The influence of behavioural and socioeconomic factors on the community injury rates of adolescents assessed by the south Korean emergency medical services: an ecological approach

This cross-sectional observational study adopted an ecological approach and was conducted in the Republic of Korea (hereafter ‘Korea’). According to the 2005 Korean census, the percentage of the population between the ages of 10 and 19 was 13.9%. [17] Suicide and road traffic accidents were the first- and second-leading causes of adolescent mortality in 2015, respectively.

To evaluate the ecological effect of risk behaviours of the peer group, we used data from the 2014 Korea Youth Risk Behaviour Web-based Survey (KYRBS). KYRBS was established in 2005 and is an ongoing national cross-sectional survey conducted by the Korea Centres for Disease Control and Prevention (KCDC). [18] KYRBS assesses health-risk behaviours among middle- and high-school students. KYRBS conducted a cross-sectional survey using samples that were newly extracted each year. Therefore, the temporal data for 2014 reflects the regional status at the time of the survey. Subjects to be surveyed were sampled using a multi-stage cluster sampling design nationally representing Korean middle- and high-school students aged 12–18 years. In 2014, 72,060 of 3,532,149 students were sampled. Students anonymously completed self-administered web-based questionnaires during one class period. [18] The 2014 questionnaire consisted of 125 items in 15 domains: tobacco use; alcohol use; physical activity; dietary behaviours; obesity and weight control effort; mental health; unintentional injury; oral health; personal hygiene; substance use; sexual behaviour; atopy and asthma; internet addiction; demographic variables and health equity; violence and perceived health status. [18] The variables that did not appear to be directly related to the occurrence of the injury or that did not show any statistical correlation were excluded. Finally, based on literature evidence [19] and the expert opinion of the investigators, three items were selected that likely have a direct impact on the incidence of injury: 1) current alcohol consumption (cAlc) defined as alcohol intake more than once within the last 30 days (excluding alcohol use related to ancestral rites or sacraments); 2) victims of violence or bullying (VicVB) defined as being treated by a doctor for physical violence within 12 months; and 3) having undergone formal education on injury prevention (Edu-IP) defined as safety education (about danger and preventive behaviour, first aid, and emergency evacuation and rescue) at school more than once within 12 months. The first two variables were risk behaviours, while the third was considered preventive behaviour. KYRBS variables were calculated as proportions after considering the complex survey design, selection probabilities, survey non-responses, and post-stratification using the survey procedures (proc survey data from SAS ver. 9.4 according to KCDC recommendations). Weighting factors were assigned to each KYRBS school related to the inverse probability of selection, enabling the data to represent the national population. This study used only results from students aged 13–15 years.

We used the 2010 Korea Census Data as an ecological index of the SES of each region. The 2010 Population and Housing Census data were collected by Statistics Korea, which is part of the Ministry of Strategy and Finance. [20] In Korea, unearned income attributable to assets such as real estate, while important, is difficult to identify, so SES cannot be evaluated using only the income level in the census. [21] Other studies have also used the residential environment and education attainment, rather than income level alone, as indicators of SES. [22] In this study, socioeconomic factors included 1) the degree of urbanisation according to administrative district standards, 2) the proportion of high educational attainment, and 3) the proportion of households not in an apartment (e.g., households in detached houses, row houses, houses within a commercial building, and others) [22] in each region. The degree of urbanisation was divided into three categories: urban, suburban, and rural. Higher educational attainment was defined as the proportion of the population with at least a high school diploma. The level of the residential environment was evaluated using the proportion of households not in an apartment. Regarding the type of housing, unlike in the United States and other western countries, most luxury houses in Korea are built as high-rise apartments. The households not in an apartment were categorised as the lower residential environment.

We set the unit of the region that performed the analysis in this study. The National Statistical Office provides census data for 255 administrative districts in Korea. By contrast, KCDC provides KYRBS data for 46 primary survey units (PSUs) to maintain the anonymity and representativeness of the survey. To integrate the data, the Korea census reconstructed the area according to the KYRBS PSUs. The representative SES value of a PSU was defined as the average of the census values surveyed in the administrative districts belonging to each PSU, and the urbanisation level was defined using the level defined by KCDC for each PSU. After integrating these datasets, the ecological analysis was performed in 43 PSUs.

As described above, each area was subdivided into three according to the degree of urbanisation. The 43 regions included 7 rural, 21 suburban, and 15 urban regions. Urban is coded with the letter ‘U’ and a number (from U01 to U15), suburban with the letter ‘S’ and a number (from S01 to S21), and rural with the letter ‘R’ and a number (from R01 to R07).

The region-specific Incidence Rates of Injury assessed by the Emergency Medical Service (IRI-EMS) were calculated. To calculate the total and gender-specific IRI-EMS, the region-specific number of injuries from the EMS records was used as the numerator and the region-specific number of registered adolescents aged 13–15 years as the denominator. In Korea, the national EMS database is maintained by EMS providers as part of their field management duties. Each provincial EMS headquarters completes the EMS run sheet, which is entered into an electronic database stored in the National Fire Agency information system. [23] The injury incidence rates were calculated for each region using the addresses of the patients recorded in the EMS database. We also used the physical address documented in the EMS for where the injury occurred. Regardless of the activity during injury and the intent to injury, if the physical address of the place of occurrence was a school, it was classified as an inside-school injury and other addresses were classified as outside-school injury. In the EMS dataset, the body part injured was described as free text by EMT and could not be used for the analysis.

The analyses were performed using Stata ver. 15. The national IRI-EMS was calculated and the mean and standard deviation of the IRI-EMS of the 43 PSUs were calculated. The IRI-EMS was calculated as described above by region and sex. The medians and interquartile ranges of the ecological factors of the 43 PSUs were calculated and the range, minimum, and maximum were calculated to identify regional gaps. To analyze the strength and direction of linear associations, we used Pearson’s correlation coefficient. Linear regression was used to examine the association between ecological factors and IRI-EMS. The linear regression model associations were visualised using scatterplots. The analysis was performed and stratified by gender. The IRI-EMS of each region was considered a dependent variable in the multiple linear regression. In this study, all independent and dependent variables were ecological, and the unit of analysis was not an individual, but a region (i.e., a population group). In multicounty or multiregional ecological studies, multiple linear regression and Pearson correlation are used to identify significantly correlated factors. [24, 25] We tested multicollinearity by variance inflation factors (VIF) and interaction between variables. For each model, heteroskedasticity was evaluated using the Breusch-Pagan and Cook-Weisberg test (Stata command: estat hettest). A model that cannot assume homoscedasticity was adjusted to a robust standard error by the Huber / White / sandwich estimator. The level of statistical significance was set at P < 0.05.