Predicting the risk of pancreatic cancer in women with new-onset diabetes mellitus
Publication Details
Ali, S.,
Coory, M.,
Donovan, P.,
Na, R.,
Pandeya, N.,
Pearson, S.,
Spilsbury, K.,
Tuesley, K.,
Jordan, S.,
&
Neale, R.
(2024).
Predicting the risk of pancreatic cancer in women with new-onset diabetes mellitus.
Journal of Gastroenterology and Hepatology, Early View (Online First).
Abstract
Background and Aim
People with new-onset diabetes mellitus (diabetes) could be a possible target population for pancreatic cancer surveillance. However, distinguishing diabetes caused by pancreatic cancer from type 2 diabetes remains challenging. We aimed to develop and validate a model to predict pancreatic cancer among women with new-onset diabetes.
Methods
We conducted a retrospective cohort study among Australian women newly diagnosed with diabetes, using first prescription of anti-diabetic medications, sourced from administrative data, as a surrogate for the diagnosis of diabetes. The outcome was a diagnosis of pancreatic cancer within 3 years of diabetes diagnosis. We used prescription medications, severity of diabetes (i.e., change/addition of medication within 2 months after first medication), and age at diabetes diagnosis as potential predictors of pancreatic cancer.
Results
Among 99 687 women aged ≥ 50 years with new-onset diabetes, 602 (0.6%) were diagnosed with pancreatic cancer within 3 years. The area under the receiver operating curve for the risk prediction model was 0.73. Age and diabetes severity were the two most influential predictors followed by beta-blockers, acid disorder drugs, and lipid-modifying agents. Using a risk threshold of 50%, sensitivity and specificity were 69% and the positive predictive value (PPV) was 1.3%.
Conclusions
Our model doubled the PPV of pancreatic cancer in women with new-onset diabetes from 0.6% to 1.3%. Age and rapid progression of diabetes were important risk factors, and pancreatic cancer occurred more commonly in women without typical risk factors for type 2 diabetes. This model could prove valuable as an initial screening tool, especially as new biomarkers emerge.
Introduction
Pancreatic cancer is the seventh leading cause of cancer-related fatalities globally.1 In more developed regions, it is predicted to advance from the fourth to the second most common cause of cancer-related deaths by the year 2030.1, 2 Survival from pancreatic cancer is strongly related to stage at diagnosis, with 5-year survival for localized and metastatic disease being 44% and 3%, respectively.3, 4 Screening to detect disease earlier may improve outcomes, but due to the low lifetime risk of pancreatic cancer (1.7%),3 population-based screening is not recommended.5 Surveillance in people with a high inherited risk of pancreatic cancer can lead to earlier stage at diagnosis,6, 7 but approximately 90% of pancreatic cancers arise sporadically,8, 9 and there is currently no established surveillance strategy in this context. It has been estimated that screening may be appropriate in population groups with a lifetime risk of 5% or greater,10 so there is considerable interest in identifying these high-risk groups.
In Australia, pancreatic cancer ranked as the third most common cause of cancer-related mortality,11 and it occurs slightly more frequently in men compared with women.11 The burden of diabetes mellitus (diabetes) has been increasing, with the number of individuals in Australia affected by diabetes surging by 2.8 times in the last two decades.12 The incidence of diabetes rises with age and is slightly higher in men than women.12 These patterns underscore the significance of gaining a deeper understanding of the connection between these two conditions. Diabetes and pancreatic cancer appears to have a bidirectional association,13 with long-standing diabetes being a risk factor for pancreatic cancer and new-onset diabetes being a manifestation of pancreatic cancer.14-16 Pancreatic cancer-associated diabetes is thought to be a paraneoplastic phenomenon, where one or more factors induced by the malignancy interfere with β-cell function, leading to diabetes.15, 17 People with new-onset diabetes may be a possible target population for pancreatic cancer surveillance. However, there is currently no clear way of distinguishing diabetes caused by pancreatic cancer from typical type 2 diabetes.
A small number of studies have found that the absolute risk of developing pancreatic cancer within 3 years of being diagnosed with diabetes ranged from 0.25% to 1%.18-21 Although this is considerably higher than in people without new-onset diabetes, this absolute risk may still be too low for routine surveillance. Using risk-prediction models to stratify this population according to the risk of pancreatic cancer may help identify a subgroup of people among whom investigations of the pancreas may be appropriate. Several pancreatic cancer risk prediction models have been reported.19, 20, 22-25 However, they were limited in that they used a small number of pancreatic cancer cases or included prediction variables that are challenging to capture in real-world settings or may be indicative of advanced disease. An alternative to this is to use medication data which can serve as proxy for other underlying risk factors or that may directly influence risk of pancreatic cancer and is easily available at the point of care. Thus, it may help to identify people who should undergo surveillance for pancreatic cancer.
The aim of this study was to estimate the absolute risk of developing pancreatic cancer, and to develop and validate a model to predict the risk of pancreatic cancer within 3 years after a new diagnosis of diabetes, within a national cohort of Australian women.
Link to Publisher Version (URL)
