**Dietary Health-Related Risk Factors for Women in the Polish and Croatian Population Based on the Nutritional Behaviors of Junior Health Professionals**

#### **Dominika Gł ˛abska 1,\*, Valentina Raheli´c 2, Dominika Guzek 3, Kamila Jaworska 1, Sandra Bival 2, Zlatko Giljevi´c 4,5 and Eva Pavi´c <sup>2</sup>**


Received: 4 August 2019; Accepted: 13 September 2019; Published: 17 September 2019

**Abstract:** In Poland and Croatia, similarly as for a number of European countries, anemia and osteoporosis are common diet-related diseases in women, while for both the proper nutritional behaviors and preventive education are crucial. However, for the proper nutritional education there are some barriers, including those associated with an educator, his own nutritional behaviors and beliefs. The aim of the study was to assess the dietary health risk factors for women in the Polish and Croatian population based on the nutritional behaviors of junior health professionals. The study was conducted in Polish (n = 70) and Croatian (n = 80) female students of the faculties associated with public health at the universities in capital cities. Their diets were assessed based on 3-day dietary records. Nutritional value and consumption of food products, as well as the dietary risk factors for anemia and osteoporosis, were compared. While assessing the risk factors for anemia, in the Polish group, the higher intake of iron and folate, as well as vitamin B12 per 1000 kcal, was observed; and for folate, the higher frequency of inadequate intake was stated for Croatian women. While assessing the risk factors for osteoporosis, in the Polish group, compared with the Croatian, the higher intake of calcium per 1000 kcal was observed, but for vitamin D, there were no differences. Differences of the intake between the Polish and the Croatian group of junior health professionals may result in various dietary health risks for women. Based on the assessment of dietary intake, for anemia, compared to Polish women, a higher risk may be indicated for Croatian women, but for osteoporosis, similar risks may be indicated for Polish and Croatian women. Therefore, for public health, adequate nutritional education of junior health professionals is necessary.

**Keywords:** diet; nutrition; intake; public health; health professionals; dietary risk

#### **1. Introduction**

According to the World Health Organization (WHO), a number of women die because of noncommunicable diseases before they reach the age of 70—it was estimated that there were 4.7 million of such deaths in 2012 [1]. Women also generally suffer due to decreased quality of life and well-being—this indicates that there is a need to work toward improving the health and well-being of women and girls, beyond maternal and child care [2].

A number of noncommunicable diseases are diet-related; hence, nutritional education is crucial to prevent these diseases. However, it has been observed that it is not enough for nutritional educators to have the proper qualifications, including the right knowledge and skills, to affect change—the educators' own beliefs may influence health education [3]. The nutritional educators' own nutritional behaviors and beliefs may influence the information that is presented to patients and may interfere with the process of imparting knowledge [4]. At the same time, nutritional educators should follow diet recommendations; previous studies have shown that if educators had excessive body mass, it negatively affected the perception of their credibility, level of trust, and inclination of the patient to follow their advice [5].

Providing dietary recommendations is a serious problem as, generally, physicians do not even engage in nutritional counseling [6], while the other nutritional educators often do not have adequate knowledge and skills [7–9]; sometimes, even dietitians do not have it [10–12]. Although some studies indicate that dietitians follow their own advice of maintaining a well-balanced diet [13], others conclude that the nutritional behaviors of dietitians generally are not in agreement with their own recommendations; both dietitians and other nutritional educators consume inadequate amounts of fruits and vegetables [14–16] and fish [14,17]—they also consume sweets [17] and fast food meals [16] too often. This data corresponds to the excessive body mass commonly reported for physicians [18], nurses [19], educators [20], and also for dietitians [21]. Moreover, malnutrition has also been reported in dietitians [15] and the lower body mass is often related with symptoms of orthorexia nervosa [22]. The improper nutritional behaviors of health professionals are associated with a diet that is characterized by inadequate intake of fiber [23,24], calcium [23,25], iron [19,23], potassium [23], vitamin A [19], vitamin D [23], riboflavin [19], folate [19,23], and vitamin B12 [19].

Taking all this information into account, two major problems can be highlighted. These problems are related with educators providing misleading information to their patients based on their own beliefs, which results in the patients maintaining an unhealthy diet; in other cases, patients do not follow the nutritional educators' recommendations because of loss of trust and confidence in them. Consequently, it may be assumed that proper nutritional behaviors of nutritional educators may be crucial for effective education; if the educators do not follow proper nutritional behavior, they may directly generate health-related dietary risk factors for patients.

Anemia [26] and osteoporosis [27] are the diet-related diseases typical for women. The two conditions may be associated, and when they occur at the same time, they cause even more serious reduction of the quality of life in the affected individuals [28].

The frequency of decreased blood hemoglobin concentration (< 120 g/L), interpreted as the occurrence of anemia according to the WHO [29] definition, is high even for the developed regions of the world, especially for women [30]. According to WHO estimations, the prevalence of anemia in nonpregnant women aged 15–49 is 22.5% (ranging from 16.4 to 30.1%) in the European region, being 23% in Poland and 24% in Croatia; for pregnant women, the frequency is even higher [31]. The primary diet-related reasons for anemia are deficiencies of iron [32], folate [33], and vitamin B12 [34], although WHO indicated iron deficiency as the most prominent contributor [35].

The frequency of osteoporosis, according to the statistics of the International Osteoporosis Foundation (IOF) [36], for European countries is also high, as for women aged over 50 (the group especially prone to osteoporosis), its occurrence is 21–24%, contributing to an estimated lifetime risk of hip fracture of 22.8%. In this age group, the frequency of osteoporosis in Poland is also estimated at about 20%, contributing to hip fractures in 3% of the female population [37]. As was observed by Cvijeti´c et al. [38], for Croatia, there are no detailed osteoporosis statistics; but it is estimated that 39% of individuals aged over 60 have osteoporosis, with 95% of them being women [39]. At the same time, based on the data for Zagreb (the capital city of Croatia), the frequency of vertebral fractures for women

in the group aged over 50 was estimated as 9.7% [40]. As calcium intake is directly associated with bone mass, it is the primary diet-related influencing factor [41]; however, vitamin D is also a factor that contributes to better calcium absorption and bone mineralization, and thus bone mineral density [42].

In this study, our aim was to assess the dietary health risk factors for women in the Polish and Croatian population based on the nutritional behaviors of junior health professionals. Taking into account the comprehensive quality of diet, we planned to assess not only the daily intake, but also the proportions of macronutrients (the share of energy contribution), as well as the nutrient density of diet [43] (expressed as daily intake recalculated per 1000 kcal). Moreover, we also included the assessment of food products intake and food products intake recalculated per 1000 kcal. In order to include issues related to the environmental effects of food production, we decided to assess, under food products intake, animal-derived products separately; under the intake of protein, we decided to assess animal and plant protein separately. Western populations generally have too high total protein intake [44]; it has also been observed that environmental impact reduction is proportional to the animal products share reduction [45]. This study was planned to be conducted by assessing the dietary risk factors for anemia and osteoporosis, which are the most common diet-related diseases for women and may influence the social sustainability level in populations.

#### **2. Materials and Methods**

#### *2.1. Ethical Statement*

The study was conducted according to the guidelines of the Declaration of Helsinki. It was approved by the Bioethical Commission of the National Food and Nutrition Institute in Warsaw (No. 0701/2015). All the participants provided their informed consent to participate in the study.

#### *2.2. Studied Group*

The study was conducted among two groups of junior health professionals who were students of faculties associated with public health at universities located in capital cities. It was decided to recruit students that in the future should be characterized by a general nutritional knowledge only, and not specialists in dietetics or medicine. Hence, students of dietary and medical courses were not included; only those who planned to become general nutritional specialists conducting group nutritional education in and outside of hospitals were recruited. For Poland, it was the Warsaw University of Life Sciences (WULS-SGGW) in Warsaw, whereas for Croatia it was the University of Applied Health Sciences (Zdravstveno Veleuˇcilište u Zagrebu—ZVU) in Zagreb.

The students were invited to participate in the study in the semester when they had their dietitian classes to assess nutritional behaviors of students characterized by a similar level of nutritional knowledge. The inclusion criteria were as follows:


The exclusion criteria were as follows:


In the indicated universities, based on the indicated inclusion and exclusion criteria, two independent groups of participants were recruited: 70 young women in the Polish group and 80 young women in the Croatian group.

#### *2.3. Assessment of the Diet*

The assessment of the diet was based on a 3-day dietary record, according to widely applicable rules [46]. The dietary assessment was conducted by respondents during three random and nonconsecutive days, which were commanded to be typical and two of them were to be weekdays and one weekend day. The study's participants had a structured form to be completed with the information about consumed meals: the place where they were consumed, consumption time, and detailed description (products included, applied culinary and cooking techniques, serving size). The serving size was to be indicated either in grams (if they possessed a kitchen scale or consumed packed products with such information on the packaging) or in a descriptive manner (as a standard household measures). The respondents were instructed about the need to note all the products and all the beverages in a scrupulous manner. Moreover, they were informed about the need to not change their typical dietary habits because of keeping the dietary record.

Both groups of respondents were given identical forms (in their native language, either Polish or Croatian) and identical instructions to conduct the 3-day dietary record. The instructions were translated into Polish/Croatian by native Polish/Croatian speakers who were dietitians to obtain an accurate description of necessary issues, including some examples as to how to keep the dietary record.

Subsequently, to obtain a comparable analysis of the diet, independent from the applied databases, the same tables of nutritional value of food products and dishes were applied for both groups. The Polish tables were selected; therefore, the dietary records obtained for the Croatian group were translated to Polish, as in the previously conducted own study [47]. The translation was independently conducted in two stages: two English-speaking Croatian dietitians translated them from Croatian to English (while describing the specific dishes and providing the photographs of typical Croatian products and information about recipes, if needed), and afterwards two English-speaking Polish dietitians independently translated Croatian dietary records from English to Polish (verifying all the doubts with Croatian dietitians, if needed). The applied procedure resulted in obtaining both Polish and Croatian dietary records translated into Polish to be analyzed using the same database (the Polish one).

The Polish 3-day dietary records and the Croatian ones, translated into Polish (in a 2-step process via English), were afterwards analyzed in four ways as follows:


were deconstructed into food products, and they were then divided into listed groups, which are commonly applied for assessing the intake of food products [49];

• The typical consumption of food products recalculated per 1000 kcal of the diet, which was assessed for the food product groups, independently recalculated to enable reliable comparison.

While assessing sodium intake, the salt added to dishes was not calculated, but only that naturally present in food products or in processed food products. This is because, in general, dietary intake assessment is not a recommended method to analyze sodium intake; the 24-hour urine collection is the 'gold standard' for this assessment [50] and is commonly applied [51]. Moreover, dietary intake assessment methods usually do not capture the amount of sodium obtained from salt added to dishes consumed [52]. This is because, in western countries, excessive sodium intake is common; in practice, the risk of inadequate intake does not exist [53].

Subsequently, the groups were compared for the obtained nutritional value and the intake of food products. Moreover, the dietary health risk factors were specified for each population, and the most common ones for young women's diet-related health risks were selected and the related nutrients were analyzed as follows: anemia (iron, folate, and vitamin B12) [54] and osteoporosis (calcium and vitamin D) [55]. Vitamin D was included in the analysis in spite of the fact that it is mainly generated in skin after sunshine exposure [56]. However, inadequate exposure that is commonly stated results in need for at least adequate dietary intake [57]. Combined inadequate sunshine exposure and inadequate intake result in vitamin D deficiency [58] that is observed both for Poland [59] and Croatia [60].

The intakes of indicated nutrients in groups were compared with the recommended intake values, as specified by National Institutes of Health [61], while the Estimated Average Requirement (EAR) values were selected as a reference to estimate the prevalence of inadequate intake [62]. We applied the following reference values: calcium (800 mg), iron (8.1 mg), folate (320 μg), vitamin B12 (2 μg), and vitamin D (10 μg) [61].

#### *2.4. Statistical Analysis*

The distribution was verified using the Shapiro-Wilk test. Afterwards, the t-Student test (for parametric distributions) and the U Mann-Whitney test (for nonparametric distributions) were applied for comparison of the typical intakes. At the same time, the chi<sup>2</sup> test was applied for comparison of the prevalence of inadequate intake.

The statistical analysis was conducted for the accepted level of significance of *p* ≤ 0.05 and while using Statistica, version 8.0 (Statsoft Inc., Tulsa, OK, USA).

#### **3. Results**

Comparison of energy value and macronutrients intake in groups of junior health professionals from Poland and Croatia is presented in Table 1. In the analyzed group, Polish women, while compared with Croatian ones, were characterized by a higher protein (*p* = 0.0001), but lower carbohydrates share in energy value of the diet (*p* = 0.0068). At the same time, they were characterized by lower intake of sucrose (*p* = 0.0109) and starch (*p* = 0.0098), but higher intake of fiber (*p* = 0.0002).

