Performance of the EndPAC risk score for pancreatic cancer in new onset diabetes patients with missing data - a database study in the British population.

Introduction: New onset diabetes (NOD) can be a first symptom of pancreatic ductal adenocarcinoma (PDAC), but less than 1 % of NOD are caused by PDAC. The EndPAC score tests for PDAC risk in NOD patients. As necessary information is often missing, the aim of our study was to assess the performance of the score using methods that adjust for missing values so the score can be successfully applied to all patients.

Patients/methods: We retrospectively followed a British cohort with NOD in the Clinical Practice Research Datalink until they developed PDAC or were censored. We calculated the EndPAC score in all patients and assessed its performance with different imputation methods for missing values. We calibrated the score for the British population.

Results: We included 197'092 NOD patients. PDAC occurred in 901 cases within 3 years after the diabetes diagnosis. Complete information to calculate the EndPAC score was available for 9.2 % of the patients. In those, the AUC (Area under the Receiver Operating Curve) of the original EndPAC score was 0.76. Including all patients, using the imputation of the population median for missing values, the AUC was 0.69. It improved to 0.71 after calibration to the UK population.

Conclusions: Use of imputation methods enabled us to use the EndPAC score for all NOD patients. However, use of the EndPAC score alone is still not sufficient to select NOD patients for diagnostic workup, with or without complete information. Its use in combination with a biomarker might lead to a better risk-benefit ratio.