Food Addiction Mediates the Relationship between Perceived Stress and Body Mass Index in Taiwan Young Adults

Perceived stress is the degree of stress experienced by an individual in the face of a stressor. Studies have shown that stress affects emotions, leads to behavioral changes, and is likely to trigger physical illnesses. According to the World Health Organization (WHO), stress is classified as a health epidemic of the 21st century; in the meantime, the percentage of adults being overweight and with obesity has continued to grow after reaching 38.9% in 2016. Hence, it is unclear whether perceived stress has become a factor affecting progressive obesity and whether food addiction (FA) is an intermediate factor. The purposes of this study were to (1) investigate the FA prevalence among young adults in Taiwan, (2) understand correlations among perceived stress, FA, and the body mass index (BMI), and (3) determine the potential mediating effect of FA due to perceived stress on BMI. The study was conducted through an online questionnaire, composed of a basic data form, the Perceived Stress Scale (PSS), and the Yale Food Addiction Scale (YFAS). We received 1994 responses and analyzed 1780 valid samples. Results showed that 231 participants met the FA criteria, accounting for 12.98%. Perceived stress was positively correlated with BMI (95% confidence interval (CI) 0.013~0.088, p-value 7.8 × 10−3), and perceived stress was positively associated to FA (95% CI 1.099~1.154, p-value < 10−4), which was also positively correlated with BMI (95% CI 0.705~2.176, p-value 10−4). FA significantly mediated the relationship between PSS and BMI with an indirect effect size of 25.18% and 25.48% in the group that scored 31~40 on the PSS. The study concluded that among people seeking weight loss, proper stress management and screening for FA in order to apply related therapies may be an effective method for weight management.


Introduction
The prevalence rate of overweight or obesity among adults increased 5.3% globally from 2006 to 2016, in which the Western Pacific (6.9%, 24.8~31.7%) and Eastern Mediterranean (6.9%, 42.1-49.0%) regions exhibited the highest increases according to the World Health Organization (WHO) [1]. Obesity is an issue worthy of study. It is directly related to mortality and chronic diseases such as heart disease, diabetes, hypertension, stroke, sleep apnea, cancers, and metabolism syndrome (MetS) [2]. However, research indicated that among people attempting to lose weight, uncontrollable eating behaviors

Perceived Stress Scale (PSS-10)
The PSS [23] was implemented to assess the degree of stress experienced by participants in the last month. Items were designed to discover how unpredictable, uncontrollable, and overloaded respondents find their lives. It also includes direct questions about current perceived stress.
Respondents were asked to score on a 5-point Likert-type scale from 0 "never" to 5 "very often". The higher the summed score, the greater the stress perceived by participants. The Cronbach's alpha value for PSS in our sample is 0.894. For trend test analysis, the summed score (range 0~40) was divided into four groups (0~10, 11~20, 21~30, and 31~40).

Yale Food Addiction Scale (YFAS)
The YFAS [24,25] evaluates eating behaviors of an individual in the past 12 months. It is a 25-item measurement that assesses FA symptoms and advises a "diagnosis" of FA. Items were designed based on seven symptoms of substance dependence listed in the Diagnostic and Statistical Manual of Mental Disorders (DSM)-IV, which are substances taken in larger amounts and for longer periods than intended, a persistent desire or repeated unsuccessful attempts to quit, use of much time/activity to obtain, use, or recover it, important social, occupational, or recreation activities given up or reduced, use continuing despite knowledge of adverse consequences (e.g., failure to fulfill role obligations and used when physically hazardous), tolerance (a marked increase in amount and a marked decrease in effect), characteristic withdrawal symptoms, and substance taken to relieve withdrawal. The Cronbach's alpha value for YFAS in our sample is 0.888. FA is recognized when an individual meets three or more of the above symptom criteria and reports clinically significant impairment or distress.