Comparison of macronutrients intake per 1000 kcal in groups of junior health professionals from Poland and Croatia is presented in Table 2. In the analyzed group, Polish women, while compared with Croatian ones, were characterized by a higher total protein (*p* = 0.0001), animal protein (*p* = 0.0190), plant protein (*p* = 0.0270), polyunsaturated fatty acids (*p* = 0.0071), lactose (*p* = 0.0205) and fiber intake (*p* < 0.0001). At the same time, they were characterized by lower intake of sucrose (*p* = 0.0098) and starch (*p* = 0.0015).


**Table 1.** Comparison of energy value and macronutrients intake in groups of junior health professionals from Poland and Croatia.

EV—Energy Value; \* nonparametric distribution (verified using the Shapiro-Wilk test; *p* < 0.05); \*\* compared using the t-Student test (for parametric distributions) or the U Mann-Whitney test (for nonparametric distributions).

**Table 2.** Comparison of macronutrients intake per 1000 kcal in groups of junior health professionals from Poland and Croatia.


\* Nonparametric distribution (verified using the Shapiro-Wilk test; *p* < 0.05); \*\* compared using the t-Student test (for parametric distributions) or the U Mann-Whitney test (for nonparametric distributions).

Comparison of minerals intake in groups of junior health professionals from Poland and Croatia is presented in Table 3. In the analyzed group, Polish women, while compared with Croatian ones, were characterized by a higher potassium (*p* = 0.0059), phosphorus (*p* = 0.0002), iron (*p* = 0.0062), magnesium (*p* < 0.0001), zinc (*p* = 0.0086), copper (*p* = 0.0001), manganese (*p* = 0.0166), and iodine intake (*p* = 0.0001).

Comparison of minerals intake per 1000 kcal in groups of junior health professionals from Poland and Croatia is presented in Table 4. In the analyzed group, Polish women, while compared with Croatian ones, were characterized by a higher potassium (*p* = 0.0007), calcium (*p* = 0.0111), phosphorus (*p* < 0.0001), iron (*p* < 0.0001), magnesium (*p* < 0.0001), zinc (*p* < 0.0001), copper (*p* < 0.0001), manganese (*p* = 0.0015), and iodine intake (*p* < 0.0001).

Comparison of vitamins intake in groups of junior health professionals from Poland and Croatia is presented in Table 5. In the analyzed group, Polish women, while compared with Croatian ones, were characterized by a higher riboflavin (*p* = 0.0063), vitamin B6 (*p* = 0.0040), folate (*p* = 0.0027), vitamin A (*p* = 0.0008), and vitamin E intake (*p* = 0.0119). At the same time, they were characterized by lower intake of niacin (*p* = 0.0043).


**Table 3.** Comparison of minerals intake in groups of junior health professionals from Poland and Croatia.

\* Nonparametric distribution (verified using the Shapiro-Wilk test; *p* < 0.05); \*\* compared using the t-Student test (for parametric distributions) or the U Mann-Whitney test (for nonparametric distributions).

**Table 4.** Comparison of minerals intake per 1000 kcal in groups of junior health professionals from Poland and Croatia.


\* Nonparametric distribution (verified using the Shapiro-Wilk test; *p* < 0.05); \*\* compared using the t-Student test (for parametric distributions) or the U Mann-Whitney test (for nonparametric distributions).


**Table 5.** Comparison of vitamins intake in groups of junior health professionals from Poland and Croatia.

\* Nonparametric distribution (verified using the Shapiro-Wilk test; *p* < 0.05); \*\* compared using the t-Student test (for parametric distributions) or the U Mann-Whitney test (for nonparametric distributions).

Comparison of vitamins intake per 1000 kcal in groups of junior health professionals from Poland and Croatia is presented in Table 6. In the analyzed group, Polish women, while compared with Croatian ones, were characterized by a higher riboflavin (*p* < 0.0001), vitamin B6 (*p* = 0.0011), folate (*p* = 0.0006), vitamin B12 (*p* = 0.0074), vitamin A (*p* = 0.0004), and vitamin E intake (*p* = 0.0015). At the same time, they were characterized by lower intake of niacin (*p* = 0.0176).

Comparison of animal-derived products intake in groups of junior health professionals from Poland and Croatia is presented in Table 7. In the analyzed group, Polish women, while compared with Croatian ones, were characterized by a higher cottage cheese (*p* = 0.0062), but lower cold cuts intake (*p* = 0.0023).

Comparison of animal-derived products intake per 1000 kcal in groups of junior health professionals from Poland and Croatia is presented in Table 8. In the analyzed group, Polish women, while compared with Croatian ones, were characterized by a higher cottage cheese (*p* = 0.0084), but lower cold cuts intake (*p* = 0.0035).

**Table 6.** Comparison of vitamins intake per 1000 kcal in groups of junior health professionals from Poland and Croatia.


\* Nonparametric distribution (verified using the Shapiro-Wilk test; *p* < 0.05); \*\* compared using the t-Student test (for parametric distributions) or the U Mann-Whitney test (for nonparametric distributions).


**Table 7.** Comparison of animal-derived products intake in groups of junior health professionals from Poland and Croatia.

\* Nonparametric distribution (verified using the Shapiro-Wilk test; *p* < 0.05); \*\* compared using the t-Student test (for parametric distributions) or the U Mann-Whitney test (for nonparametric distributions).



\* Nonparametric distribution (verified using the Shapiro-Wilk test; *p* < 0.05); \*\* compared using the t-Student test (for parametric distributions) or the U Mann-Whitney test (for nonparametric distributions).

Comparison of other products intake in groups of junior health professionals from Poland and Croatia is presented in Table 9. In the analyzed group, Polish women, while compared with Croatian ones, were characterized by a higher vegetables (*p* = 0.0049), oil (*p* = 0.0238), nuts (*p* = 0.0455), and jam/honey intake (*p* = 0.0471). At the same time, they were characterized by lower intake of potatoes (*p* < 0.0001), chocolate sweets (*p* = 0.0042), and cakes/cookies (*p* = 0.0006).

Comparison of other products intake per 1000 kcal in groups of junior health professionals from Poland and Croatia is presented in Table 10. In the analyzed group, Polish women, while compared with Croatian ones, were characterized by a higher vegetables (*p* = 0.0035), oil (*p* = 0.0114), nuts (*p* = 0.0419), and jam/honey intake (*p* = 0.0425). At the same time, they were characterized by lower intake of potatoes (*p* < 0.0001), chocolate sweets (*p* = 0.0041), and cakes/cookies (*p* = 0.0044).

**Table 9.** Comparison of other products intake in groups of junior health professionals from Poland and Croatia.


\* Nonparametric distribution (verified using the Shapiro-Wilk test; *p* < 0.05); \*\* compared using the t-Student test (for parametric distributions) or the U Mann-Whitney test (for nonparametric distributions).


**Table 10.** Comparison of other products intake per 1000 kcal in groups of junior health professionals from Poland and Croatia.

\* Nonparametric distribution (verified using the Shapiro-Wilk test; *p* < 0.05); \*\* compared using the t-Student test (for parametric distributions) or the U Mann-Whitney test (for nonparametric distributions).

Comparison of the estimated prevalence of inadequate intake of chosen nutrients, while compared with the reference values [61], in groups of junior health professionals from Poland and Croatia is presented in Table 11. In the analyzed group, for folate (*p* = 0.0023), in the Polish group there is a lower risk of inadequate intake, while compared with the Croatian one.


**Table 11.** Comparison of the estimated prevalence of inadequate intake of chosen nutrients, while compared with the reference values [61], in groups of junior health professionals from Poland and Croatia.

\* The following reference values were applied: calcium (800 mg), iron (8.1 mg), folate (320 μg), vitamin B12 (2 μg), vitamin D (10 μg), [61]; \*\* compared using the chi<sup>2</sup> test.

#### **4. Discussion**

While comparing the nutritional value of the diets and the intake of products between groups of junior health professionals from Poland and Croatia, a number of differences cropped up. While comparing the estimated prevalence of inadequate dietary intake of chosen nutrients, such differences were stated only for folate; however, in both groups, a number of respondents were characterized by an inadequate dietary intake of other nutrients. It should be mentioned that such a situation results from following an improperly balanced diet [63]. As shown by the Food and Agriculture Organization of the United Nations (FAO) [64], in a properly balanced diet that is based on nutritional recommendations, a healthy individual can easily obtain the recommended intake of a majority of nutrients. Consequently, an inadequate intake of specific nutrients may be treated as an indicator of an improperly balanced diet. FAO [65] further emphasizes that a healthy diet associated with an adequate intake of nutrients may be obtained from various combinations of food products.

#### *4.1. Potential Influence of Health Professional Dietary Habits on Their Patients*

While analyzing the nutritional value of the diets of junior health professionals, it must be above all emphasized that they will educate the Polish and Croatian populations about recommended nutritional behaviors in the future. Consequently, their nutritional inadequacies may generate nutritional inadequacies for a number of individuals and may contribute to dietary health risk factors. This is based on the association between personal dietary habits and attitudes toward preventive counseling, which was observed in a group of physicians and medical students [66]. Furthermore, their own improper dietary behaviors may generate a lack of confidence in the ability to counsel patients regarding lifestyle that is commonly observed [67].

Some patients perceive their dietitian as a role model and focus on both the presented recommendations and the body size or shape; therefore, a number of dietitians are aware that they should also follow the same dietary recommendations that they present to their patients [68]. Furthermore, these educators who include information about they themselves following recommended dietary habits are perceived by patients to be not only healthier, but also more believable and motivating than others [69]. Similarly, the review by Lobelo and Quevedo [70] reported that healthcare providers who are physically active are more likely to provide physical counseling for their patients compared to others. Furthermore, they stated that healthcare providers can play an important role by becoming early adopters of physical activity and diet behaviors and, in turn, become role models for their patients and communities [71].

As the presented study was planned to assess the dietary intake of female junior health professionals, their diets were also analyzed from the point of view of future female patients and the diet-related diseases common in this group were analyzed.

#### *4.2. Dietary Risk Factors for Anemia*

According to the WHO [72], anemia, in industrialized countries, is diagnosed for one in ten women and for one in four pregnant women. However, the systematic analysis of population-representative data by Stevens et al. [73] indicated, based on hemoglobin concentration, a higher frequency of 29% in nonpregnant women and 38% in pregnant ones. This problem may lead to significant health-related consequences such as increased mortality [74].

In the presented study, a higher intake of iron and folate was observed in the Polish group compared to the Croatian group. Also, such an observation was noted in case of intake of vitamin B12 per 1000 kcal. A higher iron and vitamin B12 intake for Polish women was observed despite a higher intake of cold cuts (which are within their sources) in Croatian women. However, the frequency of inadequate intake was rather high for both countries. Moreover, a higher frequency of inadequate intake for folate was stated more commonly for Croatian women than Polish ones, that may have resulted from the lower vegetable intake in Croatian women. The observations correspond with the results of other studies indicating inadequate intake of iron, for Polish [75] and Croatian young women [76], as well as of folate, for Polish [77] and Croatian young women [76].

Based on the assessment of intake adequacy, the risk of anemia for young women in both countries is quite high. Furthermore, for Croatia, it may be even higher than that for Poland despite a higher intake of cold cuts. Although the intake of traditional meat products is high in Croatia, corresponding to high intake of cold cuts reported in this study, it mainly results in high fat intake [78]; however, it may not influence iron intake significantly.

#### *4.3. Dietary Risk Factors for Osteoporosis*

Osteoporosis is another disease that is commonly more observed in women than men because it is estimated that one in two women, but only one in three men, will experience osteoporotic fractures [79]. However, the prevention of osteoporosis is especially important because progressive bone mass loss is observed after the age of 30 [80], contributing to the high risk for osteoporosis [81].

In the presented study, a higher intake of calcium per 1000 kcal was observed in the Polish group compared to the Croatian group. This may have been associated with a higher cottage cheese intake in Polish women. However, a very low intake was observed for vitamin D in both groups; therefore, for both countries, the frequency of inadequate intake was high. The observations correspond with inadequate intake of calcium, observed for Polish [82] and Croatian young women [83]; however, some studies reported higher intake of calcium in Croatia [84]. Furthermore, the inadequate intake of vitamin D is commonly stated for both Polish [85] and Croatian [86] young women [47].

Based on the conducted assessment of the adequacy of intake, it must be stated that the risk of osteoporosis for young women in both countries is high and comparable, resulting primarily from the inadequate intake of vitamin D. Moreover, as the vitamin D status both in Poland [59] and Croatia [60] is commonly not adequate, due to insufficient sunshine exposure, it must be emphasized that the inadequate vitamin D intake probably is not compensated by the endogenous synthesis. Taking it into account, even if the inadequate intake of calcium is not so common, the risk of osteoporosis must be stated.

Taking into account, that it is stated, that health professionals commonly do not have proper healthy lifestyle behaviors that impact chronic diseases, they are often not prepared properly for preventive counseling with their patients, and do not present dietary recommendation to their patients [86], so the necessary actions should be taken to obtain a properly balanced diet following. It may be stated, that adequate nutritional education is needed, in order to improve their nutritional behaviors, embolden them to educate their patients, and increase the number of patients being counseled.

Although the present study provided new insight for identifying potential dietary health risks for women in Poland and Croatia, some limitations must be noted. For further studies, it would be valuable to conduct similar assessments in groups consisting of other nutritional knowledge providers (including physicians); male educators should also be analyzed. It should be emphasized that, both in Poland and Croatia, nutritional faculties are not present only in capital cities, so including specialists from other regions would allow us to gain a broader perspective.

