Apr 11 2022Reviewed by Alex Smith
An artificial intelligence (AI) model programmed using sequential health data extracted from electronic health records detected a subset of persons with a 25-fold danger of developing pancreatic cancer between 3 and 36 months, according to findings showcased at the 2022 Annual Meeting of the American Association for Cancer Research (AACR) that was held from April 8th to 13th.
Image Credit: Shutterstock.com/Magic mine
At the moment, there are no reliable biomarkers or screening tools that can detect pancreatic cancer early. The purpose of this study was to develop an artificial intelligence tool that can help clinicians identify people at high risk for pancreatic cancer so they can be enrolled in prevention or surveillance programs and hopefully benefit from early treatment.
Bo Yuan, PhD Candidate, Harvard University
Bo Yuan presented the study.
Pancreatic cancer is an aggressive form of cancer that is frequently diagnosed at later stages because of the absence of early symptoms and thus has a comparatively poor prognosis, said Davide Placido, a Ph.D. candidate at the University of Copenhagen and the study’s co-first author.
Identifying pancreatic cancer sooner in the disease course may enhance treatment options for these patients, he observed.
Latest AI advances have enthused scientists to create risk prediction algorithms for different types of cancer using pathology slides, radiology images and electronic health records.
Models endeavoring to use precancer medical diagnoses—like pancreatitis, gastric ulcers and diabetes—as signs of pancreatic cancer risk have had little success, but Yuan and colleagues aimed to create more precise models by integrating concepts from language processing algorithms.
We were inspired by the similarity between disease trajectories and the sequence of words in natural language. Previously used models did not make use of the sequence of disease diagnoses in an individual’s medical records. If you consider each diagnosis a word, then previous models treated the diagnoses like a bag of words rather than a sequence of words that forms a complete sentence.
Bo Yuan, PhD Candidate, Harvard University
The scientists programmed their AI technique using electronic health records from the Danish National Patient Registry, which comprised records from 6.1 million patients treated from 1977 to 2018, about 24,000 of whom got pancreatic cancer.
The scientists entered the order of medical diagnoses from each patient to train the model in which diagnosis patterns were considerably prognostic of pancreatic cancer risk.
The team then verified the capacity of the AI tool to estimate the incidence of pancreatic cancer within intervals ranging between 3 and 60 months following a risk assessment.
At a threshold fixed to reduce false positives, individuals said to be “at high risk” were 25 times more probable to get pancreatic cancer from 3 to 36 months than patients under the risk threshold.
In comparison, a model that did not consider the order of precancer disease events resulted in a considerably lower increased risk for patients over a corresponding threshold.
The scientists additionally confirmed their findings using electronic medical records from the Mass General Brigham Health Care System.
The variances in health care and recordkeeping practices between various health care systems necessitated the model to be re-programmed on the new dataset, Yuan explained, and upon re-programming, the model executed with comparable precision; the area under the curve (a measurement of accuracy that rises as the value approaches) for this dataset was 0.88 as compared with 0.87 for the initial training set.
Although most of the decision-making by AI occurred in the “hidden layers” of an intricate neural network, making it hard for the scientists to isolate precisely what diagnosis patterns forecast risk, Yuan and colleagues discovered major associations with specific clinical features and pancreatic cancer progress.
For example, diagnoses of diabetes, gastric ulcers, pancreatic and biliary tract diseases, and others were linked with an increased danger of pancreatic cancer. While this insight may enhance traditional risk stratification in certain cases, the benefit of the AI tool is that it assimilates data regarding risk factors in the background of a patient’s disease history, Placido explained.
The AI system relies on these features in context, not in isolation.
Bo Yuan, PhD Candidate, Harvard University
The researchers—including co-first author Jessica Hjaltelin, Ph.D.; co-senior authors Søren Brunak, Ph.D., and Chris Sander, Ph.D.; and collaborators Peter Kraft, Ph.D., Michael Rosenthal, MD, Ph.D., and Brian Wolpin, MD, MPH—anticipate this study, once assessed in clinical trials, will pave the way toward detecting patients with an elevated risk of pancreatic cancer.
This could possibly help enroll high-risk patients into programs focused on prevention and better screening for early detection. If the cancer is detected early, Placido said, the probabilities of effective treatment are higher.
These results indicate the potential of advanced computational technologies, such as AI and deep learning, to make increasingly accurate predictions based on each person’s health and disease history.
Bo Yuan, PhD Candidate, Harvard University
Limits of this research include problems standardizing electronic health data between various health systems, particularly in various countries, requiring the autonomous training and application of the AI model to various data sets. Supplementary analyses are also necessary to clearly reckon for ethnic diversity.
Furthermore, prediction accuracy drops with longer time gaps between risk assessment and cancer manifestation.
This study received funding from the Novo Nordisk Foundation, the Hale Family Center for Pancreatic Cancer Research, the National Institutes of Health, the Pancreatic Cancer Action Network, the Wexler Family Fund, the Noble Effort Fund, Promises for Purple, the Lustgarten Foundation, the Bob Parsons Fund, and Stand Up To Cancer (SU2C; The AACR is the Scientific Partner of SU2C).
Placido and Yuan declare no conflict of interest.
Source: https://www.aacr.org