Early diagnosis of pancreatic cancer using machine learning applied to primary care data

What is it about?

Our pilot study showed that such a high-risk sub-population was identifiable within the general population up to 20 months prior to pancreatic cancer diagnosis with a sensitivity approaching 70%. However, it has been conducted on a relatively small sample of pancreatic cancer cases and used other-cancer patient as controls.

Featured Image

Photo by Mika Baumeister on Unsplash

Photo by Mika Baumeister on Unsplash

Why is it important?

Pancreatic cancer (PC) is rarely detected in its early stages, and therefore is one of the tumours with the lowest survival. Progress to diagnose it early, when the tumour is localised to pancreas and still treatable, has been extremely slow. Although several years may elapse between the biological initiation of PC and its clinical presentation (usually at an advanced stage), many patients display no specific alarm symptoms for most of this time. Introduction of a mass-population based screening for PC is prohibitively expensive and even potentially damaging due to the low disease incidence. Targeted screening is an alternative option. Using protein markers either alone or in combination with one clinically established biomarker, CA 19-9, have revealed promising results. The authors of these molecular studies mostly identify their application to high-risk groups containing small numbers of people. Alternatively, such tests might be applied more widely if a suitable high-risk subgroup of the general population could be identified.

Perspectives