#### **5. Conclusions**

A number of differences in the nutrients and food products intake between the Polish and the Croatian group of junior health professionals may result in various dietary health risks for women in these countries. For anemia, compared to Polish women, a higher risk may be indicated for Croatian women because of lower iron, folate, and vitamin B12 intake, and more common inadequate intake of folate. For osteoporosis, similar risks may be indicated for Polish and Croatian women because of very low vitamin D intake in both countries. Therefore, adequate nutritional education of junior health professionals is necessary.

**Author Contributions:** Conceptualization, D.G. (Dominika Gł ˛abska), V.R., D.G. (Dominika Guzek); methodology, D.G. (Dominika Gł ˛abska); formal analysis, D.G. (Dominika Gł ˛abska), V.R., D.G. (Dominika Guzek), K.J.; investigation, V.R., K.J., S.B., E.P.; data curation, D.G. (Dominika Gł ˛abska), V.R., D.G. (Dominika Guzek), K.J., S.B., Z.G., E.P.; writing—original draft preparation, D.G. (Dominika Gł ˛abska), V.R., D.G. (Dominika Guzek), K.J., S.B., Z.G., E.P.; writing—review and editing, D.G. (Dominika Gł ˛abska), V.R., D.G. (Dominika Guzek), K.J., S.B., Z.G., E.P.; project administration, D.G. (Dominika Gł ˛abska), V.R.

**Funding:** The research was financed by the Polish Ministry of Science and Higher Education with funds from the Faculty of Human Nutrition and Consumer Sciences, Warsaw University of Life Sciences (WULS), for scientific research.

**Conflicts of Interest:** The authors declare no conflict of interest.

#### **References**


© 2019 by the authors. Licensee MDPI, Basel, Switzerland. This article is an open access article distributed under the terms and conditions of the Creative Commons Attribution (CC BY) license (http://creativecommons.org/licenses/by/4.0/).

## *Article* **Research on the E**ffi**ciency of Local Government Health Expenditure in China and Its Spatial Spillover E**ff**ect**

#### **Mengying Wang and Chunhai Tao \***

School of Statistics, Jiangxi University of Finance and Economics, No. 169, East Shuanggang Road, Nanchang 330013, China; wangmengying@stu.jxufe.cn

**\*** Correspondence: taochunhai@jxufe.edu.cn; Tel.: + 86-0791-8381-6995

Received: 6 March 2019; Accepted: 22 April 2019; Published: 26 April 2019

**Abstract:** The efficiency of the local government health expenditure (GHE) in China determines the level of public health services. However, the local government does not pay much attention to that efficiency, though the scale of local GHE is increasing. In this paper, first, we use the data envelopment analysis (DEA) method to measure the static overall efficiency of the local government health expenditure (GHE) in each region of China from 2007 to 2016. Then, based on the spatial statistical theory, global and local spatial Moran's I value is utilized to investigate its spatial correlation and spatial agglomeration phenomenon. Finally, the spatial spillover effect (SSE) of the static overall efficiency of local GHE in each region is measured by constructing a spatial Durbin model (SDM). It is demonstrated that there are significant differences in the efficiency of the local GHE between different regions of China. In addition, it is shown that Moran's I value of the static overall efficiency of the local GHE from 2007 to 2016 is positive. It passed the test of the 5% significance level, indicating that there is a positive spatial correlation between the efficiency of the local GHE and a spatial spillover effect. On the other hand, the decomposition of the SDM reveals that the proportion of GHE to financial expenditure, gross domestic product (GDP) per capita, and population density have a positive effect on the efficiency of the local GHE. Hence, their growth will improve the GHE efficiency in the local region and neighboring regions. In contrast, the proportion of urban population, illiteracy, and fiscal decentralization have a negative effect. Thus, their growth will decrease the GHE efficiency in the local region and neighboring regions. The results are discussed and suggestions are given based on the analysis in this paper. The main contribution of this work is to consider the spatial spillover effect in terms with realistic meaning. The results obtained can be used as a reference for optimizing the structure and improving the efficiency of government health inputs. It breaks the government's GDP-only theory-based assessment system and helps to improve it by assessing the GHE efficiency.

**Keywords:** government health expenditure (GHE) efficiency; data envelopment analysis (DEA) method; Moran's I value; spatial spillover effect (SSE); spatial Durbin model (SDM)

#### **1. Introduction**

For a long time, China has been committed to building an efficient and sustainable social public health system (SPHS). Relevant policies are proposed by the government to improve efficiency and sustainability. Lots of science founding are established to make a substantial contribution to the improvement of public health. In addition, we know that the efficiency of government health expenditure (GHE) and healthcare sustainability can affect each other. However, the current Chinese medical and health services cannot meet the new requirements, that is, the people's growing demand for healthcare, the high quality of health services, and the wide coverage of the medical insurance. It is also very difficult to achieve the goal of an equal and sustainable healthcare service. Thus, it is important to study the GHE efficiency.

Nowadays, it is difficult and expensive to see a doctor in China, which is an outstanding social issue, and the relationship between doctors and patients is becoming terrible. In this context, the scale and efficiency of the government's social public health services supply have received much attention in all sectors of our society and they also constitute a hot research topic. On the one hand, it is an important perspective for local governments in exploring the causes of the imbalance in the national wealth and livelihood. On the other hand, local governments' insufficient input in social public health services or high-input-low-output (HILO) will also affect its supply efficiency. At present, many scholars have empirically analyzed that local governments have a policy bias toward ignoring public service expenditures in the process of fiscal expenditure [1,2]. However, there is little researche on the HILO problem. With the constraint of the relative lack of public service input in China [3], we need to determine how to improve the efficiency of local government's public service supply, in order to improve the output effect with an equal input, since this plays a decisive role in improving the efficiency of the Chinese public service. In particular, with the increase of local government social public health service input, a slight change in supply efficiency will have a great influence on the achievement of the set goals [4].

In view of this, this paper intends to use the mainstream methods of health economic supply theory and empirical research to analyze the input and output of the government's social public health service. Then, we measure the government's health expenditure efficiency in different regions of China, based on the spatial correlated test and spatial panel measurement model. According to the results of the empirical analysis, some suggestions are proposed to help design the government's new health input mechanism, which can accelerate the construction of a healthy China, improve the national health policy, and deepen the reform of the medical and health system.

#### **2. Literature Review**

Since the data envelopment analysis (DEA) method was proposed by Farrell in 1957 [5], and developed by Charnes, Cooper, and Rhodes (CCR) in 1978 [6], the quantitative evaluation theory and empirical research methods of local government social public health service supply efficiency have made great progress in the past 30 years. For example, Afonso and Fernandes (2006) used the DEA two-step analysis method to study the efficiency and influencing factors of local GHEs in Europe and Portugal [7]. Hadad et al. (2013) used the DEA estimation model to estimate the efficiency of the healthcare system and found that, in the Organization for Economic Co-operation and Development (OECD) countries, the number of stable healthcare systems is higher [8]. On the other hand, many Chinese scholars have also conducted some related research. Han and Miao (2010) used the DEA-Tobit two-stage analysis framework to measure the efficiency and influencing factors of GHEs in various provinces or cities in China [9]. Luo (2012) used the DEA-Bootstrap two-stage analysis framework to measure the efficiency of Chinese local fiscal expenditures and their influencing factors [10]. Guan et al. (2014) used the DEA four-stage analysis framework to measure the efficiency of the social public healthcare input in 30 provinces in China and the results show that the provincial social public health expenditure has an annual efficiency loss of 29.5% [11]. Jin and Song (2012) used the DEA and Malmquist productivity index to analyze the differences in government health expenditure efficiency among different regions in China [12]. Thus, in terms of evaluation methods, DEA solves the difficulty by integrating the efficiency of different project units and multiple-input-multiple-output (MIMO), so that it is easy to compare the efficiency values of different projects.

There is no typical academic and practical method to simply measure the supply efficiency of the government public health service. The most important thing to do is explore how to improve the efficiency of government health expenditure. It is found from the existing literature that scholars mainly pay attention to the evaluation of social public health expenditure efficiency, ignoring the research on the efficiency of regional public health expenditure. However, the level of economic and social development among different regions is obviously different. We need to consider the regional disparities in the calculation of the expenditure efficiency of different regions. In 2009, as a Chinese policy goal, the improvement of the utilization of health resources in the new healthcare reform was clearly proposed. Nowadays, with the increasing convenience of social transportation, the sharing of health resources among different regions is more convenient and the spatial nature of medical and health services is more obvious. This makes it more practical to study the efficiency of healthcare utilization and we can propose some policy suggestions to improve government supply efficiency based on this study.

Some scholars have studied the spatiality of government health expenditures and most studies show that the allocation of health resources in China is spatial. Wang et al. (2015) pointed out that the efficiency of health resources is a multi-dimensional and comprehensive problem closely related to factors, such as the economy, population, and regional development [13]. At the same time, the combination of traditional statistical methods and spatial statistical methods can better reveal the advantages and disadvantages of the health field, with a spatial distribution [13]. Han et al. (2016) believed that the total allocation of health resources in Chinese provinces or cities is basically consistent with the economic development of the regions [14]. Gu (2014) pointed out that when we study the allocation of health support areas, we should not only investigate the characteristics and needs of local areas, but also consider the influence of neighboring areas [15].

On the one hand, many scholars have conducted research on the output indicators in studying the efficiency of GHE. Retzlaff-Roberts (2004) [16] and Hadad et al. (2013) [8] selected life expectancy and infant mortality (survival) as output indicators. Liu et al. (2014) [17] selected the number of health institutions, health personnel and beds as intermediate output indicators. The number of outpatients and inpatients, emergency mortality, and observation room mortality were considered as final output indicators. Xiao al. (2014) [18] selected maternal mortality and infant mortality as output indicators.

On the other hand, scholars have conducted a lot of research on the factors affecting the efficiency of GHE. Most research starts from the economic, social, and political factors. For example, Tu (2012) pointed out that the factors that have a great impact on hospital efficiency include the fiscal decentralization, proportion of GHE, urbanization level, density of medical institutions, and medical technology level. Among them, only the fiscal decentralization, proportion of GHE, and medical technology level were considered to be the major factors affecting the efficiency of primary healthcare institutions [19]. Li and Wang (2015) conducted a study on the efficiency of local GHEs. The results showed that there were obvious regional differences in the efficiency of local GHEs during the sample period. The main factors affecting efficiency include the fiscal decentralization, household registration system, healthcare reform, urbanization level, economic development level, population density, and education level [20].

Among the existing research, many scholars have conducted fruitful research on the efficiency of GHE. Relevant research has a strong reference effect in the improvement of the supply efficiency of local governments' public health services in China. However, the characteristics of social public health services make it difficult to measure local governments' supply. Thus, it has not yet formed a comprehensive, consistent, and general evaluation system, methods, indicators, and research conclusions. In addition, the existing literature has the following limitations:

First, the discussion on the efficiency of GHE in the existing literature is relatively simple, if the low efficiency, caused by the uneven allocation of health resources, is not considered.

Secondly, due to the difficulty associated with accurately evaluating government health inputs and outputs, most scholars choose life expectancy and child mortality as output variables in studying the efficiency of GHE. However, health resources, such as the number of institutional beds and technicians, are the most direct outputs of GHE.

Finally, the spillover effect of health resource utilization efficiency has not been considered in the existing literature in studying the efficiency of GHE. Therefore, due to the special nature of public health services, there are still no comprehensive, consistent, general evaluation systems, methods, indicators, and research conclusions, though many scholars have conducted a lot of research on the efficiency of GHE.

#### **3. Methods and Materials**

#### *3.1. Study Settings and Potential Data Sources*

The research object of this paper was 31 provinces, municipalities directly under the central government and autonomous regions in China. The relevant indicator data of each region from 2007 to 2016 were obtained from the China Statistical Yearbook [21], China Health Statistical Yearbook [22] and China Financial Yearbook [23]. The geographical distance between provincial capitals and cities in different regions was drawn from Wikipedia.

#### *3.2. Variable Selection*

#### 3.2.1. Input Variable Selection

In this paper, the local GHE in 31 provinces (municipalities, autonomous regions) of China were selected as the sole input variable. Considering the comparability of the economy and data, Hong Kong, Macao, and Taiwan were not included in the calculation.

Figure 1 shows the local GHE in China from 2007 to 2016. From Figure 1, it can be seen that the local GHE in China had a rapid growth trend, from 199.58 billion Chinese yuan in 2007 to 1306.76 billion Chinese yuan in 2016, with a 650% increase. Among them, the largest absolute growth was in 2014, up 188.34 billion Chinese yuan from 2013. While the local government has increased health expenditure year by year, the problems in China, including it being difficult and expensive to see a doctor, are still serious. In addition, we often need to consider the spatial correlation when there is spatial distribution in the economic subjects, however, the traditional econometric models often neglect spatial correlation because of the irrelevance in the regional data. Spatial econometrics, as a branch of econometrics, incorporates the spatial weight matrix into the regression model, which is widely used in regional science, urban economics, geoeconomics, and development economics. Spatial econometrics studies how to deal with spatial heterogeneity (spatial structure) and spatial correlation (spatial interaction) in regression models. This paper will measure the efficiency of the local GHE and its spatial spillover effect. The selection of output and influencing factor variables are given as follows.

