1. Background

Workers play an important role in organizational productivity and social development. Therefore, as part of the efforts to maintain and promote the health of workers, many researchers have attempted various approaches to identify the influencing factors that explain their health and to develop interventions for them.

Among the various perspective of health, Rose has proposed a population-based approach, which suggests that individuals belong to the society and population and the explanation of the population's health should include the social context [1]. Individuals' health is dependent not only on personal risk factors but also on the environmental characteristics of the area surrounding them. For example, the social environment influences individuals by forming norms, reinforcing specific types of social control, and providing or limiting environmental opportunities for individuals to participate in certain behaviors [2]. Likewise, workers' health can be explained by the contextual factors of the population they belong to as well as the micro-level individual and workplace characteristics. Many studies on work and health have mainly dealt with occupational epidemiology focused on individual risk factors and occupational characteristics (physical and chemical risk factors) of workers, and more recently, attention has been paid to various factors including ergonomic and psychosocial work environments [3, 4].

The region is an important analytical unit where various factors work together and thereby form the quality of life of the local residents. As the interest in the place has been increased since 1990, some studies have pointed out that existing studies focused only on individual units and claimed that the region itself needs to be considered [5]. For example, the socio-economic environment of a residential area have been observed to be significantly associated with health behaviors and health status levels [6], and this association may be independent of the individuals' socio-economic characteristics [7]. These contextual effects at the regional level are independent effects of the higher analytical unit factors explaining the individuals' health behaviors and health status levels after controlling for the characteristics of members inherent in the higher analytical unit called the community, and are distinguished from the compositional effect exhibited by the group of individuals having specific characteristics.

The importance of the context for health interventions in the nursing area has long been recognized and has become evident due to an increase in community-based interventions. However, to date, the social health distribution and the impact of regional factors in relation to the health of workers have not been sufficiently taken into account in the studies of workers and interventions for them in the field of nursing. Recently, multidimensional analysis of the health-related factors of workers based on ecological models has been conducted in some studies, but they have been mainly focused on clarifying individuallevel and organizational-level factors [8, 9].

On the other hand, in other fields, studies have been accumulated to elucidate social determinants of health with a growing ecological interest in the social distribution of health. In previous studies that attempted multi-level approaches, various types of factors such as social class including socioeconomic status (SES)[10], social capital, or social relationships [11, 12] have been discussed as the social determinants of health. Various factors that represent the social structure, intervention level, and health outcomes are known to affect the differences in the health status of the population [13], and related studies have reported that regional variations in mortality, regional-level education, the financial autonomy rate, the current smoking rate, and the walking exercise rate are related to them [14]. However, regional variables that consistently affect individuals' health have not been systematically structured yet, and no studies which tested the regional variables for working population have been reported yet.

On the other hand, Benach et al.[15], explained that employment status, working conditions, and economic and policy factors of society may influence the health inequalities of workers through complex mechanisms, using theoretical models of employment relations and health inequalities (micro & macro level). The theory suggested by Benach et al.[15] has presented micro-level and macro-level contextual factors related to the health outcomes of workers. Based on these theoretical factors as well as Rose's population-based approach [1], this study intended to verify the regional factors explaining regional differences in the health status of workers. Therefore, we attempted to elucidate regional variables as the determinants of health status of workers, adjusting the indiviindividual-level factors such as socio-economic characteristics, working conditions, and occupational risk factors. The research question to be explored in this study is ‘What are the regional factors that affect the health status of workers?’

In this study, multilevel analysis was conducted as the methodology taking into account multidimensional effects including individual, environmental, and policy factors. Multilevel analysis makes it possible to simultaneously evaluate the associations between the determinants measured at various levels and health when factors affecting individuals' behaviors are present not only at the individual level but also at a higher level, such as a group or region, and may be suitably used as a research model for explaining the regional differences in the health status of workers. The data of the Korean Working Condition Survey used in this study are the sample data of Korean workers, and they allow for verification of various factors related to the employment conditions of workers. Therefore, this study aimed to present the grounds and direction of the regional-level approach to the interventions for workers through the multi-level analysis using the data of the Korean Working Conditions Survey and the regional factor indicators of the Statistics Korea.