Statistical Analysis
Population characteristics were described by the number (n) and size of portions (%) of the sample. A Chi-squared test was employed to analyze whether a certain variable (feature) of two groups had the same distribution. Student's t-test was implemented to verify whether the average of two independent groups was the same. The Cochran-Armitage trend test was administered to assess whether there was a correlation between a categorical variable of two groups and another sequential scalar-type variable. In addition, we used a path analysis concept to test the significance of direct or indirect effects among perceived stress, FA, and BMI. The multiple logistic regression analysis was applied to model the effect of perceived stress on FA status adjusting for sociodemographic Nutrients 2020, 12,1951 4 of 11 characteristics (sex, age group, employment status, and educational attainment) and the estimated odds ratios (OR) and corresponding 95% confidence interval (CI) were presented. The effect of FA status on BMI was estimated by a multiple linear regression adjusting for sociodemographic characteristics. The indirect effect of perceived stress on BMI mediated through FA was estimated by a multiple linear regression with BMI as the dependent variable, PSS as an independent variable in the model, and controlling for FA and other sociodemographic characteristics. The overall effect of perceived stress on BMI was estimated by a similar multiple regression but not controlling for FA in the model. The mediation effect for FA from perceived stress to BMI was calculated with a mediation analysis by comparing the two multiple linear regressions with controlling for and without controlling for FA [26].
All analyses are performed with SAS version 9.4 (SAS Institute, Cary, NC, USA). An alpha level of p < 0.05 was considered statistically significant.

Sample Description
Participant characteristics are shown in Table 1. The sample included 1387 females (77.92%) and 393 males (22.08%): 73.71% of participants were 20~29 years old, and the average age was 26.96 years. Just over half (51.46%) possessed a BMI in the normal range, 23.13% were obese, and overall, subjects had an average BMI of 23.98 kg/m 2 . Participants predominantly held an educational level of college or university (73.82%) or master's degree or above (23.60%), 53.99% were full-time employees, 28.54% were students, and 87.7% of participants reported a PSS score in the range of 11~30 (Table 1).  Table 1). The proportion of female subjects (14.92%) who were assessed as having FA was significantly greater than that of males (6.11%) ( Table 1). Between the FA and non-FA groups, there were significant distribution differences in participants who were underweight, normal, overweight, and with obesity ( Table 1). The average BMI level of the FA group was significantly higher than that of the non-FA group (Table 1), and proportions of FA subjects in the four BMI categories were 9.62%, 10.59%, 15.61%, and 17.77%, respectively. In the trend test, the proportion of FA participants increased as BMI increased ( Table 1). The mean value of perceived stress scores in subjects with FA was also significantly greater than that of those without an FA (Table 1). In the trend test, as the level of perceived stress increased, the number of FA subjects significantly increased ( Table 1).
The proportion of women significantly increased as the level of perceived stress scores increased ( Table 2). After adjusting for perceived stress, age, employment status, and educational attainment, sex was significantly related to a diagnosis of FA. Compared to men, women had increased odds of FA (OR = 1.126, 95% confidence interval (CI) 1.099~1.154) ( Figure 1). sex was significantly related to a diagnosis of FA. Compared to men, women had increased odds of FA (OR = 1.126, 95% confidence interval (CI) 1.099~1.154) ( Figure 1).

Effects of Perceived Stress on BMI
After adjusting for sex, age, employment status, and educational attainment, the perceived stress level was positively associated with a higher odds of a diagnosis of FA (odds ratio 1.126, 95% CI 1.099~1.154) ( Figure 1).

Figure 1.
Odds ratio (OR) of food addiction (FA) for perceived stress after adjusting for sociodemographic characteristics.

Effects of Perceived Stress on BMI
After adjusting for sex, age, employment status, and educational attainment, the perceived stress level was positively associated with a higher odds of a diagnosis of FA (odds ratio 1.126, 95% CI 1.099~1.154) (Figure 1).

Effect of FA on BMI
After adjusting for sex, age, employment status, and educational attainment, a diagnosis of FA was related to a greater BMI (Figure 2).

Effect of FA on BMI
After adjusting for sex, age, employment status, and educational attainment, a diagnosis of FA was related to a greater BMI (Figure 2).