**Figure 1.** Chinese local government health expenditures from 2007 to 2016.

#### 3.2.2. Output and Influencing Factor Variables Selection

The basic goal of GHE is to achieve the optimal supply of public health services under the condition of limited health resources in order to obtain the maximum health output. From this point of view, we can divide the target of GHE into an intermediate target and a final target (see Figure 2).

**Figure 2.** Government health expenditure (GHE) target.

Therefore, according to the final goal of the local GHE, this paper chooses the following seven indicators to show the inputs and outputs of local governments in the field of health services: (1) Severe malnutrition rate among children under five years old and the mortality rate, reflecting the health level of residents; (2) the number of outpatient visits and the medical expenses per capita of outpatients and inpatients, reflecting public health services; (3) infectious disease incidence rate of Class A and B statutory reports and infectious disease mortality rate, reflecting the residents' diseases control; and (4) hospital bed utilization rate, reflecting the quality of healthcare services. The output of local government public health services from 2007 to 2016 is shown in Table 1.



Based on the existing research results, this paper will consider the urban and rural, cultural, economic, and demographic factors, which may have an important impact on the spatial spillover effect of the local GHE efficiency. Therefore, the proportion of urban population, illiteracy rate, the GHE accounting for the proportion of fiscal expenditure, gross domestic product (GDP) per capita, fiscal decentralization, and population density are selected as explanatory variables, which are shown in detail in Table 2.


#### *3.3. Method*

#### 3.3.1. DEA Method

The DEA method, occasionally called frontier analysis, was developed by Charnes, Cooper, and Rhodes (CCR) in 1978 [6]. DEA models are classified into the model of Banker, Charnes, and Cooper (BCC) [24] and the CCR model. The BCC model is an output-oriented model and the CCR model is an input-oriented model. This paper chooses the output-oriented BCC model to study the efficiency of the local GHE. The specific model is shown as follows.

Assuming that the DEA model has *J* decision-making units (DMUs), DMU*j*(*j* = 1, 2, ··· , *J*), each DMU has *M* input items, *Xj* = (*x*1*j*, *x*2*j*, *x*3*j*, ··· *xMj*), and *N* output items, *Yj* = (*y*1*j*, *y*2*j*, *y*3*j*, ··· *yNj*). Thus, the overall efficiency of *j*0-th DMU can be obtained from the following linear programming.

$$\begin{aligned} \max \quad & \eta\_{j0} = \frac{\sum\_{n=1}^{N} a\_n y\_{nj\_0}}{\sum\_{m=1}^{M} \beta\_m x\_{mj\_0}} \\ \text{s.t.} & \begin{cases} \sum\_{n=1}^{N} a\_n y\_{nj} \\ \frac{n-1}{M} \sum\_{m \ge n\_{\text{inj}}} \beta\_m x\_{mj} \\ a\_n \ge 0 & (n = 1, 2, \cdots, M), \beta\_m \ge 0 (m = 1, 2, \cdots, M) \end{cases} \end{aligned} \tag{1}$$

where η*j*<sup>0</sup> is the desired overall efficiency and α*<sup>n</sup>* and β*<sup>m</sup>* are the weights for the outputs and inputs, respectively.

The purpose of this paper is to study the input of local governments in the field of public health services and to maximize the output of medical and health services with a certain amount of scale of health resources. Thus, this paper chooses the BCC model to calculate the efficiency of the local GHE in China. Furthermore, with the increasing convenience of social transportation and the more convenient sharing of health resources among different regions, the spatial spillover effect of public health services in different regions will be more obvious. In view of this, we continue to explore the temporal and spatial correlation and evolution trend of the local GHE efficiency in order to obtain a deeper understanding of the efficiency of the local GHE in China, which has great significance for optimizing the results of financial expenditure and improving the GHE efficiency.

#### 3.3.2. Spatial Econometric Model Method

#### (1) Morland Index

The First Law of Geography asserts that everything is related and the closer the things are, the higher the degree of correlation will be [25]. We record the geospatial data of *n* regions as *xi*, *xj n i*=1,*j*=1 , where *i* and *j* represent region *i* and region *j*, respectively. The distance between region *i* and region *j* is recorded as *wij*, which can be defined as a spatial weight matrix as follows.

$$\mathcal{W} = \begin{pmatrix} w\_{11} & \cdots & w\_{1n} \\ \vdots & \vdots & \vdots \\ w\_{n1} & \cdots & w\_{nn} \end{pmatrix} \tag{2}$$

where the elements on the principal diagonal are equal to 0, i.e., *w*<sup>11</sup> = ··· = *wnn* = 0 (the distance of the same region is expressed as 0). It should be noted that the commonly used spatial weight matrix is the spatial adjacent weight matrix, which can be expressed as follows:

$$w\_{ij} = \begin{cases} 1, & \text{regions } i \text{ and } j \text{ are neighbors} \\ 0, & \text{regions } i \text{ and } j \text{ are not neighbors} \end{cases} \tag{3}$$

As we know, the spatial measurement method can be used when there is spatial correlation among the data. If there is no spatial correlation, the general measurement method is used. "Spatial autocorrelation" means that regions have similar values of variables to other similar regions. Spatial autocorrelation can be divided into positive spatial correlation and negative spatial correlation. The most widely used method to measure spatial correlation is the Moran index, which is given as follows.

$$I = \frac{\sum\_{i=1}^{n} \sum\_{j=1}^{n} w\_{ij} (\mathbf{x}\_i - \overline{\mathbf{x}}) (\mathbf{x}\_j - \overline{\mathbf{x}})}{s^2 \sum\_{i=1}^{n} \sum\_{j=1}^{n} w\_{ij}},\tag{4}$$

where *s*<sup>2</sup> = *n i*=1 (*xi*−*x*) 2 *<sup>n</sup>* is the sample variance and *n i*=1 *n j*=1 *wij* is the sum of all spatial weight matrixes.

The value range of the Moran index is [−1,1]. If the Moran index *I* is greater than 0, there is a positive spatial correlation among different regions. If *I* is less than 0, there is a negative spatial correlation. If *I* is close to 0, the correlation among different regions is weak. The Moran index can be decomposed into the global Moran index and the local Moran index. The global Moran index represents the overall correlation, while the local Moran index decomposes the Moran index of each region in a certain year, which indicates a clustering phenomenon in this region. The Moran index can be regarded as the correlation coefficient between the observed value and its spatial lag.

The global Moran index is decomposed to obtain the Moran index value of each sample individual. The spatial dependence of each sample individual and its adjacent individual can be judged by plotting the value of each sample individual and its spatial adjacent individual as a scatter plot. Generally, four quadrants are obtained using scatter plots of the local Moran index. Each quadrant corresponds to a spatial structure, representing a set of special relationships between the value of individual variables and the mean value of adjacent individual variables, and each sample individual is grouped into a quadrant. Specifically, the value of individual variables in the first quadrant is high (H) and that of adjacent individuals is also high (H) and is expressed as HH, which is a common spatial expression pattern of a high-high cluster. The value of individual variables in the third quadrant is low (L), and the value of variables in adjacent individuals is low (L) and expressed by LL, which is a typical low-low cluster. The number of individuals in the second quadrant and the fourth quadrant is relatively small, where the value of individual variables is low, with high-value neighboring individual variables, and the value of individual variables is high, with low-value neighboring individual variables; this can be expressed by the LH and HL clusters, respectively. Furthermore, positive spatial correlation refers to the high-high or low-low cluster, while negative spatial correlation refers to the high-low or low-high cluster. There is no spatial correlation among the HL or LH cluster regions with random distribution. In this way, we can find the identity of the population and economic clusters in China by identifying the Chinese provinces using the HH and LL characteristics and the local Moran index.

(2) Spatial Econometric Model

➢ Model Selection

First, the spatial autoregressive model of the panel should be examined. The specific formula is as follows.

$$y\_{il} = \rho w\_{i}' y\_{l} + \mathbf{x}\_{il}' \boldsymbol{\beta} + \mu\_{l} + \varepsilon\_{il\prime} (i = 1, \cdots, n; t = 1, \cdots, T), \tag{5}$$

where *w <sup>i</sup>* is the *i*-th row of the spatial weight matrix *W*, *w i yi* = *n <sup>j</sup>*=<sup>1</sup> *wijyjt*, *wij* is the (*i*, *j*)-th element of the spatial weight matrix *W*, and μ*<sup>i</sup>* is the individual effect of region *i*. If μ*<sup>i</sup>* is related to *xit*, it is a fixed effects model; otherwise, it is a random effects model. The usual Hausman test can be used to determine whether a fixed-effect or a random-effect model should be used and it is shown as follows.

$$\begin{cases} y\_{it} = \pi y\_{i\downarrow -1} + \rho u\_j' y\_t + x\_{il}\beta + d\_i' X\_l \delta + u\_i + \gamma\_l + \varepsilon\_{il} \\ \varepsilon\_{il} = \lambda m\_j' \varepsilon\_l + v\_{il} \end{cases} \tag{6}$$

where *yi*,*t*−<sup>1</sup> is the first-order lag of the interpreted variable *yit*, *<sup>d</sup><sup>i</sup> i Xt*δ is the spatial lag of the explanatory variable, *d<sup>i</sup> <sup>t</sup>* is the *i*-th row of the corresponding spatial weight matrix *D*, γ*<sup>t</sup>* is the time effect, and *m i* is the *i*-th row of the disturbance item space weight matrix *M*. The following items explain how to distinguish the spatial autoregressive models [26].


#### ➢ Decomposition Mechanism of the Special Spillover Effect

The spatial Durbin model can decompose the explanatory variable-spillover effect using the partial differential method proposed by LeSage and Pace in 2009 [27], considering the total, direct, and indirect effects. Among them, the direct effect is the influence on the local region caused by the explanatory variables, while the indirect effect is the influence on the result of the explanatory variables in the neighboring regions. The specific calculation method is given as follows.

$$\mathcal{Y} = \left(I\_{\rm nl} - \ell \mathcal{W}\right)^{-1} a l\_{\rm nl} + \left(I\_{\rm nl} - \ell \mathcal{W}\right)^{-1} \left(X\_{l} \beta + \mathcal{W} X\_{l} \theta\right) a l\_{\rm nl} + \left(I\_{\rm nl} - \ell \mathcal{W}\right)^{-1} \varepsilon,\tag{7}$$

which can be rewritten as

$$Y = \sum\_{r=1}^{k} S\_r(\mathcal{W}) \mathbf{x}\_r + V(\mathcal{W}) l\_n \alpha + V(\mathcal{W}) \varepsilon\_r \tag{8}$$

where we have *Sr*(*W*) = *V*(*W*)(*In*β*<sup>r</sup>* + *W*θ*r*), *V*(*W*)=(*In* − *W*) <sup>−</sup><sup>1</sup> and *In* is an *n*-order identity matrix. In addition, the above equation can be converted into a matrix form, shown as

$$
\begin{bmatrix} y\_1 \\ y\_2 \\ \vdots \\ y\_n \end{bmatrix} = \sum\_{r=1}^k \begin{bmatrix} S\_r(\mathcal{W})\_{11} & S\_r(\mathcal{W})\_{12} & \cdots & S\_r(\mathcal{W})\_{1n} \\ S\_r(\mathcal{W})\_{21} & S\_r(\mathcal{W})\_{22} & \cdots & S\_r(\mathcal{W})\_{2n} \\ \vdots & \vdots & \ddots & \vdots \\ S\_r(\mathcal{W})\_{n1} & S\_r(\mathcal{W})\_{n1} & \cdots & S\_r(\mathcal{W})\_{nn} \end{bmatrix} \begin{bmatrix} \mathbf{x}\_{1r} \\ \mathbf{x}\_{2r} \\ \vdots \\ \mathbf{x}\_{nr} \end{bmatrix} + V(\mathcal{W})\_t \mathbf{a} + V(\mathcal{W}) \varepsilon\_r \tag{9}
$$

Thus, the total, direct, and indirect effects are obtained as

$$\overline{M}(r)\_{\rm ATI} = n^{-1} I\_{\hbar} \mathcal{S}\_{r}(\mathcal{W})\_{I\_{\hbar}r} \tag{10}$$

$$
\overline{M}(r)\_{\text{ADI}} = n^{-1} \text{trace} \mathbf{S}\_r(\mathcal{W}),
\tag{11}
$$

$$\overline{M}(r)\_{\text{AII}} = n^{-1} \{ I\_n \mathcal{S}\_r(\mathcal{W})\_{I\_n} - \text{trace} \mathcal{S}\_r(\mathcal{W}) \}. \tag{12}$$

#### **4. Results**

#### *4.1. E*ffi*ciency Analysis of Local Government Health Expenditure*

In this paper, we used the output-oriented BCC model and Deap 2.1 software to process the input and output data in order to obtain the efficiency of the local GHE from 2007 to 2016. The results are shown in Figure 3.

