AI-based population screening could speed pancreatic cancer diagnosis

An artificial intelligence tool has successfully identified people most at risk for pancreatic cancer up to three years before diagnosis using only patient medical records, according to new research by researchers from Harvard Medical School and the University of Copenhagen, in collaboration with VA Boston Healthcare System, Dana-Farber Cancer Institute and Harvard TH Chan School of Public Health.

The conclusions, published on 8 May in natural medicine, suggest that AI-based population screening could be useful in finding people at high risk for the disease and could speed up diagnosis of a disease found too often in later stages when treatment is less effective and the results are dismal, the researchers said. Pancreatic cancer is one of the deadliest cancers in the world, and its balance sheet is expected to increase.

Currently, there are no population-based tools to broadly screen for pancreatic cancer. Those with a family history and certain genetic mutations that predispose them to pancreatic cancer are screened in a targeted manner. But such targeted screenings may miss other cases that don’t fall into those categories, the researchers said.

One of the most important decisions clinicians face on a daily basis is who is at high risk for disease and who would benefit from further testing, which can also mean more invasive and expensive procedures that involve their own risks. An AI tool that can focus on those most at risk for pancreatic cancer who will benefit the most from additional testing could go a long way to improving clinical decision-making. »

Study Co-Principal Investigator Chris Sander, Faculty Member of the Department of Systems Biology at HMS Blavatnik Institute

Applied on a large scale, Sander added, such an approach could speed up the detection of pancreatic cancer, lead to earlier treatment, improve outcomes and extend the life of patients. “Many types of cancer, especially those that are difficult to identify and treat early, exert a disproportionate impact on patients, families, and the healthcare system as a whole,” said the study’s co-lead investigator, Søren Brunak, Professor of Disease Systems Biology and Director of Research. at the Novo Nordisk Foundation Center for Protein Research at the University of Copenhagen. “AI-based screening is an opportunity to alter the trajectory of pancreatic cancer, an aggressive disease notoriously difficult to diagnose early and treat quickly when the chances of success are highest.”

In the new study, the AI ​​algorithm was trained on two separate datasets totaling 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 in the recordings. Based on combinations of disease codes and when they occurred, the model was able to predict which patients are likely to develop pancreatic cancer in the future. Notably, many symptoms and disease codes were not directly related to, or originated from, the pancreas.

The researchers tested different versions of the AI ​​models for their ability to detect people at high risk of developing the disease on different time scales -; 6 months, one year, two years and three years. Overall, each version of the AI ​​algorithm was significantly more accurate in predicting who would develop pancreatic cancer than current population-wide disease incidence estimates; defined as the frequency with which a condition develops in a population over a specific period of time. The researchers said they believe the model was at least as accurate in predicting the onset of the disease as current genetic sequencing tests which are generally only available for a small subset of patients in data sets. .

“The Angry Organ”

Screening for some common cancers such as breast, cervix and prostate is based on relatively simple and very effective techniques -; a mammogram, a Pap test and a blood test, respectively. These screening methods have transformed outcomes for these diseases by providing early detection and intervention at the most treatable stages.

In comparison, pancreatic cancer is harder and more expensive to detect and test. Doctors primarily look at family history and the presence of genetic mutations, which, while important indicators of future risk, are often missed by many patients. A particular advantage of the AI ​​tool is that it could be used on all patients for whom health records and medical history are available, not just those with a known family history or genetic predisposition to the disease. disease. This is particularly important, the researchers add, because many high-risk patients may not even be aware of their genetic predisposition or family history.

In the absence of symptoms and no clear indication that a person is at high risk for pancreatic cancer, clinicians may understandably be cautious and recommend more sophisticated and expensive tests, such as CT scans, MRIs, or endoscopic ultrasound. When these tests are used and suspicious lesions are found, the patient must undergo a procedure to obtain a biopsy. Located deep inside the abdomen, the organ is difficult to access and easy to provoke and inflame. His irritability earned him the nickname “the angry organ”. An AI tool that identifies those most at risk for pancreatic cancer would ensure clinicians are testing the right population, while sparing others unnecessary testing and additional procedures, the researchers said.

About 44% of people diagnosed in the early stages of pancreatic cancer survive five years after diagnosis, but only 12% of cases are diagnosed that early. The survival rate drops to 2-9% in those whose tumors have grown beyond their original site, the researchers estimate.

“This low survival rate is despite marked advances in surgical techniques, chemotherapy and immunotherapy,” Sander said. “So, in addition to sophisticated treatments, there is a clear need for better screening, more targeted testing and earlier diagnosis, and this is where the AI-based approach emerges as the first critical step in this continuum.”

Past diagnoses predict future risk

For the current study, the researchers designed several versions of the AI ​​model and trained them on the health records of 6.2 million patients in Denmark’s national healthcare system spanning 41 years. Of these patients, 23,985 developed pancreatic cancer over time. During training, the algorithm discerned patterns indicative of future pancreatic cancer risk based on disease trajectories, i.e. whether the patient had certain conditions that occurred in a certain sequence over time.

For example, diagnoses such as gallstones, anemia, type 2 diabetes, and other gastrointestinal problems predicted an increased risk of pancreatic cancer within 3 years of evaluation. Less surprisingly, inflammation of the pancreas was strongly predictive of future pancreatic cancer within an even shorter time frame of two years. The researchers caution that none of these diagnoses by themselves should be considered indicative or causative of future pancreatic cancer. However, the pattern and sequence in which they occur over time offer clues for an AI-based surveillance model and could prompt doctors to monitor high-risk individuals more closely or perform live testing. result.

Next, the researchers tested the best-performing algorithm on an entirely new set of patient records it had not encountered before; a US Veterans Health Administration dataset of nearly 3 million records spanning 21 years and containing 3,864 people diagnosed with pancreatic cancer. The tool’s predictive accuracy was somewhat lower on the US dataset. This was likely due to the fact that the US dataset was collected over a shorter period and contained somewhat different patient population profiles –; the entire population of Denmark in the Danish dataset compared to current and former military personnel in the veterans dataset. When the algorithm was retrained from scratch on the US dataset, its predictive accuracy improved. According to the researchers, this underscores two important points: first, ensuring that AI models are trained on rich, high-quality data. Second, the need to access large representative datasets of nationally and internationally aggregated clinical records. In the absence of such globally valid models, AI models should be trained on local health data to ensure that their training reflects the idiosyncrasies of local populations.