Establish a Predictive Model for Pancreatic Cancer

According to the health promotion administration ministry of health and welfare, Taiwan latest cancer registration report in 2016, 2,202 people were diagnosed with pancreatic cancer and 1,996 died of pancreatic cancer. It ranks 8th in the male and 5th in the female of cause of cancer death. Compared with other cancers, the incidence of pancreatic cancer is not high, but “the incidence is almost equal to the mortality rate”, which is the most terrible situation for pancreatic cancer. In this study, the NHIRD was used to try to find out the diseases and symptoms associated with pancreatic cancer, and to design a prediction model for pancreatic cancer within three years. Allow patients with high risk of pancreatic cancer to check as early as possible to achieve early diagnosis. This study is a Cohort Study. Subjects were pancreatic cancer patients between 2000 and 2009, and the patient’s health record was tracked three years prior to diagnosis. Finally, the total of 7 diseases were selected as disease characteristics, including Other symptoms involving abdomen and pelvis (ICD – 9: 789), Diseases of pancreas (ICD – 9: 577), Peptic ulcer, site unspecified (ICD – 9: 533), Chronic liver disease and cirrhosis (ICD – 9: 571), Functional digestive disorders, not elsewhere classified (ICD – 9: 564), Gastric ulcer (ICD – 9: 531), and Diabetes mellitus (ICD – 9: 250). Complete the linear regression equation to calculate the probability of disease and remind potential patients to make an early diagnosis. The accuracy of the research results is 68% accuracy, Sensitivity = 77.8 and Specificity = 66.3, it reveals 73% AUROC (95% CI 0.67–0.79) model precision. © 2020, Springer Nature Singapore Pte Ltd.