Developing Statistical Model for Identifying New Cancer Cases from a Medial Accounting System for Enhancing Cancer Registry Tasks: Multicenter Study

Abstract

　The national cancer registration project that began in January 2016 requires medical institutions to submit medical cases and makes it possible to collect what would be universal information on cancer. However, such medical institutions have a hard time being efficient in finding registered cancer cases in many outpatients and admitted patients that they have. Therefore, this research developed the statistical model that discriminates registered cancer cases in clinical data from several institutions and evaluated its discrimination abilities. The discrimination model, developed based on retrievable data from a medical accounting system, had AUC of 0.953. The discrimination ability (95％ confidence interval) of the model using evaluation datasets had a 92.0％ (90.5％-93.3％) sensitivity and an 89.1％ (88.7％-89.6％) specificity with accuracy. It excluded 82.2％ of cases that are subject to a search for cancer registration. This discrimination model can be used in many hospitals that provide healthcare services through insurance and has the potential to increase efficiency in operations related to registered cancer case searches.