Data sources

We used commercially available health insurance claims and enrollment data of approximately three million individuals, maintained by the JMDC Inc. (Tokyo, Japan) [15], between January 1, 2005 and March 31, 2014 [16]. The JMDC database represents about 2.4% of the Japanese population as of 2014 and the data is anonymized. Data on beneficiaries in the 20- to 74-year age group who were workers in private firms along with their dependents, including the inpatient and outpatient claims in hospitals, and the dispensing claims of pharmacies were included. The insurance claims data contained information regarding the use of health care services, diagnoses, medical procedures, and the use of drugs. As the inpatient claims have the same format as the outpatient claims in Japan, the information concerning drug use during hospitalization was obtained from the inpatient claims. The claims database did not contain data of those aged 75 years and older as they were not covered by corporate health insurance but by the public health insurance (late-stage medical care system for the elderly) for all Japanese citizens aged 75 or older [17]. The enrollment data contained the year of birth, sex, and the dates of enrollment and disenrollment of the insured. In the claims data, more than 20,000 local drug codes were used to specify the trade name and generic name for all the approved drugs [18]. The generic name for drugs was coded by the Anatomical Therapeutic Chemical (ATC) code [19] and the diagnoses were coded using the International Classification of Diseases, 10^th revision (ICD-10) [20].