New risk tool could identify people with new-onset diabetes at highest risk for pancreatic cancer

Pancreatic cancer is associated with poor outcomes and less than a quarter of people diagnosed survive past a year. Being able to diagnose pancreatic cancer earlier could improve outcomes, but this is challenging to do as symptoms can be non-specific and there are currently no simple diagnostic tools or screening programmes to help identify the disease.  

Early diagnosis could be improved by identifying people at increased risk of developing the disease  so that they can be monitored and receive effective treatment when needed.   

One such group is adults with newly diagnosed type 2 diabetes, of which, up to 1% get diagnosed with pancreatic cancer within 3 years. It’s not completely clear whether this association is due to diabetes being a risk factor for pancreatic cancer, or a tumour causing the diabetes (so-called type 3 diabetes), or both.   

In the UK, the National Institute for Health and Care Excellence (NICE)  recommends adults aged 60 years and older with newly diagnosed diabetes and weight loss  are sent for urgent imaging to look for a possible tumour in the pancreas. However, by considering multiple other factors about the individual, people with newly diagnosed diabetes could benefit from more nuanced risk prediction that goes beyond just using age and a single (albeit important) symptom.   

 In the Primary Care Epidemiology group, we explored this idea using the QResearch database to develop and evaluate a new risk prediction model. QResearch collects anonymised health record data from millions of individuals in the UK in primary care, and is linked to data from the cancer registry, hospital care data and also the national death register.  

Using data from over a quarter of a million adults aged 30-85 when they were diagnosed with type 2 diabetes, we developed a new risk prediction model that considers age, sex, BMI, blood test results, and whether or not the individual had specific symptoms in the preceding 6 months (pain in the abdomen, weight loss, jaundice, heartburn, indigestion or nausea). By including these factors, the new model could offer more nuanced risk estimates than current guidance and aid the referral of people with newly diagnosed diabetes for further testing. 

The model had promising results on evaluation. For example, it had a better balance of true positives and false negatives than would be attained if everyone with type 2 diabetes was sent for an urgent scan, and the model was more sensitive at picking up cases of pancreatic cancer than current NICE criteria.  

Whilst this new model has encouraging results, it needs to be validated in an independent data source before it is used on real individuals. Our study provided an understanding of how well the new model performs statistically, but further research is needed to explore the best way to use it in practice. For example, a simulation study could be conducted to identify the most effective strategies in which the model could be used, considering thresholds, numbers of scans and number of cancers detected. 

To help improve patient survival rates for pancreatic cancer, it’s vital that we develop methods to improve early detection. Identifying factors that could increase an individual’s risk of developing pancreatic cancer is a key step in this process. This new model, once validated, could help us identify those patients who are at higher risk so they can be monitored and receive preventative measures and effective treatment as early as possible.   

Read the full paper, ‘Predicting the risk of pancreatic cancer in adults with new-onset diabetes: development and internal–external validation of a clinical risk prediction model’ in the British Journal of Cancer.