Introduction

Patients with new onset diabetes (NODM) have a higher risk of pancreatic cancer (PAC) development. We aimed to (1) compare the incidence of PAC among NODM and non-DM patients and (2) develop a risk prediction model for incident PAC within 1 year after diagnosis of DM.

Aims & Methods

We included adult patients who were newly diagnosed with type 2 DM (i.e. NODM group) or had any clinical consultation without a diagnosis of DM (i.e. non-DM group) between 2008 and 2015 from a territory-wide electronic healthcare database in Asia. Patients with type 1 DM or prior history of PAC were excluded. Each patient was followed from the diagnosis date of DM (or first consultation in non-DM group) till PAC development, death or 1 year. The incidence rate of PAC was calculated under Poisson distribution and compared between NODM and non-DM groups. The NODM cohort with complete data was further randomly split in a ratio of 7:3 into a training set (to develop the prediction model) and a validation set, respectively. Logistic regression with forward stepwise selection was conducted to identify the potential predictors of PAC. Model performance against the validation set was assessed by the area under receiver operating characteristic curve (AUC).

Results

Among 4,002,826 patients identified, 261,046 (6.5%) of whom had NODM. The incidence of PAC at 1 year was 11.68 per 10,000 persons (95% CI: 10.41-13.07) for NODM group and 1.82 per 10,000 persons (95% CI: 1.68-1.96) for non-DM group, with an incidence rate ratio of 6.43 (95% CI: 5.60-7.37). To develop a prediction model for PAC at 1 year, 19,210 NODM patients (28 PAC cases) were included, of whom 13,447 (18 PAC cases) were in the training set and 5,763 (10 PAC cases) in the validation set. After forward stepwise selection, the final prediction model included age, baseline body mass index (BMI), BMI change in 1 year, baseline blood glucose, prescription of anti-diabetic medications (oral drugs or insulin) and high-density lipoprotein cholesterol (HDL-C) change in previous 1 year. NODM patients who were older (OR: 1.06, 95% CI: 1.01, 1.11), had higher baseline blood glucose (OR: 1.01, 95% CI: 1.00, 1.01), were prescribed anti-diabetic medications at baseline (OR: 3.40, 95% CI: 1.92, 9.70), had larger drop in BMI (OR: 1.24, 95% CI: 1.05, 1.45) and HDL-C (OR: 3.28, 95% CI: 1.33, 8.09) in previous year were more likely to be diagnosed with PAC (Table 1). Using the validation set, the model showed an AUC of 0.89 (95% CI:0.81-0.96) with a sensitivity of 90.0% and specificity of 72.3%. The number-needed-to-screen (NNS) to identify one PAC case among all NODM and high-risk NODM patients was 763 and 234, respectively.
Table 1: The final model for 1-year risk prediction of pancreatic cancer using logistic regression with forward stepwise.

Conclusion

Patients with NODM had 6.4-times higher risk of developing PAC within 1 year than patients without DM, and our new prediction model may help to identify high risk patients.