Effect of Perceived Stress and FA on BMI
After adjusting for the perceived stress level, sex, age, employment status, and educational attainment, a diagnosis of FA was related to a greater BMI ( Figure 3).
To calculate the effect through a mediation analysis, we set the two models as follows: Model A: where Y is a variable and X is an independent variable, then c is an estimated coefficient representing the degree of correlation of a unit of X variation with respect to Y variation). Model B: X→M→Y (Y_i = α−2 + c^' X_i + bM_i + ε_i, where Y is a variable, X is an independent variable, M is also an independent variable, c^' is an estimated coefficient, representing the degree of correlation of a unit of X change with respect to Y variation, b is also an estimated coefficient representing the degree of correlation of the unit of M change for Y, while the M variable may also be affected by the X variable, and b might not only include the influence of the M variable but also the combined influence of X and M). Therefore, using the indirect effect calculation of a mediation analysis, the indirect effect percentage is 1−c^'/c [26].

Effect of Perceived Stress and FA on BMI
After adjusting for the perceived stress level, sex, age, employment status, and educational attainment, a diagnosis of FA was related to a greater BMI (Figure 3).

Figure 2.
Effect of food addiction (FA) on the body mass index (BMI) after adjustment for sociodemographic variables. The model was obtained from a multiple linear regression, and was adjusted for sex, age group, employment status, and educational attainment. *** p-value < 0.0001. Coeff., coefficient; CI, confidence interval; Ref., reference.

Effect of Perceived Stress and FA on BMI
After adjusting for the perceived stress level, sex, age, employment status, and educational attainment, a diagnosis of FA was related to a greater BMI (Figure 3).
To calculate the effect through a mediation analysis, we set the two models as follows: Model A: X→Y (Y_i = α−1+cX−i+ε−i, where Y is a variable and X is an independent variable, then c is an estimated coefficient representing the degree of correlation of a unit of X variation with respect to Y variation). Model B: X→M→Y (Y_i = α−2 + c^' X_i + bM_i + ε_i, where Y is a variable, X is an independent variable, M is also an independent variable, c^' is an estimated coefficient, representing the degree of correlation of a unit of X change with respect to Y variation, b is also an estimated coefficient representing the degree of correlation of the unit of M change for Y, while the M variable may also be affected by the X variable, and b might not only include the influence of the M variable but also the combined influence of X and M). Therefore, using the indirect effect calculation of a mediation analysis, the indirect effect percentage is 1−c^'/c [26].  To calculate the effect through a mediation analysis, we set the two models as follows: Model A: X→Y (Y_i = α−1+cX−i+ε−i, where Y is a variable and X is an independent variable, then c is an estimated coefficient representing the degree of correlation of a unit of X variation with respect to Y variation). Model B: X→M→Y (Y_i = α−2 + cˆ' X_i + bM_i + ε_i, where Y is a variable, X is an independent variable, M is also an independent variable, cˆ' is an estimated coefficient, representing the degree of correlation of a unit of X change with respect to Y variation, b is also an estimated coefficient representing the degree of correlation of the unit of M change for Y, while the M variable may also be affected by the X variable, and b might not only include the influence of the M variable but also the combined influence of X and M). Therefore, using the indirect effect calculation of a mediation analysis, the indirect effect percentage is 1−cˆ'/c [26].
The indirect effect size of perceived stress on BMI through FA was 25.18% ( Figure 4). The indirect effect size of the perceived stress 31~40 score group on BMI through FA was 25.48% (data not shown).
The indirect effect size of perceived stress on BMI through FA was 25.18% (Figure 4). The indirect effect size of the perceived stress 31~40 score group on BMI through FA was 25.48% (data not shown).

