Researching the cures


Artificial intelligence shows potential in identifying those at risk of pancreatic cancer

A pilot study funded by PCRF has shown that by analysing electronic health records with sophisticated computer modelling techniques, it is possible to identify people at risk of developing pancreatic cancer up to 20 months before their diagnosis.

The study was led by Dr Laura Woods and her colleagues Professor Bernard Rachet and Dr Ananya Malhotra at London School of Hygiene & Tropical Medicine. Its findings show that using artificial intelligence to sift through huge amounts of patient medical data can highlight patterns of early symptoms, and identify a group of patients who would benefit from further investigations.

The study analysed just under 1200 anonymised electronic health records of people who had been diagnosed with pancreatic cancer and over 4,000 anonymised electronic health records of patients who had been diagnosed with a different cancer.

The team created a model based on patient data – including symptoms, prior diseases and medications – in the two years before their diagnosis, which predicted the risk of a new patient developing pancreatic cancer.

In those under 60 years of age in the study group, the model correctly predicted over half of the patients who went on to develop pancreatic cancer, up to 20 months before they were diagnosed.

Presenting the findings at the World Congress on Gastrointestinal Cancer in Barcelona, statistician Dr Ananya Malhotra explained: “We’re talking about such a large amount of data, it would be almost impossible to identify these trends with the naked eye.”

Modelling tools are not routinely used in primary care settings, but the researchers believe that GPs could one day use this type of predictive modelling on their patients’ medical records to identify those who would benefit from further investigations.

Patients who develop pancreatic cancer often consult their GP with non-specific symptoms such as gastrointestinal problems or back pain in the months and years prior to diagnosis. Individually, these symptoms are unlikely to trigger further investigations for cancer.

Lead researcher Dr Laura Woods said: “These early results are exciting and enough to know we’re onto something here, but more needs to be done. We’ve shown that it’s possible to predict which patients may develop pancreatic cancer, even among a group of cancer patients. We will continue to refine our model and plan to apply it to data from the general population.”

She added: “We hope that ultimately it could be paired with a blood or urine biomarker test, followed by scans or biopsies for those whose tests were positive. This would enable much earlier diagnosis for a significant number of patients who can then start appropriate treatment immediately.”

“We funded this research project because we believed it to be both innovative and timely,” said Maggie Blanks, PCRF’s CEO. “Pancreatic cancer needs to be tackled on all fronts, so as we work to discover and develop new drugs and better treatments, we also need to really push to diagnose the disease earlier. While there are several promising early detection tests using blood or urine currently being validated, we need to know how to spot those who would benefit from these tests. Using artificial intelligence to do this, would be an enormous step forward.”