Project Description

Today, many cancers are often diagnosed at a late stage resulting in poor treatment options and survival rates. One of the most difficult cancers to detect early on is pancreatic cancer, which has one of the lowest 5-year relative survival rates at only 9%. Symptoms of pancreatic cancer, such as weight loss, abdominal discomfort and occasionally diabetes, are often mistaken for signs of less severe illnesses and overlooked in clinical practice. Other examples of cancer types diagnosed in a late stage are lung, colorectal, ovarian, stomach and liver cancer. Recently, deep learning has demonstrated to be a highly effective methodology to learn complex data structures, such as healthcare data. Danish health registries are some of the largest and most comprehensive healthcare datasets in the world. These comprise disease history, clinical notes, laboratory measurements, drugs and diagnostic images. Due to the linkability by the Central Person Register (CPR) identification number, it is possible to achieve individual-level linking across all data types.