Discussion
Our aim in this study was to examine interrelationships among perceived stress, FA, and BMI. Whether FA acts as a mediator in the relationship between perceived stress and an increased BMI in young adults was also a main concern.
Our results showed that perceived stress during the past month was positively correlated with BMI. However, a study of 5077 Hispanic/Latino adults indicated no associations between perceived stress in the past month and being overweight or with obesity [27]. A five-year longitudinal study in Australia demonstrated that although perceived stress in the past month was not associated with BMI, it was significantly associated with life health behaviors such as daily energy intake and physical activity [28]; therefore, as long as a perception of stress persists, it is highly likely to increase the risk of obesity. Additionally, research also showed stronger associations between perceived stress and weight gain in participants who were normal weight, overweight, or younger [28], which may explain why we found significant such relationships in our sample (an average age of 26.96 years and a mean BMI in the normal range) while others did not.
In terms of long-term stress, a prospective study of adults in Australia showed that people who had three or more stressors in the past year had significantly higher weight gains than those who did not [28]. The number of chronic stressors had a higher obesity OR and was significantly associated with the waist circumference and body fat percentage; in heavier-weight groups, more chronic stressors were carried [27]. The evidence of higher perceived stress producing a greater BMI was reiterated.
As to the relation between perceived stress and FA, this study showed that the higher the perceived stress score, the higher the odds of FA. This result is consistent with several studies. A French study of 1349 college students showed that psychological distress from perceived stress, anxiety, and depression was significantly positively correlated with the number of FA symptoms [29]. A study among 408 type 2 diabetic patients indicated that subjects with higher symptom counts of FA reported a higher degree of stress [30]. Research has pointed out that stress tolerance was significantly negatively correlated with emotional eating, exogenous eating, uncontrolled eating, and FA [31]. The association between perceived stress, food addiction, and BMI. Perceived stress = perceived stress scale with range 0-40; Food addiction = food addiction or non-food addiction recognized by Yale Food Addiction Scale; BMI = Body Mass Index (kg/m 2 ). Regression coefficients of multiple linear regressions are unstandardized and the coefficient for the indirect relationship between perceived stress and BMI controlling for food addiction in parentheses. The indirect effect size of perceived stress on BMI through FA is 25.18%. * p < 0.01.

