Study population and data
We used the Shizuoka Kokuho Database for this study, which is an administrative claims database of insured persons of the National Health Insurance (NHI) and Late Elders' Health Insurance (LEHI) in the Shizuoka prefecture, Japan. The Shizuoka prefecture is located approximately at the center of Japan on the Pacific coast, with a population of approximately 3.6 million as of 2020; it is the tenth largest prefecture among the 47 prefectures in the country.
There are three main types of health insurances in Japan: the Employee’s Health Insurance (EHI), NHI, and LEHI; the EHI and NHI are for those who are aged ≤ 74 years, while the LEHI is for those who are aged ≥ 75 years [11]. The EHI is provided to employed workers (company employees) and their dependents and insured by many insurers (number of insurers in Japan is more than 1,500), which is mostly dependent on the size of the company. Meanwhile, the NHI is designed for people who are not company employees (hence, not eligible to be members of the EHI), are aged < 74 years, and are insured by the prefectural and municipal governments (villages, towns, and cities). Those who are aged > 75 years, including self-employed persons aged > 75 years, are enrolled in the LEHI, which is insured by the prefectures. The Shizuoka Kokuho Database does not contain insurance claims data from the EHI.
The Shizuoka Kokuho Database also contains data on health check-ups, which are performed annually as part of the NHI and LEHI systems on a voluntary basis for those aged > 40 years at designated community centers and medical institutions [11]. A health check-up notification is sent to each household every year, based on the city’s family registry. The check-up comprises a physical examination, blood test, and self-reported medical history with a lifestyle survey.
In this study, we considered both the insurance claims data, which included data on prescribed medicines (detailing the year and month of prescription), and the health check-up data for all insured persons enrolled in the NHI and LEHI in the Shizuoka prefecture between April 2012 and March 2018 (2012–2017). These data were tied to individuals by anonymized individual identifiers for research purposes. More details about the database can be found elsewhere [12].
Eligibility criteria for analyses
In this study, we considered only individuals who had health check-up records (aged > 40 years) and could be followed up from 2012 to 2017. The database also included data on the dates when insured persons were enrolled into and withdrew from the NHI and LEHI schemes, and we included only those who were confirmed to have enrolled from 2012 to 2017. Insured persons who withdrew during this period were those who transferred their resident cards to another prefecture or those who transferred their insurance to the EHI scheme.
We included individuals who had health check-ups in both 2012 and 2013. We excluded individuals who self-reported undergoing diabetes treatment or dialysis therapy during the health check-ups between 2012 and 2013. In addition, we excluded those who were newly prescribed with diabetes medications, including injection drugs, between 2012 and 2013; we confirmed this from the insurance claims data [13]. We also excluded those without HbA1c data.
Statistical analyses
The specific objectives of this study were, among people without a history of diabetes, to assess the associations of HbA1c levels at the 2013 health check-up and transition trends in the HbA1c levels from the previous year with the likelihood of initiating diabetes treatment within the next 4 years (by 2017). Treatment initiation was defined as a case in which a drug was prescribed more than once every three months, and whether the treatment was an oral drug or injectable drug was based on the type of drug used in Japan [13].
To evaluate the associations, we constructed two logistic regression models: (1) the associations of the likelihood of initiating diabetes treatment by 2017 with (a) HbA1c levels at a health check-up in 2013, (b) trends in the HbA1c level changes from 2012 to 2013, and (c) number of health check-ups after 2013 and (2) the associations of the likelihood of using injection drugs among those who began diabetes treatment by 2017 with (a), (b), and (c).
For (a), the HbA1c level was treated as a categorical variable, and upon considering the ease of clinical and policy decision-making as well as sample size, the two groups were as follows: normal group (< 6.5%, including the suspicious zone for prediabetes) and diabetes group (≥ 6.5%) [14,15]. For (b), the trends in HbA1c levels from 2012 to 2013 indicated changes in these groups and were defined as three categories: improving, no change, and worsening. Based on (a) and (b), we created the following categorical variables and included them in the regression models: normal group with no trend changes; normal group with improving trend; diabetes group with no trend changes; and diabetes group with a worsening trend (hereafter referred to as HbA1c Groups A, B, C, and D, respectively).
In the regression models, the selection of variables was based on the backward-stepwise method with a p-to-remove value of > 0.05. Covariates of primary interest, including the HbA1c levels and trends (represented by HbA1c Groups A–D) and number of health check-ups received after 2013 (c), were entered into the models, regardless of their significance and as long as stable models were obtained.