**Figure 3.** Overall efficiency of Chinese local government health expenditures from 2007 to 2016.

Figure 3 shows the changing trend of the static overall GHE efficiency of 31 regions from 2007 to 2016. During these 10 years, it can be seen that the overall efficiency increased from 0.62 to 0.73.

As shown in Figure 4, we used the average overall efficiency of the local GHE from 2007 to 2016 as a representative to analyze the differences in the total efficiencies among different regions. From Figure 4, we can see that the lowest overall efficiency is in Guizhou and the highest is in Ningxia. Among them, the overall efficiencies of 22 regions are under 0.8, for example, in Guizhou, Inner Mongolia, and Heilongjiang. There are nine regions whose overall efficiencies are greater than 0.9. Furthermore, it is not difficult to see that the overall efficiencies of the eastern region are generally higher, while those of the northeast and central regions are generally lower.

**Figure 4.** Ranking of the average overall efficiency of Chinese local government health expenditures from 2007 to 2016.

#### *4.2. Spatial Spillover E*ff*ect Analysis of Local Government Health Expenditure*

#### 4.2.1. Global Moran Index Calculation

The efficiency of GHE is a multi-dimensional and comprehensive problem closely related to factors such as the economy, culture, population, and regional development. Moreover, factors such as economic development, population cluster, and other factors in neighboring regions may also have an impact on the efficiency of the local GHE. Therefore, theoretically speaking, the efficiency of the local GHE has spatial correlation. In order to verify this hypothesis, this paper used the Moran index.

In this paper, we used Stata software to test the global spatial autocorrelation of the static overall efficiency of the local GHE from 2007 to 2016. The results are shown in the following Table 3.



Table 3 shows that the Moran's I value of the static overall efficiency of the local GHE from 2007 to 2016 is positive. Moreover, from the test of the 5% significance level, the efficiency of the local GHE is shown to have a positive spatial correlation in the past 10 years, that is, the efficiency of local GHE is spatial agglomeration. At the same time, we can see from the changing trend of the Moran's I value that the spatial correlation degree of the Chinese local GHE became stronger and stronger from 2007 to 2014 and slightly declined from 2015 to 2016.

#### 4.2.2. Local Moran Index Calculation

The results of the global Moran index show that the efficiency of the local GHE has a positive spatial correlation from 2007 to 2016. However, it cannot reveal which regions are in high-cluster areas or low-cluster areas. In view of this, this paper chose the three years when China completed five key tasks of deepening the reform of medical and health systems from 2009 to 2011. We used Stata software to calculate the local Moran index of the local GHE efficiency, in order to further show whether there was a local spatial agglomeration phenomenon for the efficiency of the local GHE. In this paper, Arcmap 10.2 software was used to describe the distribution of Moran scatter plots. The results are shown in the following figures.

Figures 5–7 show that the positive spatial correlation regions are mainly concentrated in the northeastern, eastern, and western regions of China. Among them, the low-low cluster areas are mainly concentrated in Inner Mongolia, Liaoning, Jilin, and Heilongjiang in Northeastern China and Sichuan, Yunnan, Chongqing, Guizhou, Guangxi, etc. in Western China, belong to the low GHE efficiency regions. The high-high cluster areas, mainly concentrated in Beijing, Jiangsu, Shanghai, Zhejiang, etc. in Eastern China and Tibet and Qinghai in Western China, belong to the high GHE efficiency regions. Henan had a negative spatial correlation agglomeration in the past three years, showing high-low clustering phenomenon, which indicates that the GHE efficiency of its neighboring region is very low.

**Figure 5.** The spatial agglomeration of Chinese local government health expenditures overall efficiency in 2009.

**Figure 6.** The spatial agglomeration of Chinese local government health expenditures overall efficiency in 2010.

**Figure 7.** The spatial agglomeration of Chinese local government health expenditures overall efficiency in 2011.

#### 4.2.3. Spatial Econometric Model Analysis

After verifying the spatial correlation of Chinese local GHE efficiency, we built a spatial panel econometric model to further analyze the impact of the proportion of urban population, illiteracy rate, the proportion of government health expenditure to fiscal expenditure, GDP per capita, and population density on the efficiency of the local GHE.

In this paper, the idea in [28] was used as a reference in choosing a spatial model, that is, an SDM model without any constraints was firstly estimated and then whether the SDM model can be simplified was tested. According to Equation (6), the SDM model estimation was first selected, then we processed the data using Stata software. Finally, Wald and likelihood ratio (LR) tests [29] were carried out to verify whether the SAR model and SEM model were nested in the SDM model. The results are shown in Table 4.



Note: \*, \*\*, \*\*\* indicate significant levels at 10%, 5%, and 1%, respectively.

From Table 4, the results of Wald and LR tests show that the hypothesis of τ = 0 and δ = 0 or τ = ρ = 0 and δ = 0 is negated. The SAC model and SDM model were further tested to see which was more suitable. The results showed that the absolute values of the Akaike information criterion (AIC) and Bayesian information criterion (BIC) of the SDM model were less than those of the SAC model, so the SDM model was more suitable for this paper. The Hausman test of the SDM model had a *p*-value of 0.000 < 0.05, which indicated that the random effect model hypothesis was rejected at the 5% significance level. Therefore, the fixed effect model of SDM was more suitable.

On the other hand, we can see that the spatial lag coefficient (rho) of explanatory variables is not significantly zero according to the estimation results of SDM. LeSage and Pace (2009) proposed that there will be systematic errors in measuring the spillover effect using the coefficient of SDM when the spatial lag term coefficient of the explanatory variables is not significantly zero [27]. In this case, the spatial effect of SDM should be further decomposed to accurately reflect the direction and extent of the impact of each explanatory variable on the GHE efficiency in the local and neighboring regions [27]. In view of this, this paper decomposed the SDM to obtain the total, direct, and indirect effects of each explanatory variable, as shown in Table 5.


**Table 5.** Spatial spillover effect decomposition with SDM Model.

Note: \*, \*\*, \*\*\* indicate significant levels at 10%, 5%, and 1%, respectively.

From Table 5, we can see that the total, direct, and indirect effects of the proportion of the urban population are significantly negative among the factors affecting the local GHE efficiency. It is also shown that, with the increase of the proportion of the urban population, both the local and neighboring GHE efficiencies decrease. The reason is that, with the increase of the urbanization proportion, people's healthcare needs increase correspondingly. However, with the increasingly convenient transportation and health services, people choose much more developed regions or countries for their healthcare service, which leads to a decline in the GHE efficiency.

The total, direct, and indirect effects of the illiteracy rate are all negative, but the direct and indirect effects have not passed the significant level test. This shows that the increase of the illiteracy rate will lead to a decrease of the local GHE efficiency, but the influence of spatial decomposition effect on the local and neighboring regions is very small. This may be due to the lower cultural quality of residents, because they know less about the transmission and prevention of some common diseases and are more likely to get sick. Moreover, the lower the education level of residents, the more difficult it will be to communicate with hospitals, ask for medical and health services, choose the right medical institutions, and obtain accurate healthcare information when they are sick. In addition, it will have some influence on people's awareness of the need to cooperate with the government's healthcare supervision, which will also decrease GHE efficiency.

The total and direct effects of the proportion of GHE to fiscal expenditure are significantly positive, while the indirect effects are negative and do not pass the significant level test. This shows that the increase of the proportion of GHE to fiscal expenditure will have a positive impact on the local GHE efficiency. The greater local governments' input in the field of healthcare, the higher the GHE efficiency will be. The reason is that, with the increase of local government input in the field of healthcare, it will be possible to provide better health services and thus improve the GHE efficiency.

The total, direct, and indirect effects of GDP per capita are all significantly positive, which indicates that the improvement of GDP per capita will improve the local GHE efficiency. This influence is reflected not only in improving the local GHE efficiency, but also the significant promotion effect on neighboring regions. This can be explained by the fact that residents in rich regions have a higher demand for high-quality health services, which may put more pressure on local governments to improve their health expenditure efficiency by increasing the health service level.

The total effect of fiscal decentralization is negative and has passed the significance level test, while the direct and indirect effects are negative but not significant. This indicates that the greater the degree of fiscal decentralization, the less the local GHE efficiency will be improved. This may be due to government competition caused by fiscal decentralization, which may lead to an inadequate provision of public goods by local governments and the reduction of the GHE scale, decreasing the efficiency of health expenditure.

The total, direct, and indirect effects of the population density are all significantly positive. The higher the population density, the higher the GHE efficiency. We believe that a higher population density can reduce the cost of government management and supervision and thus help to improve the GHE efficiency.

#### **5. Discussion**

In this paper, after using the DEA, spatial Moran's I value, and SDM model to measure the GHE efficiency and its spatial spillover effect from 2007 to 2016, we found that there are obvious differences in the local GHE efficiency due to the different levels of economic development and the high or low foundation of medical and health industry among different regions of China. We also found that the GHE efficiency in China increased year by year from 2007 to 2014, decreased slightly in 2015, and continued to increase in 2016. Finally, we found that the GHE efficiency has a significant positive spatial spillover. The proportion of GHE to fiscal expenditure, GDP per capita, and population density have a positive impact on the GHE efficiency. Their growth will promote the GHE efficiency in the local region and adjacent areas. Conversely, the proportion of urban population, illiteracy rate, and fiscal decentralization have a negative impact on the GHE efficiency. Their growth will reduce the efficiency in the local region and adjacent areas.

The strength of this study is to consider the spatial spillover effect in terms with realistic meaning. The results obtained can be used as a reference for optimizing the structure and improving the efficiency of government health inputs. It breaks the government's GDP-only theory-based assessment system and helps to improve it by assessing the GHE efficiency. On the other hand, our study makes a significant contribution to the literature on public health services with spatial spillover effect.

Based on the obtained results and analysis, we have some suggestions and policy recommendations, which are as follows:

Develop a reasonable health insurance system. In the current context of aging and urbanization, China should formulate a scientific and reasonable healthcare insurance system, which can not only minimize the waste of health resources, but also improve the efficiency of the local GHE.

Improve the fiscal decentralization system. China should strengthen the supervision and management of financial expenditure in various regions, so as to improve the efficiency of the local GHE.

Reduce the regional unbalance. From the calculation and analysis of the local Moran index of the GHE efficiency, the GHE efficiency of Beijing, Jiangsu, Shanghai, and Zhejiang in the eastern region and Tibet and Qinghai in the western region is high, showing a high-high cluster. Conversely, the GHE efficiency of Inner Mongolia, Liaoning, Jilin, and Heilongjiang in the northeastern region and Sichuan, Yunnan, Chongqing, Guizhou, and Guangxi in the western region is generally low, showing a low-low cluster. Henan Province in the middle region shows negative spatial correlation, which is a high-low cluster area. The central part is located in the east–west junction grounding zone, which is driven, to a certain extent, by the developed areas in the east, but the development is still lagging behind. Therefore, the overall economic development pattern of China is not coordinated and is unbalanced among different regions. Correspondingly, the GHE efficiency is also unbalanced in terms of regional development. In order to maximize the GHE efficiency, we should vigorously implement the strategy of regional coordinated development, strengthen the health construction in the central and western regions, improve the level of health efficiency, and induce regional spillover of the GHE efficiency.

Strengthen the health cooperation. It is very important to strengthen the overall coordination among different regions and cross-administrative health cooperation. We should standardize inter-regional health rules and policies, improve the formulation and implementation of inter-regional environmental policies, reduce vicious competition in health expenditure efficiency and avoid blind competition and excessive competition. For the eastern region, which is in a high-high cluster area, regional exchanges and cooperation can be promoted to achieve coordinated development and common progress among regions. In order to reduce the efficiency of health expenditure and improve the "free-rider" behavior, it is necessary to establish standardized and unified health regulations and standards for the western and northeastern regions in low-low cluster areas.

Utilize the spatial spillover effect. Chinese GHE has a significant spatial spillover effect. In order to promote the coordinated development of the regional economy, the government should guide the flow of health resources in the region through fiscal and tax policies and optimize the layout of regional health centers. On the other hand, the government can fully utilize the spatial spillover effect of the health expenditure efficiency through the rational layout of regional health centers in order to promote regional development. In addition, we can use the spatial spillover effect to break down administrative barriers. Local government health services should break down the administrative barriers between regions, to make it convenient for people to enjoy the healthcare services in different places. At the same time, we need to strengthen the healthcare exchanges and cooperation with neighboring regions and make full use of regional spillover effect mechanism in order to better realize the effective supply of local basic healthcare services.

Optimize the urban population structure. China should rationally optimize the urban population structure, promote population aggregation, and play its role in improving the efficiency of health expenditure in local region and adjacent areas. Moreover, we need to accelerate the balanced development of the regional economy. From the analysis results concerning the spatial measurement of the health expenditure efficiency in China, the impact of Chinese economic development on the efficiency of health expenditure is obviously unbalanced. The growth of economics and the proportion of GHE to financial expenditure have promoted the efficiency of health expenditure in local regions and adjacent areas, while fiscal decentralization has inhibited it. The government can further adjust the proportion of GHE to financial expenditure to maintain economic growth and promote health expenditure efficiency. In order to comprehensively promote the construction of social education and improve people's cultural quality, the government needs to further promulgate policies to promote the development of Chinese education. The efficiency of health expenditure can be significantly improved if the illiteracy rate is reduced.

