The study was conducted in accordance with the declaration of Helsinki and approved by the Institutional Review Board at National Center for Global Health and Medicine (Approval No. NCGM-G-003098-01). Patient identifiers were not included in our data used in this study and therefore no informed consent was required to conduct the present study.

This was an observational ecological study using the Japanese national database (NDB) in 2019. All methods were carried out in accordance with the STROBE statement. The NDB is a database of anonymized electronic health insurance claims covering more than 99% of all national claims with data regarding medical outpatient and inpatient services, diagnostic procedural combinations, dental services, and dispensed medications [19, 20]. Our proposal document for the use of the NDB was reviewed by the Japanese Ministry of Health, Labour and Welfare (MHLW), which granted approval for its use. The outcome of this study was the defined daily doses (DDD) per 1000 inhabitants per day (DID). DDD was calculated according to the World Health Organization (WHO) definition in 2022 [21]. Population referred to that on October 1 in the study year [22]. Using dependent variables as described below, a multivariable negative binomial regression analysis was conducted to evaluate the impact of each variable on the outcome.

All systemic antimicrobials, defined as J01 in the Anatomical Therapeutic Chemical (ATC) classification, were included in the study [23]. AMU data from January 1 to December 31, 2019 were extracted for all 47 prefectures. The location of prescribing healthcare facilities was used to identify the prefecture of each prescription. Then, the DID was calculated by prefecture. AMU data extracted from the NDB was categorized using ATC classification [23].

We hypothesized that antimicrobial stewardship measures are associated with reduced AMU. For example, the national action plan includes the enhancement of education related to AMR as well as the development of infectious disease specialists. Therefore, the average age of physicians (assuming young physicians receive more AMR-related education) and the number of infectious disease physicians are selected as variables. In Japan, physicians are the only practitioners who can prescribe antimicrobials. In hospital settings, antimicrobial stewardship teams often comprise physicians (e.g., infectious disease physicians), pharmacists (e.g., board certified pharmacists in infection control), nurses (e.g., certified nurses in infection control), and clinical laboratory technicians (e.g., infection control microbiological technologists) that support and advise physicians on antimicrobial prescriptions. In community settings, community pharmacists assist and advise physicians on antimicrobial prescribing.

As the action plan also focuses on the reduction of AMU for upper respiratory infections (URIs), the number of URIs was also selected as a variable in this study. In addition, the action plan encourages center hospitals to take initiative in their local stewardship efforts (additional reimbursement for infection prevention in Supplemental Table S1). The adjustment was conducted with population factors (population age, sex, nationality, income, and education level) and other healthcare facility factors (the number of clinics, the number of hospitals, and the proportion of large hospitals).

On the basis of the hypotheses described above, the variables evaluated in this study included population sex (the proportion of female individuals), population age (the proportion of population < 15 years and ≥ 65 years), the average income per household, the population education level (the proportion of upper secondary graduates going to further education), the number of clinics per 10,000 inhabitants, the number of hospitals per 100,000 inhabitants, the proportion of hospitals with ≥ 500 beds, the average age of physicians, the number of certified infectious disease specialist physicians per 100,000 inhabitants, the number of hospitals with additional reimbursement for infection prevention 1 per 100,000 inhabitants, the number of hospitals with additional reimbursement for infection prevention 2 per 100,000 inhabitants, and the annual number of diagnoses related to URI per 1000 inhabitants per day. In this study, all variables were presented at the prefecture level.

Population age and sex, the average income per household, the proportion of upper secondary graduates going to further education, and the average age of physicians were obtained from the Statistic Bureau, Ministry of Internal Affairs and Communications [24]. The physician’s age was included in the variables because some studies showed physicians’ experience is related to antibiotic prescriptions [13, 17]. The number of clinics, the number of hospitals, the proportion of hospitals with > 500 beds, and the number of healthcare facilities with additional reimbursement for infection prevention 1 and 2 were obtained from regional bureaus of Health and Welfare (e.g., Kanto-Shinetsu) [25, 26]. The requirements for hospitals with additional reimbursement for infection prevention in 2019 are presented in Supplemental Table S1 [27]. The number of certified infectious disease specialist physicians was obtained from the Japanese Association for Infectious Diseases [28]. The annual number of diagnoses related to URIs was extracted from the NDB. The URI-related diagnoses included in this study are presented in Table S2.

Characteristics of variables were presented with median and interquartile range (IQR). To investigate the relationship between the outcome of DID and each variable, multivariable negative binomial regression analyses were performed to calculate adjusted rate ratios (aRR) with 95% confidence intervals. The aRRs of proportional variables were presented for a 1% increase in each variable. The negative binomial regression model was selected because of the significant overdispersion in the multivariable Poisson regression model and because of DDD being the count data. To avoid multicollinearity, variance inflation factors (VIFs) in all variables were evaluated. After discussion amongst the authors, the proportion of ≥ 65 years, the average income per household, and the number of hospitals per 10,000 inhabitants, factors with high VIF, were excluded from the multivariable analysis. The log of the population in the prefecture was used as an offset term in the negative binomial regression analyses. The analyses were implemented for all systemic AMU, and AMU by class (1, J01C penicillins; 2, J01D other beta-lactams; 3, J01M quinolones; 4, J01F macrolides, lincosamides, and streptogramins; 5, J01 all other systemic antimicrobials).

Two-sided p values less than 0.05 were considered to show statistical significance. The statistical analyses described above were conducted using Stata 14.2 (College Station, TX, USA).