In the National Health Information DB of the National Health Insurance Service, data from July 2008 to June 2018 were used, based on the sickness details tables, medical services statement tables, and the qualification and insurance tables.8 In addition, since health insurance database is a claims data, it was reconstructed as epidemiologic research unit, episode of care.

Influenza disease was reset from July 1 to June 30 of the following year, considering the characteristics of the disease that occurs during the winter season.

Health insurance database is a claims data that may be claimed separately for a variety of reasons, even though it was actually one episode of case. Because a single disease uses different medical services for different periods and times, separate claims are generated, and therefore separate claims can be used as epidemiological data only after they are connected and integrated.10

In this study, the analysis process grouped all claims that were directly connected in time as one episode of care, that is, the time between the end date and the date within 1 day of the end date and the next start date. The interval between the leading end dates and trailing start dates for the same person, which is a window period, is as follows (Fig. 1). Assuming that claim events within the same season are considered as one and the same episode of care, 100 days were set as a window period. In this way, 94.9% of all segregation claims were included within 100 days. When the claims data was converted to episode of care after applying 100 days of window period, there were some differences from year to year, but about 99% of all influenza patients experienced one outbreak. That is, only about 1% of patients experience more than one outbreak (Table 1).

In the health insurance claims data, primary diagnosis refers to the disease that causes the greatest demand for the treatment or examination (the disease that causes the most use of medical resources) and is listed first. Additional diagnosis is a condition that occurs with the main condition during the treatment period.8

Unlike severe diseases such as myocardial infarction, stroke, or cancer, influenza is less likely to be justified as a primary diagnosis. That is, if influenza occurs among existing chronic patients or other hospitalized patients, it is not likely to occupy the position of primary diagnosis but is likely to be added as an additional diagnosis or other illness. In consideration of these points, the influenza corporal subjects included all primary, additional and other diagnoses. The disease code was defined as J09, J10, and J11.

Using the 40T in the National Health Information DB, all cases of medical statement in 20T, qualifications, and insurance were combined, with the claims diagnosed as all sicknesses including primary, additional, and other diagnoses of influenza. A total of 24,605,290 claims were used to construct a 100-day episode of care. The final study consisted of 100-day episode of care constructed from all cases of influenza, and a total of 14,400,754 episodes were selected for analysis.

In the National Health Information DB, sickness details tables (40T), medical services statement tables (20T), qualification and insurance tables, and population group tables were used. 20T is a collection of medical services statement claimed by medical care institutions when patients use medical services. 40T includes disease codes extracted from 20T.8 The parameters of sex, age at onset of care, start year and month of treatment, start date of treatment, form code, disease number 1 (primary diagnosis), disease number 2 (secondary diagnosis), and disease treatment distribution were selected from sickness details tables. In the medical services statement tables, variables such as medical institution number, medical institution type code, medical department code, total hospital visit days, total prescription days, and payment status were selected. In the qualification and insurance tables, the month of care start date, adjusted address code, subscriber classification, and insurance variables were selected. Sex, base month, municipality, income level, and age variables were used in the population table and the insurance tables. The two tables were used as denominators representing the total population when we calculated incidence rate and APC effects. The keys connecting all the tables were a common key and a personal identification number.

The health insurance claims used in this study were reconstructed into 100-day episodes of care to be used as epidemiological data to perform basic analysis, calculation of CIR, and APC effect analysis.

In the basic and incidence analysis, the age was grouped on the unit of 5, the cohort was initiated with 3 years unit and reset to 5 years unit group. The incidence rate of influenza by age and cohort by year was calculated by using the number of the corresponding cases as a numerator. The entire population table was created using the population table and the insurance table was used as the denominator.

The APC effect analysis analyzed three effects related to time to summarize the social variation, causes of disease, increase in age, and population characteristics, which is different from the original method that uses only age and period. The identification problem of the APC analysis, in which the three effects cannot be estimated simultaneously due to collinearity, in which the sum of age and cohort is the period variable, was solved using the Glimmix model using the statistical package SAS (SAS Institute Inc., Cary, NC, USA).11 In other words, the hierarchical APC (HAPC) analysis was performed using the mixed effect model. The specification of the model is as follows:

i, age group; j, time period; k, birth cohort.

In addition, assuming the Poisson distribution and correcting the population with the offset option, the relative risk was able to be calculated directly, rather than the odds ratio. In the HAPC analysis model, the fixed effect variables were analyzed using age, metropolitan city, income level by insurance type, and random effect variables with period and cohort (= birth year) variables.

This study was exempted from deliberation by the Hanyang University Institutional Review Board (IRB) for the exemption from IRB deliberation that does not include personally identifiable information (HYU-2019-04-021).

In terms of the number of episodes diagnosed as influenza, the number of episodes between the ages of males and females aged 0–9 was the highest in the 2017–2018 season after the 2009–2010 season (Table 2). In the 2016–2017 season a large number of episodes from 10 to 19 years were seen. In the 2017–2018 season, males and females over 45 years of age had more episodes than in the 2009–2010 season.

The incidence rate of males and females diagnosed with influenza CIR, (per 1,000 person) showed that between 2009 and 2010 0–9 years had the highest incidence rate (Table 3). Also, the incidence rate in school-age and adolescent ages was high. In 2016, the 2015–2016 season, the incidence rate of males aged 0–4 was 171.02 episodes per 1,000 people, and females were 173.31. In 2017, the 2016–2017 season, the incidence rate was 139.48 for males aged 10–14, 140.94 for females, 52.96 for males aged 15–19, and 60.07 for females. The incidence rate of the 2017–2018 season, in the ages of 40–50 and 60–64 was higher compared to that in the pandemic influenza season.

According to the CIR, (per 1,000 person) diagnosed as influenza, on 2010, the 2009–2010 season the incidence rate of males born in the 1923–1942 was 31.46–31.29 episodes per 1,000 people, 66.24 for males who were born between 1988–1992, and 172.24–253.39 for males who were born on 1993–1997 and higher (Table 4). The incidence of females was 63.32 for those who were born between 1978 and 1982, 69.54 for those who were born between 1983–1987, and 166.10–240.32 for those who were born on 1993–1997 and higher. There was an overall decline in 2009–2010 season after the pandemic influenza outbreaks but from 2014, cohort incidence rates continued to increase. In 2017, the 2016–2017 season, the incidence rate was 52.66 for males born in 1998–2002, 59.67 for females, 138.39 for males from 2003–2007, 139.76 for females, 176.06 for males born in 2008–2012, and 184.11 for females, which showed high incidence rates. In 2018, the season of 2017–2018, males were found to have a high incidence of 29.19 for 1963–1967 born and 243.79 for 2013–2017 born. Females had a higher incidence for 1953–1967 born and 1978–1987 born, 41.13 for the 1993–1997 born and 251.38 for 2013–2017 born. Overall incidence rate was higher in females.

The APC effect analysis was conducted in all episodes diagnosed with influenza. As a result, on age effect, the relative risk was 1.38 for males and 1.48 for females at age 4 based on age 13 (Fig. 2). In addition, it was confirmed that the relative risk was 1.23 for males and 1.32 for females at age 8 or school age. At age 13, higher relative risks were observed in females. The period effect was identified at the highest relative risk in 2010, the season of 2009–2010, and has been continuously increasing since the 2013–2014 season. The cohort effect was gradually high in males and females born in the 1937's or lower. Males were found to have higher relative risks for 1970–1988 born and females in 1977–1988 born and the relative risk was higher in males. From people born in 2008 or later, the relative risk was high in females. Males born in 2015, and females born between 2012 and 2013, the relative risk was high.