#### **6. Conclusions**

In this paper, the DEA and spatial regression models are used to measure the local GHE efficiency and its spatial spillover effect in 31 provinces (municipalities and autonomous regions) of China from 2007 to 2016. The conclusions are summarized as follows.

Due to the different levels of economic development and the high or low foundation of the medical and health industry in various regions of China, there are obvious differences in the local GHE efficiency.

As far as individual regions are concerned, the overall efficiency of local GHE in Shandong, Zhejiang, Guangdong, Shanghai, Qinghai, Tibet, Hainan, Tianjin, and Ningxia is higher than 0.8. However, the overall efficiency of other regions is generally not high, which indicates that there is a large waste of GHE in these regions.

The results of the global Moran index calculations show that there is a positive spatial correlation between the local GHE efficiency from 2007 to 2016, that is, the efficiency of the local GHE is a spatial agglomeration. At the same time, we can also see from the trend of the Moran index that the spatial correlation degree of the local GHE from 2007 to 2014 shows an upward trend, while the spatial correlation degree between 2015 and 2016 slightly declines. At the same time, from 2007 to 2016, the proportion of the urban population, the illiteracy rate, the proportion of GHE to fiscal expenditure, the GDP per capita, fiscal decentralization, and population density are the main factors affecting the efficiency of the local GHE and they are all spatially correlated.

The calculation results of the local Moran index show that the positive spatial correlation regions are mainly concentrated in the northeast, east, and west of China. Among them, the low-low cluster areas are mainly concentrated in Inner Mongolia, Liaoning, Jilin, and Heilongjiang in the northeast region and Sichuan, Yunnan, Chongqing, Guizhou, Guangxi, etc. in the western region. High-high cluster areas are mainly concentrated in Beijing, Jiangsu, Shanghai, Zhejiang, etc. in the eastern region and Tibet and Qinghai in the western region. Henan is a spatial negative correlation cluster region, showing a high-low clustering phenomenon.

The results of the spatial econometric model in this paper show that, from the perspective of total, direct, and indirect effects, the proportion of GHE to fiscal expenditure and GDP per capita are two important indicators representing economic factors and the population density represents the population factors. They have a positive impact on the health and financial expenditure efficiency of local government and their growth will improve the efficiency of the local and neighboring regions, while the three indicators of the proportion of urban population, illiteracy rate, and fiscal decentralization have a negative impact on the health and financial expenditure efficiency of local governments and their growth will reduce the efficiency of local and neighboring regions.

**Author Contributions:** M.W. and C.T. conceived and designed the study; M.W. collected and analyzed the data and drafted the paper; M.W. and C.T. read and revised the draft critically. All authors read and approved the final manuscript.

**Funding:** This research is funded by the National Social Science Foundation of China (Grant No. 16BTJ004; 14BJY160), the Bidding Project of Cooperative Innovation Center of Jiangxi University of Finance and Economics, "Research on Development of Biomedical Industry and Policy Support" (Grant No. 2016-03), Special Project of the National Social Science Foundation of China (Grant No. 18VSJ016), and the National Natural Science Foundation of China (Grant No. 81760619).

**Acknowledgments:** We would like to thank Lisu Yu and Xiaohui Liu for their help to improve the paper.

**Conflicts of Interest:** The authors declare no conflict of interest.

#### **References**


© 2019 by the authors. Licensee MDPI, Basel, Switzerland. This article is an open access article distributed under the terms and conditions of the Creative Commons Attribution (CC BY) license (http://creativecommons.org/licenses/by/4.0/).

## *Article* **Depression, Acculturative Stress, and Social Connectedness among International University Students in Japan: A Statistical Investigation**

**Minh Hoang Nguyen 1, Tam Tri Le <sup>1</sup> and Serik Meirmanov 2,\***


Received: 23 December 2018; Accepted: 3 February 2019; Published: 8 February 2019

**Abstract:** (1) This study aims to examine the prevalence of depression and its correlation with Acculturative Stress and Social Connectedness among domestic and international students in an international university in Japan. (2) Methods: A Web-based survey was distributed among several classes of students of the university, which yielded 268 responses. On the survey, a nine-item tool from the Patient Health Questionnaire (PHQ-9), the Social Connectedness Scale (SCS) and Acculturative Stress Scale for International Students (ASSIS) were used together with socio-demographic data. (3) Results: The prevalence of depression was higher among international than domestic students (37.81% and 29.85%, respectively). English language proficiency and student age (20 years old) showed a significant correlation with depression among domestic students (β = −1.63, *p* = 0.038 and β = 2.24, *p* = 0.048). Stay length (third year) also displayed a significant correlation with depression among international students (β = 1.08, *p* = 0.032). Among international and domestic students, a statistically significant positive correlation between depression and acculturative stress, and negative associations of social connectedness with depression and acculturative stress were also found. (4) Conclusions: The high prevalence of depression, and its association with Acculturation stress and Social Connectedness, among the students in this study highlight the importance of implementing support programs which consider the role of Acculturation and Social Connectedness.

**Keywords:** depression; acculturation stress; social connectedness; international students; university students; ASSIS; Mindsponge; multicultural

#### **1. Introduction**

The rapid spread of globalization has made countries worldwide increasingly interconnected in many aspects, including in education. The number of globally mobile students increased by 25.3% from 2012 to 2017 [1]. Understanding the mental health condition among international students will help sustain the development of the social public health system, especially in countries with a high number of international students.

Depressive disorder is a major public health concern which affects 322 million people globally. In 2015, global depression prevalence was 4.4%. Together with HIV/AIDS and heart disease, depression was projected to be one of the three leading causes of burden of disease until 2030 [2]. If becoming too severe, depression can lead to suicide—the second leading cause of death for people aged 15–29 [3].

High depression prevalence (4.2%) is a serious problem to the sustainable public health development in Japan [3]. In 2008, depression entailed a substantial burden to the Japanese economy by costing approximately \$11 billion [4]. Depressive disorder was also among the top causes of suicide in Japan [5,6].

Depression is more prevalent in university students compared to the general population [7], even in Japan. Major depressive disorder prevalence among first-year university students in Japan was reported at 20.7%, 53.4%, and 23.3% according to three different measurements [8]. Another paper found that around 30% of university students in Japan had depressive state [9]. As the number of international students in Japan increased rapidly by 45% in the last 3 years (from 184,155 in 2014 to 267,042 in 2017) [10], the mental health of international students has become a major concern. The prevalence of depression of international students in Japan was 41% [11]. A proportion of 34% of students visiting the mental health service at Tsukuba University health center informed being depressed [12].

Studies of factors contributing to depression in university students vary among countries regarding their findings [13–19]. For instance, a study in Kenya showed significant correlations between higher depression and year of study, academic performance, religion, age [14], while another study in Malaysia indicated significant associations of depression with age, gender, and economic status [15].

There were several studies on the predictors of depressive disorder in Japan. For instance, personality traits, such as self-directedness and harm avoidance, were significantly correlated with major depressive disorder in Japanese university students [20]. Negative automatic thought was found to be positively corelated with depression [21]. University life satisfaction and lifestyle (irregularity of meal and wake-up time) also contributed to depression among Japanese students. Meanwhile, gender, course category, and residential arrangement significantly correlated with depression among international students in Japan [11]. Overall, few studies about predictors of depression among domestic and international students in Japan have been conducted.

The results of cultural and psychological changes from moving to a foreign country or living in a new cultural environment are known as acculturation [22]. As international students start living in a new environment, intercultural contacts can cause acculturative stress, so they need to adapt to harmonize potential conflicts. Findings of many studies showed a significantly positive correlation between acculturative stress and depression [23–26]. Knowing more how acculturative stress is linked to depression in an international university will help improve the support systems for student's mental health. Findings of research about acculturation are mostly about acculturative stress, psychological adjustment, social belonging, depression, and anxiety [27]. While there are plenty of studies about acculturative stress in general immigrants [23,25,26] and international students in the US [28–30], the amount of studies on students in an international university in other country is still limited. University students have a worse quality of life [31]. The mental health condition of international university students needs special attention, especially considering the challenges they need to overcome when living in a new environment.

Social connectedness reflects an individual's opinion of themselves in relation to other people within a social context [32]. When the sense of connectedness of a person declines, that person starts to feel distant and different from other people and recall where it belongs [32]. Social connectedness is a predictor of depression among college students [33] and acculturative stress among international students [34]. International students who leave their home country to study in an entirely new environment become disconnected with their old relationships and connections. These losses might pose threat to student's mental health, such as depression. However, studies regarding the association between social connectedness and depression among students at an international university remain limited.

An international university is where students and faculties from different societies and cultural background are placed together. According to the Times Higher Education, the international outlook of a university is assessed based on three factors: the proportion of international students, the proportion of international staffs, and its international collaboration [35]. As a result, the greater the proportion of international students and faculties from various countries and regions is, the more multicultural the international university becomes.

It is worth nothing that studying in an international university is a double-edged experience. On the one hand, individuals being in a multicultural environment might receive many positive outcomes, for example, having a higher degree of cultural additivity [36] and behaving appropriately under multiple cultural contexts [37]. On the other hand, living in a multicultural environment could lead to problems of acculturation, such as feeling being rejected from multiple cultural contexts [38]. Not only international but even domestic students might experience difficulties in adjusting to the environment of international university. This is why it is important to understand the impact of acculturation and an individual's sense of connectedness on depression level in an international university.

To our knowledge, there was no comparative study concerning depression, acculturative stress, and social connectedness among domestic and international students in Japan. Since the literature about depression prevalence and the association of depression with acculturative stress and social connectedness in an international university is still limited, this study aims to examine: (1) the prevalence and predictors of depression and (2) the hypotheses of associations of depression with acculturative stress and social connectedness among domestic and international students at an international university. On the basis of the examined literature, the study seeks to answer the following research questions (RQ1 and RQ2) and hypotheses (H1 and H2) based on data collected from an international university in Japan:

**RQ1:** What is the prevalence of depression among domestic and international students?

**RQ2:** What are the socio-demographic predictors of depression among domestic and international students?

**H1: Acculturative stress will be significantly positively associated with depression in both domestic and international students.** Students in an international university were expected to have higher depression levels due to extensive conflicts during acculturation process.

**H2: Social connectedness will be significantly negatively associated with depression in both domestic and international students.** A greater sense of connectedness with others will make individual feel more comfortable and confident within a social context, which will prevent depression.

#### **2. Materials and Methods**

#### *2.1. Study Site*

In this study, we selected Ritsumeikan Asia Pacific University (APU), which is situated at Beppu City, Oita Prefecture in southern Japan, as our study area because it is Japan's first truly international university and is currently the most diverse university in Japan in terms of international faculty and student [39]. As of 2017, the number of international students and faculties on APU campus consisted of 50.1% of total 5,887 students and 49.4% of 166 faculty members, respectively [40]. The diversity of this university is not only limited to the proportion of students and faculties but also their variety of origins. According to official statistics of the university in 2018, international students came from 86 countries and regions, while the faculty members are from 22 different countries and regions [41]. The diversity of APU made an appropriate site for studying students in a multicultural environment—a subject found to be limited within the extant scholarship.

#### *2.2. Participants*

The study collected web-based questionnaires (Google Forms) of 268 students from a variety of countries who are currently studying at APU. A proportion of 75% of the number of participants were international students (N = 201), while domestic students accounted for 25% (N = 67). Among 201 international students, the highest percentages of students were from South East Asia (75%), including Vietnam, Indonesia, Thailand, and Malaysia. Students originating from East Asia, which consists of

China Korea, and Taiwan, accounted for 25%. Students from South Asia and other areas accounted for 9% and 5%, respectively.

Participants consisted of 170 females (63.4%) and 98 males (36.6%), with half of them being freshmen or who had been staying in APU within one year (see Table 1). As for language proficiency, around 76% of international students could speak English with high proficiency, whereas most of the native students acquired medium English proficiency and only 20% of them could use English fluently. The rate of international students able to speak the Japanese language fluently was quite low with only 12.6%, and almost half of them informed acquiring low Japanese proficiency. More than half of the participants, both domestic and international students, reported that they had no intimate partner at the time of filling the questionnaire (seven participants failed to report whether they had an intimate partner). The number of international students reported being religious was higher than that domestic students (37.31% compared to 23.88%).


**Table 1.** Socio-demographic characteristics of domestic and international students.

#### *2.3. Instruments*

#### 2.3.1. Measures of Depression

The PHQ-9, a nine-item tool from the Patient Health Questionnaire (PHQ), was used to measure Depression. The questionnaire consists of 9 questions based on the Diagnostic and Statistical Manual for Mental Disorders—4th edition (DSM-IV) criteria for diagnosis depression. With only nine questions, the PHQ-9 can be used for dual purposes, being the diagnosis of depressive disorder and grade depressive symptom severity [42]. By asking the participants about the frequency of various symptoms over the past two weeks, the study then categorized the respondents as having major depressive disorder or other depressive disorder. The respondents are diagnosed positive to major depressive

disorder or other depressive disorder if 5 or 2 depressive symptoms respectively present at least "more than half of the days" over the past two weeks, and one of the symptoms needs to be depressed mood or anhedonia [43]. Notably, the symptom "thoughts that you would be better off dead or of hurting yourself in some way" is counted regardless of the duration. To estimate the severity of depression, the DSM-IV criteria are scored as "0" (not at all) to "3" (nearly every day) in the PHQ-9, and thus the depressive severity score ranges from 0 to 27. The severity of depression is also categorized into five levels (minimal depression, mild depression, moderate depression, moderately severe depression, and severe depression) based on the score 1–4, 5–9, 10–14, 15–19, and 20–27 accordingly.

