Background: Pancreatic cancer (PC) has an overall five-year survival rate of 7%, although the survival rate differs greatly between stages (65% for stage IA and 4% for stage IV). Early detection is therefore of outmost importance, although only 2% are diagnosed in stage IA. Due to low prevalence, a general screening for PC is not optimal and identification of high-risk groups are needed. The aim of this project is to discover risk predictive diagnoses and blood test results for developing PC, and thereby identify high-risk populations, potentially eligible for screening in the future. Methodology: This is a retrospective case-control study. The study cohort was selected through searches in The Skåne Healthcare Register (SHR). Base data (gender, age) were collected for cases (presence of ICD-code C25.X, pancreatic cancer) and controls (absence of ICD-code C25.X) from 2003 and onward. Cases and controls with less than five ICD-codes the last five years were excluded. Cases were matched with four randomly assigned controls based on gender and age. Exposure (ICD-codes and blood test results) were acquired five years before the diagnosis of pancreatic cancer for cases and, for controls, five years before the age of matching with the correspondent case. The exposures were time-stamped with six months intervals to create an exposure-to disease-progression timeline. A deep machine learning algorithm will be trained, evolved and tested on these timelines. The algorithm will be evaluated in different time intervals up to 5 years before the diagnosis of pancreatic cancer. Significance Identification of pancreatic cancer-specific composition of time dependent diagnoses may help define high-risk populations, which may be potential candidate for future screening.