Decision support tools for pancreatic cancer detection: external validation in Australian primary care - a retrospective cohort study.

[BACKGROUND] Pancreatic cancer is often diagnosed at an advanced stage with poor survival. Risk assessment tools have been developed to aid early diagnosis of pancreatic cancer in primary care settings (QCancer, electronic Risk Assessment Tool [eRAT], and the Queensland Institute of Medical Research [QIMR] Berghofer Pancreatic Cancer Decision Support Tool [QPaC Tool]) but have not been validated in the Australian setting.

[AIM] To estimate and compare the performance of these tools for identifying patients with undiagnosed pancreatic cancer using Australian primary care data.

[DESIGN AND SETTING] A cohort study was conducted using linked primary care and cancer registry data from Victoria, Australia.

[METHOD] Patients presenting to primary care with signs and/or symptoms included in the tools (recorded in the primary care 'reason for encounter') were included. Diagnostic accuracy statistics for each tool (and their individual signs and symptoms) were compared.

[RESULTS] Patients with pancreatic cancer were more likely (<0.001) to present with new-onset diabetes, jaundice, and unexpected weight loss pre-diagnosis than patients without pancreatic cancer. The most common pre-diagnostic presentations in patients with pancreatic cancer were jaundice (29.0%), abdominal pain (25.6%), change in bowel habits (17.6%), and new-onset diabetes (14.8%). Jaundice, steatorrhoea, and pancreatitis had the highest positive predictive values (PPV) for pancreatic cancer (1.96%, 1.77%, and 0.89%, respectively). Among the tools, eRAT had the highest PPV of 1.37% (95% confidence interval [CI] = 1.12 to 1.66); the PPV for QPaC was 1.01% (95% CI = 0.82 to 1.22) and QCancer was 0.8% (95% CI = 0.54 to 1.15).

[CONCLUSION] When applied to Australian primary care data, none of the tools were strongly predictive of pancreatic cancer. New diagnostic models incorporating additional data could potentially improve their predictive performance.