The validity of the PHQ-9 was tested for correlation with diagnosis by many mental health studies [44,45]. In the original study, Spitzer et al. acquired the validation of sensitivity and specificity at 73% and 98%, respectively, for significant depression among primary care patients [46]. The PHQ-9 was used to measure depression of not only patients but also a wide range of populations, including international students [44,45,47,48]. The Cronbach alpha measured in this study were 0.81 and 0.80 for international and domestic students respectively, which were acceptable.

#### 2.3.2. Measures of Social Connectedness

The measure of Social Connectedness in this research was the Social Connectedness Scale (SCS) developed by Lee and Robbins to evaluate an individual's emotional distance or connectedness between themselves and other people [32]. The Social Connectedness Scale consists of 8 items representing three aspects of belongingness: Connectedness (4 items), Affiliation (3 items), and Companionship (1 item). Each item is rated on a 6-point Likert scale ranging from 1 (Strongly Disagree) to 6 (Strongly Agree). A sample is "I feel disconnected from the world around me". The total score is the sum of 8 items, with higher scores indicating higher social connectedness. The potential score of Social Connectedness ranges from 6 to 48.

In the original study of Social Connectedness scale, the internal reliability or coefficient of Cronbach's alpha estimated was 0.91 [32], whereas alpha coefficients in other studies ranged from 0.83 to 0.93 [34,49–52]. The Alpha Coefficient in this study of international and domestic students was similar with 0.95. The construct validity was also supported by a negative association with anxiety, a positive association with self-esteem [53], and a negative correlation with acculturative stress [34].

#### 2.3.3. Measures of Acculturative Stress

This study measured acculturative stress using the Acculturative Stress Scale for International Students (ASSIS) developed by Sandhu and Asrabadi [54]. The ASSIS, a 36-item questionnaire about acculturative stress of international students, covers 7 major factors: Perceived Discrimination (8 items), Homesickness (4 items), Perceived Hatred (5 items), Fear (4 items), Culture Shock (3 items), Guilt (2 items), and Miscellaneous (10 items). Each item is scored on a 5-point Likert scale ranging from 1 (strongly disagree) to 5 (strongly agree). The sum of all 7 major factors represents the total score of acculturative stress. The higher the total score, the higher the degree of acculturative stress international students undergo. However, since this study was conducted in an English-speaking campus in a country where English is not the native language, the survey was modified suitably for the study area. Specifically, based on the item "I feel nervous to communicate in English", there was on additional item related to student's Japanese proficiency: "I feel nervous to communicate in Japanese." On the other hand, as the factor related to social connectedness was covered in the SCS, the item "I feel intimidated to participate in social activities" was omitted to keep the total score from 36 to 180.

In other research, the alpha coefficient for acculturative stress total score ranged from 0.92 to 0.95 for international students [28,29,34,50]. In this study, the alpha coefficient was 0.95 for both international students and domestic students, indicating a high average inter-item correlation. The construct validity of ASSIS was supported by the positive correlation with depression [28,29] and mental health [55], and the negative correlation with condom-use intentions [55].

#### *2.4. Procedure*

The questionnaire was approved by the Ethical Committee Board of APU. Google Forms was selected as the platform to conduct the questionnaire because it is common and easy to manage. After a brief explanation of the project, the link to the questionnaire was distributed in several classes and among the APU Vietnamese community from the end of 2018. We chose the Vietnamese community as two authors are originally from Vietnam. After the announcement, survey was available online and students had the chance to answer the survey at their convenience (campus, home, library, etc.). The total response rate was 40.05% (268/669).

Informed consent was obtained from each participant: after following the survey link, each participant read the informed consent text together with an explanation on the research's goals (based on APU regulations), below which the participants had the choice to participate or not by choosing "Agree" or "Not agree." The participants had an option to quit survey by choosing "Not agree" or any time later, just by not submitting the survey. In the case of agreeing to participate, each participant was shown the list of questions to which he/she needs to answer.

#### *2.5. Statistical Analysis*

The statistical analysis of the research consists of two main parts. The first part is for estimating the prevalence of depression and potential socio-demographic predictor of depression, whereas the second part is to test the association of two predictors (social connectedness and acculturative stress) with the severity of depression.

The correlations between depression and socio-demographic factors, such as, gender, age, length of stay, language proficiency, e.g., were examined to find out potential predictors of depression. People who suffered from major or other depressive disorder were considered as being depressed. Dichotomous logistic regression was used as a statistical analysis tool for testing the potential socio-demographic predictor, because of three main points: (1) predetermined number of variables was comprised in the model making it show each variable's significance more clearly; (2) the model can indicate all odd ratios simultaneously between dependent variable and other categories [56]; and (3) binary logistic regression was widely used in other studies with the same topic [45,57–59]. Thus, the examined model of being depressed as dependent variables was presented as follows:

$$\begin{array}{rcl} \ln\left(\frac{p}{1-p}\right)\_{\text{depression}} &=& \ln\left(\text{Odd Ratio}\right)\_{\text{depression}} \\ &=& a + \beta\_{1\text{j}}\text{gender}\_{1\text{j}} + \beta\_{2\text{j}}\text{age}\_{2\text{j}} + \beta\_{3\text{j}}\text{stay}\_{3\text{j}} + \beta\_{4\text{j}}\text{Eng}\_{4\text{j}} + \beta\_{5\text{j}}\text{Iap}\_{5\text{j}} \\ &+ & \beta\_{6\text{j}}\text{partner}\_{6\text{j}} + \beta\_{7\text{j}}\text{religion}\_{7\text{j}} + e\_{\text{}} \end{array}$$

with

*p*: the probability of being depressed *α*: intercept *β*: coefficient which is the logarithm of Odd Ratios *j*: categorical factor of independent variables *gender*, age, etc.: independent variables *e*: error term.

The binary logistic model was applied to both domestic and international students for comparative purpose. However, as no domestic students reported low Japanese proficiency and very few reported average Japanese proficiency, Japanese proficiency variable was omitted when the model was employed on the data set of domestic students.

Pearson Correlation Coefficient (PCC) analysis was first employed to look at the correlation among main variables used in the study for two reasons: (1) PCC indicates the strength of linear relationship of two variables [60]; and (2) PCC was widely used in studies in the same field [61–64]. According to the Cauchy-Schwartz inequality, PCC (r) has a value of between −1 and +1, where +1

indicates total positive linear correlation, 0 indicates no linear correlation, and −1 shows total negative linear correlation [65]. Next, to test the two hypotheses H1 and H2, multiple regression analysis was conducted to explore the contribution of main variables on predicting depression severity. Multiple regression analysis has been widely used in many other studies of the same field [23,25,66].

Raw data were cleaned in MS Excel and saved as CSV files. The data were then transferred into the database of STATA statistical software (version 15.1) to run statistical analysis. STATA statistical software was used for running dichotomous logistic regression, Pearson Correlation Coefficient, and multiple regression analysis. Additionally, robust statistics was also comprised in the dichotomous logistic regression model to omit outliners [67]. The *p*-value indicated the significance of independent variables in the models. It is conventional to choose *p*-value < 0.05 as a required statistical significance [68].

#### **3. Results**

#### *3.1. Descriptive Results*

According to the PHQ-9, 37.81% of international students and 29.85% of domestic students were found to be positive with depression (see Table 2). The proportion of depressed international students was higher than that of domestic students by almost 8%. While international students possessed a lower rate of major depressive disorder (14.93%) than other depressive disorder (22.89%), major depressive disorder (17.91%) was more significant than other depressive disorder (11.94%) among domestic students. Among the depressed domestic students, there was not much difference between males and females. On the other hand, international female students (39.84%) had a higher rate of depression than male international students (34.25%).


**Table 2.** Prevalence of depression.

Among international students, the percentage of students being depressed varied according to their regions of origin. International students who were originally from South East Asia had the highest percentage of depression (39.34%), while students from South Asia possessed the lowest rate of depression (27.78%), which was even lower than domestic students (29.85%). Students from other regions had relatively similar rates of depression at 38%. Similar to domestic students, students from South Asia suffered from major depression (16.67%) more than other depression (11.11%).

Of the 67 domestic students and 201 international students, more than half of them only reported minimal depression or mild depression. More domestic students (40.3%) indicated suffering moderate depression or higher, compared to international students (34.33%). A higher percentage of moderate and more severe depression was reported in domestic male students (44%) than in domestic female students (38.09%). On the contrary, the number of international female students (37.5%) undergoing moderate to severe depression exceeded the number of international male students (28.77%)

Based on the PHQ-9, ASSIS, and SCS, the average total scores of each concept was presented in Figure 1. As can be seen, domestic students (8.61) had higher average stress level compared to that of international students (8.04), but the difference was negligible. Domestic and international students shared a relatively same level of social connectedness. The result also indicated that acculturative stress was perceived more strongly among international students (75.56) than domestic students (62.88).

**Figure 1.** The total scores for depression, social connectedness, and acculturative stress among international and domestic students.

#### *3.2. Main Analysis*

#### 3.2.1. Binary Logistic Regression Analysis

The results of the dichotomous logistic regression with the dependent variable "depression" on the independent socio-demographic variables were displayed in Table 3. In Table 3, there are also the estimated coefficients and the *p*-value of each variable for both domestic students and international students (with the coefficient being the logarithm of Odd Ratio).

The results showed multiple socio-demographic characteristics predicting the depression in domestic students at *p*-value < 0.05, whereas only one potential predictor can be found in international students. Domestic students in the age of 20 were more likely to have depression than those with ages from 17 to 19 (β = 2.24, *p* = 0.048). Another predictor of depression among domestic students was English proficiency. Domestic students speaking average English had a lower rate of being depressed than those who acquired low English proficiency (β = −1.63, *p* = 0.038). On the other hand, the only predictor for international students was the length of stay. International students living in Japan for three years suffered from a higher risk of being depressed than first-year students (β = 1.08, *p* = 0.032).


**Table 3.** Coefficient (β) and *p*-value for predictors of depression.

Note: \* and \*\*\* are statistically significant at 0.05 and 0.001, respectively.

#### 3.2.2. Pearson Coefficient Correlation Analysis

In pairwise correlation analysis (see Tables 4 and 5), all relationships were found to be statistically significant at *p*-value < 0.001 for both international and domestic students. According to the finding, our hypotheses were confirmed that social connectedness significantly negatively correlated with depression in both domestic and international students (r = −0.6 and r = −0.54, respectively), and that acculturative stress significantly positively corresponded with depression in international students (r = 0.41). In other words, students with high social connectedness would be more likely to have higher depression levels, while international students with higher acculturative stress levels reported higher depression levels.

**Table 4.** Correlational relationship among Depression, Acculturative Stress, and Social Connectedness (Domestic students).


Note: \* and \*\*\* are statistically significant at 0.05 and 0.001, respectively.

**Table 5.** Correlational relationship among Depression, Acculturative Stress, and Social Connectedness (International students).


Note: \* and \*\*\* are statistically significant at 0.05 and 0.001, respectively.

The finding also indicated a significantly negative relationship between social connectedness and acculturative stress in international students (r = −0.58). International students suffering from high acculturative stress reported lower sense of connectedness with surroundings. Additionally, there were statistically significant correlations of acculturative stress with depression and social connectedness estimated among domestic students (r = 0.45 and r = −0.55, respectively). The decrease of acculturative stress corresponded with the decrease of depression among domestic students, whereas the growing feeling of being connected to others correlated with less stress occurring during acculturation.

#### 3.2.3. Multiple Regression Analysis

The normality of the dependent variable was first evaluated using the Skewness and Kurtosis normality test to make sure the variable was normally distributed. The Skewness and Kurtosis of international students were 0 and 0.06, respectively, indicating a normal distribution. On the other hand, those of domestic students were 0.14 and 0.41, indicating a mild nonnormality [28]. Therefore, we used square-root transformation for the dependent variable of domestic student [69]. After transformation, the Skewness and Kurtosis of domestic students were 0 and 0.15, indicating a normal distribution similar to those of international students. The transformed dependent variable of domestic students was used in the remaining of the analysis.

*Domestic students.* Regressing two main variables on transformed depression, results showed that the model accounted for 35.7% (R<sup>2</sup> = 0.357) of the variance (see Table 6). The *F* value of the model, *F* (2,64) = 17.77 was statistically significant at *p*-value < 0.001. Both of the main predictors had statistically significant association with depression. Social connectedness was negatively correlated with depression (β = −0.044, *p*-value < 0.01). This indicated that people with stronger feeling of connectedness were likely to have lower severity of depression. Meanwhile, acculturative stress had a positively significant relationship with level of depression (β = 0.014, *p*-value < 0.05). In other words, domestic students with higher acculturative stress would likely suffer from more severe depression.



