been positive, but we continue to discuss these issues with all those involved.”