National Database of Health Insurance Claims and Specific Health Checkups of Japan (NDB): Outline and Patient-Matching Technique

Abstract Background The National Database of Health Insurance Claims and Specific Health Checkups of Japan (NDB) is a comprehensive database of health insurance claims data under Japan’s National Health Insurance system. The NDB uses two types of personal identification variables (referred to in the database as “ID1” and “ID2”) to link the insurance claims of individual patients. However, the information entered against these ID variables is prone to change for several reasons, such as when claimants find or change employment, or due to variations in the spelling of their name. In the present study, we developed a new patient-matching technique that improves upon the existing system of using ID1 and ID2 variables. We also sought to validate a new personal ID variable (ID0) that we propose in order to enhance the efficiency of patient matching in the NDB database. Methods Our study targeted data from health insurance claims filed between April 2013 and March 2016 for hospitalization, combined diagnostic procedures, outpatient treatment, and dispensing of prescription medication. We developed a new patient-matching algorithm based on the ID1 and ID2 variables, as well as variables for treatment date and clinical outcome. We then attempted to validate our algorithm by comparing the number of patients identified by patient matching with the current ID1 variable and our proposed ID0 variable against the estimated patient population as of 1 October 2015. Results The numbers of patients in each sex and age group that were identified with the ID0 variable were lower than those identified using the ID1 variable. By using the ID0 variable, we were able to reduce the number of duplicate records for male and female patients by 5.8% and 6.4%, respectively. The numbers of children, adults older than 75 years, and women of reproductive age identified using the ID1 patient-matching variable were all higher than their corresponding estimates. Conversely, the numbers of these patients identified with the ID0 patient-matching variable were all within their corresponding estimates. Conclusion Our findings show that the proposed ID0 variable delivers more precise patient-matching results than the existing ID1 variable. The ID0 variable is currently the best available technique for patient matching in the NDB database. Future patient population estimates should therefore rely on the ID0 variable instead of the ID1 variable.