Note: \* and \*\*\* are statistically significant at 0.05 and 0.001, respectively.

*International students.* The regression analysis of the international student sample showed similar results to the domestic student sample. The goodness of fit of the international sample (R<sup>2</sup> = 0.303) was smaller than that of domestic samples, while the F value was still significant at *p*-value < 0.001 with F (2,198) = 43.15. All two main variables significantly predicted depression level. International students feeling connected to surrounding people were more likely to have low depression level (β = 0.033, *p*-value < 0.001). Higher acculturative stress could also predict more serious depression in international students (β = 0.033, *p*-value < 0.05).

The results from both domestic and international students implied the impact of social connectedness and acculturative stress on depression. According to the correlational results from the PCC test, social connectedness and acculturative stress were negatively correlated, social connectedness and acculturative stress might have not only direct impacts but also indirect impacts on the level of depression. Social connectedness could possibly directly affect depression levels and indirectly increase depression levels through escalating acculturative stress, and vice versa for acculturative stress.

#### **4. Discussion and Conclusions**

By collecting questionnaires from both domestic and international students at APU, the current study presents a primary investigation on depression, social connectedness, and acculturative stress in a multicultural environment. In the cross-sectional questionnaire of 67 domestic students and 201 international students, the study found that 29.85% of domestic students and 37.81% of international students were positive to major depressive disorder or other depressive disorder. In addition, the findings also highlight several socio-demographic predictors (age, English proficiency, and length of stay) and main associations (social connectedness and acculturative stress) of depression. The Mindsponge concept [70] was used to explain the findings.

#### *4.1. Prevalence of Depression*

According to this study, 30% of domestic students at APU had depression. This figure is quite common among studies of depression prevalence of students in Japan. In a study using The Center for Epidemiologic Studies Depression Scale (CES-D), one third of the total 105 students reported having mild depression and higher [9]. Based on The Zung Self-Rating Depression Scale (SDS), a study in 2011 also showed 30% of 2197 of Japanese dental college students having symptoms of moderate or severe depression [71]. As can be seen, the depression rate of Japanese college students was around 30%, which was close to the prevalence of depression among university students (30.6%) reported in a systematic review of 11 countries [7].

The depression prevalence of international students in this study was 37.81%, while 34% of international students who visited a mental health service at Tsukuba University health care were found to be depressed [12]. Another study employing the Center for Epidemiologic Studies Depression Scales reported 41% of 480 international respondents to be depressed [11]. In all cases, the situations of international students were more serious than those of domestic students, which required policy makers, schools, or anyone in charge to pay more attention to the mental health of international students.

The fact that more international students suffer from depression than domestic students can be the consequence of multiple sources. First, international students at an international university received higher level of acculturative stress (see Figure 1), which elevated the risk of being depressed [28–30,55]. Second, international students have lower access to mental health support than domestic students [44,72]. Third, international students have fewer choices of help-seeking sources than their domestic counterparts. For example, it is difficult for international students to seek help from parents and relatives due to the geographical distance [73].

Compared to the depression of students in other countries, the depression prevalence at APU and in Japan, in general, was not high. For example, a study in India employing the University Student Depression Inventory (USDI) showed 53.2% of students being positive to depression [13]. Using Depression Anxiety Stress Scale-21 (DASS-21), 60.8% of Egyptian students, 37.2% of Malaysian students, and 33% American students reported being depressed [15,19,74]. A web-based questionnaire among 4330 students using BAPI depression scale showed that 28.25% of Turkish students exhibited symptoms of depression [17].

Different levels of prevalence among countries are attributable to several reasons. Different types of questionnaire were used to measure depression (PHQ-9, The Center for Epidemiologic Studies Depression Scale (CES-D) [9], The Zung Self-Rating Depression Scale (SDS) [71], the University Student Depression Inventory (USDI) [13], Depression Anxiety Stress Scale-21 (DASS-21) [15,19,74], and BAPI depression scale [17]). Moreover, the difference might also result from macro-scale, micro-scale, and personal-scale factors. Among the macro-scale factors, depression can be influenced by a socio-economic background in which the university is located, such as income inequality and cultures [75], while the micro-scale factors, which include the living arrangement on campus and academic environment, might play an essential role in driving depression [13]. Apart from that, personal activities, beliefs, and issues might also be crucial contributors to depression [13,75].

#### *Sustainability* **2019**, *11*, 878

Different levels of prevalence by gender were also noted. In both international and domestic students, a higher proportion of female students was found to be depressed than that of male. This difference might be explained by the fact that female students seem to have higher emotional, physiological, and behavioral reactions to stressors [76].

#### *4.2. Socio-Demographic Factors Associated with Depression*

In this study, socio-demographic data such as age, gender, length of stay in a new environment, and language proficiency, e.g., were selected to examine the association with depression using dichotomous logistic regression. Compared to another finding of the depression among international students [11], the results here confirmed that age and Japanese language proficiency were not potential predictors of depression among international students in Japan. On the other hand, studies of international students in the United States [77] revealed English proficiency as a predictor for depression. Along this note, the current study did confirm English proficiency as a predictor of depression among the Japanese students. A high percentage of foreign staffs and students, and mandatory English-based subjects for domestic students might be the answers.

Apart from English proficiency, the results also showed two other statistically significant predictors for depression among domestic students. First, age was a significant predictor for depression. Domestic students aged 20 reported having higher rate of being depressed than students with ages between 17 and 19. This finding was consistent with the findings in Kenya and Malaysia [14,15], while the fact that depression did not vary according to age was pointed out in another study in the U.S. [78]. Thus, it might depend on the cultural and social context of the study site to say whether age can be a predictor. In this case, Japanese teenagers were afraid of becoming adults [79], and when Japanese teenagers become 20, they will be considered as adults after the Coming of Age Day (*Seijin no Hi*). As a result, depression might be more likely to occur among 20-year-old Japanese students.

Another notable result in this study was that third-year international students had a higher chance of being depressed than first-year students. There are several possible explanations: (1) third-year students at APU need to take major courses which are more difficult than the introductory and basic courses in the first and second years; (2) the end of the third year is the period that students have to think about their career paths, and career indecision was positively correlated with depression [80,81].

Findings that age and length of stay were contributors to depression provided some insights of emerging adulthood theory [82] in an Asian country, a topic that has not been explored much. Some people in their late teens to their mid-to late 20s experience serious mental health problem as they do not want to take adult responsibilities and obligations [82,83]. Additionally, facing difficulties and lacking educational information when entering into the labor market might be very stressful for emerging adult students [84,85]. In general, depression related to the fear of being an adult or career indecision during the emerging adulthood period does not happen only in Western countries, but also in the Asian settings.

#### *4.3. Mindsponge Theory as an Explaination for Depression, Social Connectedness and Acculturation*

Findings of the current study confirmed the two main hypotheses that (H1) acculturative stress was significantly positively associated with depression, and (H2) social connectedness was significantly negatively correlated with depression among both domestic and international students. The Mindsponge concept [70] could serve as an explanation for the underlying mechanism among depression, social connectedness, and acculturation among students from a multicultural aspect.

Mindsponge is a concept designed to illustrate the mechanism of how a person absorbs and integrates new cultural values into a his/her own mindset and the reverse of ejecting inappropriate core values (see Figure 1 in the work of Vuong & Napier [70]).

*Acculturative stress.* This study found a positive correlation between acculturative stress and depression among domestic and international students. This judgement was supported by similar findings from other studies [23–26,28,29]. Acculturative stress happens during the acculturation process as a result of an individual facing various unfamiliar aspects in his/her daily life while trying to adapt to the new environment. Here, students studying in a new environment are prone to acculturative stress, especially those living in a multicultural environment (e.g., climate, food, language, race, landscape, culture) [86].

Living in different environments means facing different cultural and ideological values. Studying in a multicultural environment, students are more likely to be exposed to many new cultural values, which requires adjustment and adaptation. However, it is not easy to for the mindset to integrate and absorb new values. From the external environment to the core value, new cultural values need to get through points of filter. During the filtering process, old and new values are evaluated, connected, and compared [87] to integrate, synthesize, and incorporate values that are compatible or eject waning values. The filtering process is also where the acculturative conflict takes place and causes acculturative stress [22].

The extent to which an individual trusts a new value to be compatible with his/her mindset is the key in the filtering process. Living among new cultural values that an individual has not yet trusted will lead to sustained stress, which may cause brain dysfunction (certain types of depression) [88]. As the conflict becomes more serious, similar to the natural resistance of a human body, the filtering membrane of an individual might receive signal from the mindset to grow thicker to protect the old core values. This, in turn, makes the core values become even more distant with the external values, which greatly raises the level of distrust, intensifies the stress level, and eventually escalates the depression level. For example, low language proficiency makes an individual unable to communicate. Without communication, new cultural values cannot be interpreted, which may lead to distrust of new values, conflict and stress. Eventually, that may give rise to depression.

*Social connectedness.* The negative association between social connectedness and depression was found in the current study. In other words, social connectedness may help lessen the negative impacts of depression in a multicultural context. The social connectedness reflects "one's opinion of self in relation to other people" in term of emotional distance [32]. When students start to live in a new environment, their social networks begin to change. Living in a new environment where the ideological setting and surrounding people are different, students have to adapt and make new friends. This process can also be explained by the Mindsponge concept.

Assuming the self of an individual is also his/her own mindset. Making new friends is equivalent to accepting new values, so it requires a point of filter. At that point, a friend's personality, perception, and opinion will need to get through a filtering process to make sure whether a self/mindset trusts its new friend/value. If a new friend/value is trustworthy, it can possibly come closer to the self/mindset. This is when a student has a high sense of connectedness. On the other hand, if a new friend/value is in conflicts with the self/mindset, it will be considered as untrustworthy. Functionally, the filter will become thicker, increase the emotional distance from self/mindset to a new friend/value. In other words, feeling difficult to have new friends, students may have no sense of connectedness which may result in strong feeling of hopelessness [89,90] and gradually contribute to loneliness [91]. As loneliness is among the top predictors of depression [91–94] and hopelessness is also positively correlated with depression [95], the lower social connectedness, the higher depression will be. This way of explanation may also be compatible with the association between social connectedness and depression in other population.

In addition, social connectedness under a multicultural context may also have a buffering effect on acculturative stress [34]. As discussed, the mindset increasingly distrusts new value, the distance between mindset and new cultural value is widened due to the rising acculturative conflicts. Social connectedness may contribute to enhance trust of mindset toward new cultural values, which may facilitate the acculturation process. Eventually, since acculturation happens more smoothly, acculturative stress decreases, depression is less likely to happen.

#### **5. Recommendations for School Policy**

Based on the above findings, the study suggested some recommendations to have a healthier and more sustainable educational environment in international universities.


#### **6. Limitation and Recommendation for Further Research**

This paper has some limitations. In the data collection process, the team used sampling and needed to modify the model questionnaire slightly. In addition, findings were based on self-reported measures.

Regarding of the results on depression, the comparison with the prevalence in other studies might not be precise due to different types of questionnaire used in each study. The questionnaire was collected between November and January, when winter signs were apparent, so depression might be affected by seasonality, primarily when a large number of students originated from tropical areas. The impacts of seasonality on depression were confirmed in other studies [99,100]. Besides that, the current study had higher proportion of females than males, while depression among female students was more prevalent than among male students. The gender disparity in sampling might, therefore, leads to a higher percentage of depression.

The imbalances between the numbers of international and domestic students might be a limitation of this study. Besides that, different proportions of students from different origins might cause the results to have a regional bias among the surveyed international students.

There is a need to find out cause-effect relationships among the correlation, as well as for qualitative studies to provide more in-depth information. The study also recommend a meta-analysis for the mental health of international students in Japan like the following work [101]. This will increase the integration and synthesis of mental health research in Japan. Moreover, the Mindsponge concept was recommended to explain results in mental health study.

**Author Contributions:** Conceptualization, M.H.N., T.T.L. and S.M.; methodology, M.H.N. and S.M.; software, M.H.N.; validation, M.H.N. and S.M.; formal analysis, M.H.N.; investigation, M.H.N and T.T.L.; resources, M.H.N., T.T.L., and S.M.; data curation, M.H.N.; writing—original draft preparation, M.H.N. and T.T.L.; writing—review and editing, M.H.N. and S.M.; visualization, M.H.N.; supervision, S.M.; project administration, M.H.N. and S.M.; funding acquisition, S.M.

**Funding:** This research received no external funding.

**Acknowledgments:** We thank Ho Manh Tung (Ritsumeikan Asia Pacific University) and Nguyen To Hong Kong (Vuong & Associates) for giving advices and comments that greatly help us during the course of this research. We would also like to show our gratitude to the Research Office of Ritsumeikan Asia Pacific University for facilitating our research process.

**Conflicts of Interest:** The authors declare no conflict of interest.

#### **References**


© 2019 by the authors. Licensee MDPI, Basel, Switzerland. This article is an open access article distributed under the terms and conditions of the Creative Commons Attribution (CC BY) license (http://creativecommons.org/licenses/by/4.0/).