2. Purpose

This study was conducted to determine whether the regional factors related to the areas where workers reside affect the health status of workers independently of the individual characteristics. The hypotheses of this study are as follows:

1. Study Design

This study is a descriptive study to elucidate individual and regional factors that affect the self-rated health of workers, and multi-level analysis was performed using hierarchical data based on the theoretical model of Benach et al. [15], which explains workers' health inequalities (Figure 1).

Figure 1 The conceptual model of this study. Click for larger image

2. Subjects and Data Collection

In this study, two-level data were collected for the construction of hierarchical data set. For the level 1 research data, the data from the 4th Korean Working Condition Survey (2014) conducted by the Korea Occupational Safety and Health Research Institute (KOSEF) was utilized. The Korean Working Condition Survey has been carried out since 2006 with the approval of the Statistics Korea (approval No. 38002), and it is a large-scale survey that provides the overall information on the work environment, including employment type, working conditions, type of occupation, and exposure to occupational risk factors, benchmarking the European Working Condition Survey. In this study, we used coded raw data excluding the data for which it is possible to estimate the subjects provided after the review by the data provider organization. The population was employed people aged 15 years or older in all households in Korea. The survey population was sampled through the secondary probability proportion stratified cluster sampling method by reflecting the criteria of the 2010 Population and Housing Census. In this study, among a total of 50,007 persons of the raw data, only wage workers aged 20 years or older were included. Among them, 25,069 workers were included in the final analysis excluding the residents in Sejong City, for whom it was difficult to produce regional indicators, soldiers, and missing data for major variables.

The level 2 regional units were classified based on the area codes classified in the Korean Working Condition Survey, and the regional units of 16 cities and provinces excluding Sejong City were included in the analysis. The data of each region were based on the statistical data by region of the Statistics Korea and the results of the research reports of the Korea Institute for Health and Social Affairs.

3. Measures

4. Data Analysis

In this study, χ² test and t-test were conducted through PASW statistics 24.0 (IBM Co., Armonk, NY, USA) to examine the difference in self-rated health according to individual-level and group-level factors of workers, and correlations between regional-level variables were determined by Pearson's correlation coefficient.

For the hierarchical analysis of individual- and regional-level variables, HLM 7.01 for Windows (SSI Inc., Skokie, IL, USA) was used, and multi-level logistic analysis was performed assuming the Bernoulli distribution of dependent variable. In the multi-level model, only variables of which statistical significance was confirmed through univariate analysis were entered (p<.05). For age, which is an individual-level variable, and regional-level variables, grand-mean centering was performed, which makes it possible to maintain the regional variations of variables and estimate the effects of regional-level variables more efficiently. In Model 1, which is the null model that does not include explanatory variables, the significance of regional differences for the dependent variable was determined and the validity of model setting was verified by calculating regional variance. In Model 2 (means outcome model) and Model 3 (random effect ANCOVA), regional-level and individual-level variables were applied, respectively, to examine the decrease in the regional-level variance. In Model 4, which is the variance component model including both individual-level and regional-level variables, the regional variables that explain the self-rated health of workers were identified while controlling for individual-level factors. The multi-level model was estimated by the Penalized quasi-likelihood (PQL) method. The distribution of major variables is shown in Table 1.

Table 1 Distribution of Variables Click for larger image

1. General Characteristics of Subjects

In terms of self-rated health, a total of 25,069 subjects were divided into the fair health status group of 24,020 persons (96.5%) and the poor health status group of 866 persons (3.5%). The average age of the subjects was 43.78±12.67 years, and males accounted for 51.5%. As for the level of education, 50.7% of the subjects were junior college graduates or higher. With respect to occupational characteristics, the proportion of workers earning the average monthly income of 1~1.99 million won was the highest (36.4%), and regular workers accounted for 77.0%. 52.7% of the subjects worked 35~47 hours per week, and 61.1% had atypical work experience. Regarding the exposure to occupational risk factors, the experience rate of the ergonomic risk factor was the highest at 59.5%. 11.2% and 14.8% of the respondents had low social support from their colleagues and supervisor, respectively. Regarding the workplace size, 72.7% were working in the workplaces with 50 employees or more (Table 2).

