According to new research led by investigators at Harvard Medical School and the University of Copenhagen, in collaboration with VA Boston Healthcare System, Dana-Farber Cancer Institute, and the Harvard T.H. Chan School of Public Health, an artificial intelligence tool has successfully identified people at the highest risk for pancreatic cancer up to three years before diagnosis using the patient’s medical records solely.
The findings, published May 8 in Nature Medicine, suggest that AI-based population screening could be valuable in finding those at elevated risk for the disease and expedite the diagnosis of a condition found all too often at advanced stages when treatment is less effective. Outcomes are dismal, the researchers said. Pancreatic cancer is one of the deadliest cancers in the world, and its toll is projected to increase.
Currently, there are no population-based tools to screen broadly for pancreatic cancer.
Those with a family history and certain genetic mutations predisposing them to pancreatic cancer are screened in a targeted fashion. But such targeted screenings can only cover cases that fall outside of those categories, the researchers said.
“One of the most important decisions clinicians face day to day is who is at high risk for a disease, and who would benefit from further testing, which can also mean more invasive and more expensive procedures that carry their own risks,” said the study co-senior investigator Chris Sander, a faculty member in the Department of Systems Biology in the Blavatnik Institute at HMS. “An AI tool that can zero in on those at highest risk for pancreatic cancer who stand to benefit most from further tests could go a long way toward improving clinical decision-making.”
Applied at scale, Sander added, such an approach could expedite the detection of pancreatic cancer, lead to earlier treatment, improve outcomes and prolong patients’ life spans.
“Many types of cancer, especially those hard to identify and treat early, exert a disproportionate toll on patients, families and the healthcare system as a whole,” said study co-senior investigator Soren 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 that is notoriously hard to diagnose early and treat promptly when the chances for success are highest.”
The new study trained the AI algorithm on two separate data sets totalling 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 combinations of disease codes and the timing of their occurrence, 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 stemming from the pancreas.
The researchers tested different versions of the AI models for their ability to detect people at elevated risk for disease development within different time scales — 6 months, one year, two years, and three years. Overall, each version of the AI algorithm was substantially more accurate at predicting who would develop pancreatic cancer than current population-wide estimates of disease incidence — defined as how often a condition develops in a population over a specific period. The researchers said the model is at least as accurate in predicting disease occurrence as current genetic sequencing tests are usually available only for a small subset of patients in data sets.
The “angry organ” Screening for certain common cancers, such as those of the breast, cervix, and prostate gland, relies on relatively simple and highly effective techniques — a mammogram, a Pap smear, and a blood test, respectively. These screening methods have transformed disease outcomes by ensuring early detection and intervention during the most treatable stages.
By comparison, pancreatic cancer is harder and more expensive to screen and test for. Physicians look mainly at family history and the presence of genetic mutations, which, while important indicators of future risk, often miss many patients. One particular advantage of the AI tool is that it could be used on any and all patients for whom health records and medical history are available, not just in those with a known family history or genetic predisposition for the disease. This is especially important, the researchers add, because many patients at high risk may not even be aware of their genetic predisposition or family history.
In the absence of symptoms and without a clear indication that someone is at high risk for pancreatic cancer, clinicians may be understandably cautious about recommending more sophisticated and expensive testing, such as CT scans, MRI or endoscopic ultrasounds. When these tests are used, and suspicious lesions are discovered, the patient must undergo a procedure to obtain a biopsy. Positioned deep inside the abdomen, the organ is hard to access and easy to provoke and inflame. Its irritability has earned it the moniker “the angry organ.”
The researchers said an AI tool that identifies those at the highest risk for pancreatic cancer would ensure that clinicians test the right population while sparing others unnecessary testing and additional procedures.
About 44 per cent of people diagnosed in the early stages of pancreatic cancer survive five years after diagnosis, but only 12 per cent of cases are diagnosed that early. Researchers estimate that the survival rate drops to 2 to 9 per cent in those whose tumours have grown beyond their site of origin.
“That 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 where the AI-based approach comes in as the first critical step in this continuum.”
Previous diagnoses portend 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 from Denmark’s national health system spanning 41 years. Of those patients, 23,985 developed pancreatic cancer over time. During the training, the algorithm discerned patterns indicative of future pancreatic cancer risk based on disease trajectories, that is, whether the patient had certain conditions that occurred in a certain sequence over time.
For example, diagnoses such as gallstones, anaemia, type 2 diabetes, and other GI-related problems portended greater risk for pancreatic cancer within three years of evaluation. Less surprisingly, inflammation of the pancreas was strongly predictive of future pancreatic cancer within an even shorter period of two years. The researchers caution that none of these diagnoses should be indicative or causative of future pancreatic cancer. However, the pattern and sequence they occur over time offer clues for an AI-based surveillance model. They could prompt physicians to monitor those at elevated risk or test accordingly closely.
Next, the researchers tested the best-performing algorithm on an entirely new set of patient records it had not previously encountered — a US Veterans Health Administration data set of nearly 3 million records spanning 21 years and containing 3,864 individuals diagnosed with pancreatic cancer. The tool’s predictive accuracy was lower on the US data set. This was most likely because the US dataset was collected over a shorter time and contained somewhat different patient population profiles — the entire population of Denmark in the Danish data set versus current and former military personnel in the Veterans’ Affairs data set.
When the algorithm was retrained from scratch on the US dataset, its predictive accuracy improved. The researchers said this underscores two important points: First, ensuring that AI models are trained on high-quality and rich data. Second, the need for access to large representative datasets of clinical records aggregated nationally and internationally.
Such globally valid models are necessary for AI models to be trained on local health data to ensure their training reflects the idiosyncrasies of local populations.
(with inputs from ANI)
