AI-based population screening may facilitate diagnosis of pancreatic cancer

A new study by researchers from Harvard Medical School and the University of Copenhagen in collaboration with the VA shows that an artificial intelligence tool can predict pancreatic cancer up to three years before diagnosis using only a patient’s medical record. We have successfully identified people at highest risk. Boston Health Care System, Dana-Farber Cancer Institute, Harvard TH Chan School of Public Health.

Survey results announced on May 8 natural medicine, AI-based population screening could help identify people at high risk for the disease, and could hasten the diagnosis of conditions often discovered in advanced stages where treatment is less effective and outcomes disastrous. It suggests. Pancreatic cancer is one of the deadliest cancers world, and its tolls are projected to increase.

Currently, there are no population-based tools for broad screening for pancreatic cancer. People with a family history of certain genetic mutations that predispose to pancreatic cancer are screened in a targeted manner. However, such targeted screening may miss other cases that do not fall into these categories, researchers say.

One of the most important decisions clinicians face every day is who is at higher risk for disease and who might benefit from further testing. This can mean that more invasive and expensive procedures come with their own risks. AI tools that can target those at highest risk of pancreatic cancer and who stand to benefit most from further testing could greatly help improve clinical decision-making. “

Chris Sander, Research Co-Principal Investigator, Department of Systems Biology, Bravatnik Institute, HMS

Applied on a large scale, such an approach could enhance pancreatic cancer detection, lead to earlier treatment, improve outcomes and extend patient life, Sander added. These types of cancer, especially those that are difficult to detect and treat early, wreak disproportionate toll on patients, families and the entire healthcare system,” said co-principal investigator of the study, Disease Systems Biology. Professor and Research Director Søren Brunak said. At the Novo Nordisk Foundation Center for Protein Research at the University of Copenhagen. “AI-based screening represents an opportunity to change the trajectory of pancreatic cancer, a notoriously difficult, progressive disease that is difficult to diagnose early and treat promptly when it has the highest chance of success.”

In the new study, the AI ​​algorithm was trained on two separate datasets of a total of 9 million patient records from Denmark and the United States. The researchers “asked” the AI ​​model to look for telltale signs based on the data contained in the records. Based on the combination of disease code and its time of onset, the model was able to predict which patients were more likely to develop pancreatic cancer in the future. Notably, many of the symptoms and disease codes were not directly related to or derived from the pancreas.

Researchers tested different versions of the AI ​​model for its ability to detect people at high risk of developing the disease on different time scales. 6 months, 1 year, 2 years, 3 years. Overall, each version of the AI ​​algorithm was significantly more accurate in predicting who would develop pancreatic cancer than current population-wide estimates of disease incidence. It is defined as the frequency with which a condition occurs in a population over a specified period of time. The researchers believe the model can predict disease development at least as accurately as current gene-sequencing tests, which are usually available only for a small subset of patients in the dataset.

“Anger Organ”

Screening for certain common cancers such as breast, cervical, and prostate relies on relatively simple and highly effective techniques. mammograms, Pap smears, and blood tests, respectively. These screening methods have changed the outcome of these diseases by ensuring early detection and intervention at the most treatable stages.

By comparison, pancreatic cancer is difficult and expensive to screen and test. Physicians focus primarily on family history and the presence of genetic mutations, which, while important indicators of future risk, are often overlooked in many patients. One particular advantage of AI tools is that they can be used on all patients for whom health records and medical histories are available, not just those with a known family history or genetic predisposition to the disease. This is especially important, the researchers add, as many at-risk patients may not even be aware of their genetic predisposition or family history.

In the absence of symptoms and clear indications of an increased risk of pancreatic cancer, it is not surprising that clinicians recommend more advanced and expensive tests, such as CT scans, MRIs, and endoscopic ultrasound. If these tests are used and suspicious lesions are found, the patient must undergo the procedure to have a biopsy. Organs located deep within the abdomen are difficult to access and prone to irritation and inflammation. An AI tool that identifies people at highest risk for pancreatic cancer would ensure clinicians are testing the right population while avoiding other unnecessary tests and extra steps, study says said the person.

Approximately 44% of people diagnosed in the early stages of pancreatic cancer survive 5 years after diagnosis, but only 12% of cases are diagnosed at that early stage. Researchers estimate that survival rates drop to 2-9% for patients whose tumors have grown beyond their site of origin.

“This low survival rate is despite significant advances in surgical techniques, chemotherapy and immunotherapy,” Sander said. “Therefore, in addition to advanced treatments, there is a clear need for better screening, more targeted testing and earlier diagnosis, with AI-based approaches emerging as the first critical step in this continuum. ”

Previous diagnoses portend future risks

For the current study, researchers designed several versions of AI models and trained them on the health records of 6.2 million patients from the Danish National Health System over 41 years. Of these patients, 23,985 developed pancreatic cancer over time. During training, the algorithm identified patterns indicative of future pancreatic cancer risk based on disease trajectory. That is, whether the patient went through a particular condition and occurred in a particular sequence.

For example, diagnoses such as gallstones, anemia, type 2 diabetes, and other gastrointestinal-related problems were predictors of increased risk of pancreatic cancer within 3 years of evaluation. Not surprisingly, pancreatic inflammation was strongly predictive of future pancreatic cancer within an even shorter period of 2 years. The researchers caution that none of these diagnoses should be considered signs or causes of future pancreatic cancer. It may provide clues to the model and encourage physicians to monitor more closely those at risk or test accordingly.

The researchers then tested the best-performing algorithms on an entirely new set of patient records never encountered before. The U.S. Veterans Health Administration dataset has approximately 3 million records over 21 years and includes 3,864 of her individuals diagnosed with pancreatic cancer. The predictive accuracy of this tool was slightly lower for the US data set. This is most likely because the US dataset was collected in a shorter time and contained somewhat different patient population profiles. The total Danish population in the Danish data set and current and former military personnel in the Veterans Affairs data set. Retraining the algorithm from scratch on the US dataset improved the prediction accuracy. According to researchers, this highlights her two key points. The first is ensuring that AI models are trained on high-quality, rich data. Second, the need for access to large, representative datasets of nationally and internationally aggregated clinical records. In the absence of such globally valid models, AI models must be trained on regional health data to ensure that the training reflects the idiosyncrasies of local populations.