Table 2 Self-rated Health according to Individual-level Factors (N=25,069) Click for larger image

2. Individual- and Regional-level Factors and Workers' Self-rated Health

There were significant differences in the self-rated health of workers between the two groups according to demographic factors such as gender, age, and education level among the individual-level factors (p<.001). In terms of occupational characteristics, it was found that the group with the average monthly income of 3 million won or more and the group working for 35~47 hours per week were most likely to belong to the group with good self-rated health, and there was a significant difference in the self-rated health status between the two groups according to the employment type and the level of social support (p<.001). As for the relationship between self-rated health and exposure to occupational risk factors, there were differences in self-rated health according to the exposure to each of physical, chemical, and ergonomic risk factors (p<.001), but there was no statistical difference according to the exposure to the psychological risk factor (p=.139). In addition, there was no significant difference in the self-rated health of workers depending on the size of the workplace (Table 2).

The self-rated health of workers according to regional-level factors was obtained through a single dimension analysis by assigning regional-level variables according to the area codes assigned to individuals (Table 3). There were significant differences in the self-rated health of workers according to the current smoking rate, financial autonomy rate, number of physicians working in medical institutions per 1,000 people, and social trust. The harmful environment work rate and high-risk drinking rate in the region were not significantly related to the difference in self-rated health of workers.

Table 3 Self-rated Health according to Regional-level Factors Click for larger image

3. Effects of Individual- and Regional-level Factors on Workers' Self-rated Health

Based on the results of the univariate analysis of the individual- and regional-level variables, variables other than atypical work, the exposure to psychological risk factors, and the size of the workplace among individual-level variables and harmful environment work rate, and high-risk drinking rate among regional level variables were entered in the multi-level model. As a result of correlation analysis of the regional-level variables, there was no significant correlation between the variables (r<0.5, p<.05), and multicollinearity was not detected, so the results about this are not presented separately in this study.

Model 1 is the unconditional model where individual- and regional-level variables were not entered, and there were statistically significant regional differences in the self-rated health of workers (χ²=111.68, p<.001)(Table 4). The level 2 variance (τ) was 0.1141 and the ICC value of the basic model was 0.0335 as a result of calculating the value by applying the theoretical residual (π²/3) for the level 1 of the logistic model. Although the general suggested criterion for the ICC value is 0.05 or higher, multilevel analysis can be carried out if there are empirical study results on regional variations [19]. Considering the relatively small sample size at the regional level in the data of this study, the influences of variables were confirmed by verifying the variables in a stepwise manner based on the statistical significance of the variance value (τ).

Table 4 Multilevel Logistic Analysis for Wage Worker's Self-rated Health Click for larger image

In Model 2 in which only the regional level variables were applied, the financial autonomy rate (OR=1.03, p<.05) and current smoking rate (OR=0.97, p<.05) significantly explained regional differences in the self-rated health status of workers. In other words, the probability that the self-rated health of individual workers was good increased by 1.03 times when the financial autonomy rate of the region was increased by one unit, and decreased by 0.97 times as the current smoking rate of the region became lower. The ICC value for Model 2 was 0.0079, and the regional level variables explained 77% of the level 2 variance (τ) in the initial model, when individual factors were not taken into account (χ²=36.58, p<.001).

In Model 3, in which only the effects of individual-level variables were verified (χ²=84.64, p<.001), the odds ratio decreased by 0.75 times for women, by 0.96 times as age was increased by one year, by 0.77 times in case of weekly working hours of 48 hours or more, and by 0.80 and 0.51 times, respectively, when exposed to chemical and ergonomic occupational risk factors, showing that they are factors influencing workers' self-rated health negatively. On the contrary, the odds ratio was 1.97 times higher when the education level was junior college graduation or higher. Also, it increased with an increase in the average monthly income, and especially, it was 1.93 times higher in the case of the average monthly income of 2~2.99 million won. In addition, when the levels of social support of colleagues and supervisors were high, the odds ratio was 1.32 times and 1.26 times higher, respectively, indicating a higher likelihood that the self-rated health status of workers is good. These individual-level variables accounted for 18.4% of the regional variance (τ).

In Model 4, both individual- and regional-level variables were entered, and it was found that there were regional differences in the health status of workers with adjusted individual factors. The level 2 variance (τ) was lowered to 0.0316, and the input variables explained 72.3% of the total regional variance and were statistically significant (χ²=39.29, p<.001). The significance of the variables was not different from the results in Model 3, but overall, the odds ratio decreased slightly. In the case of regional-level variables, only the current smoking rate had a statistically significant effect on the self-rated health of workers (OR=0.97, p<.01), and the financial autonomy rate was not significant after entering individual-level variables (Table 4).