Discussion
Our aim in this study was to examine interrelationships among perceived stress, FA, and BMI. Whether FA acts as a mediator in the relationship between perceived stress and an increased BMI in young adults was also a main concern.
Our results showed that perceived stress during the past month was positively correlated with BMI. However, a study of 5077 Hispanic/Latino adults indicated no associations between perceived stress in the past month and being overweight or with obesity [27]. A five-year longitudinal study in Australia demonstrated that although perceived stress in the past month was not associated with BMI, it was significantly associated with life health behaviors such as daily energy intake and physical activity [28]; therefore, as long as a perception of stress persists, it is highly likely to increase the risk of obesity. Additionally, research also showed stronger associations between perceived stress and weight gain in participants who were normal weight, overweight, or younger [28], which may explain why we found significant such relationships in our sample (an average age of 26.96 years and a mean BMI in the normal range) while others did not.
In terms of long-term stress, a prospective study of adults in Australia showed that people who had three or more stressors in the past year had significantly higher weight gains than those who did not [28]. The number of chronic stressors had a higher obesity OR and was significantly associated with the waist circumference and body fat percentage; in heavier-weight groups, more chronic stressors were carried [27]. The evidence of higher perceived stress producing a greater BMI was reiterated.
As to the relation between perceived stress and FA, this study showed that the higher the perceived stress score, the higher the odds of FA. This result is consistent with several studies. A French study of 1349 college students showed that psychological distress from perceived stress, anxiety, and depression was significantly positively correlated with the number of FA symptoms [29]. A study among 408 type 2 diabetic patients indicated that subjects with higher symptom counts of FA reported a higher degree of stress [30]. Research has pointed out that stress tolerance was significantly negatively correlated with emotional eating, exogenous eating, uncontrolled eating, and FA [31].
In stress-related diseases, FA was positively correlated with post-traumatic stress disorder (PTSD). In addition to the indication that the prevalence of FA increased with the symptom count of PTSD, a study of 49,408 female nurses showed that those who had the highest number of PTSD symptoms (six or seven symptoms) had more than twice the FA rate compared to those without PTSD symptoms or a traumatic history [32].
In terms of sex differences, this study was similar to other studies in that women had a higher level of perceived stress than men [27,29], and had a higher rate of FA [33] or eating disorders [34]. One study pointed out that the proportion of FA and the number of symptoms in women aged 18~34, 35~54, and over 55 years were significantly higher than men in the same age groups, and there was no significant difference among the groups [35].
However, another study showed that although women had significantly higher negative emotional effects (especially anxiety and perceived stress) than men, emotional eating and FA symptoms, after adjusting for anxiety and perceived stress, a sex difference only occurred for the emotional eating score but not on the FA symptom counts. On this basis, that study believed that a true sex difference lies in the emotion-driven eating behavior rather than clinical addiction symptoms like disordered eating behaviors [29]. Inconsistent with our findings, after adjusting for sociodemographic variables of perceived stress, age, employment status, and educational attainment, the odds of FA for females was still significantly higher than that of men. Under the same PSS, the difference may have been caused by the use of a simpler modified YFAS of that study, different adjusted variables, or the effects of European and Asian ethnicities and cultures.
Our results showed a positive relation between FA and BMI, which is consistent with other research [4,34,36,37]. Individuals who were overweight or with obesity had a higher relative risk of FA than those who were normal or underweight according to either BMI or body fat percentage measurement [33]. A study using neuroanatomy to examine relationships among brain structure, FA, and BMI showed that a higher BMI predicted a significantly lower thicknesses of the (pre)frontal, temporal, and occipital cortices and an increased volume of left nucleus accumbens [37]. The former is believed to be related to the ability to regulate or suppress emotions and self-control, while the latter is considered to play an important role in rewards, happiness, laughter, addiction, aggression, fear, and placebo effects [38]. Although the study claimed that symptoms of FA did not account for the major part of the structural brain variances associated with BMI in the general population, it may still explain additional structural differences in the orbitofrontal cortex, a hub area of the reward network [37].
As we hypothesized, the present study identified a positive path effect of perceived stress to FA and then to BMI. FA is a mediator of perceived stress that affects BMI. To our best knowledge, there is only one other study that examined the path of these three. Despite the study also showing a mediating role of FA and higher psychological distress being indirectly related to an increase in weight via addictive-like eating symptoms, that study found a negative association between psychological distress and weight that did not reach statistical significance [29]. It was explained as having an unexpected suppressive effect on the statistical analysis [29]. Therefore, we believe a conclusion of proper stress management and screening for FA that would benefit the population seeking to lose weight can still be made.
Since the degree of perceived stress varies in individuals, understanding characteristics of people who have less stress tolerance and are highly reactive to stress is important [39]. Research has shown that individuals with high impulsivity levels are more likely to result in obesity [40][41][42], and activities such as exercise, music, and meditation can help sooth emotions and avoid overeating [43]. Although the content of FA has been debated over whether it is more of a "substance addiction" or a "behavior addiction" [43,44], pharmacologic therapy and cognitive behavioral therapy are both believed to be effective ways to treat substance addiction and addictive behaviors, and were proven to elevate the efficiency when both therapies were applied as treatment [45,46].
The strengths of the study was its large sample size in Taiwanese young adults and that it is the first study to examine interrelationships among perceived stress, FA, and BMI with Asian criteria of BMI. Moreover, it provides the prevalence rate of FA in a general young adult population, which is also not yet fully discussed in the literature. However, there were significantly more female participants than male participants in our sample. Since we found a sex difference among the odds of FA, the prevalence rate of FA in the general population may be lower than 12.98% in our overall population. Further Nutrients 2020, 12, 1951 9 of 11 investigation from the angle of sex differences may provide valuable insights. The limitation of the study was its cross-sectional design. In spite of the statistical method used to analyze the pathway, a longitudinal study is still needed to support a firm causal relation from perceived stress and FA to BMI. Another limitation that is worth mentioning is that there may be selection biases in those who completed the online study (i.e., those with access to internet, computer, social media) that might limit the generalizability of the findings.

Conclusions
The study indicated that perceived stress was positively correlated with BMI, perceived stress is positively associated to FA, and FA was also positively correlated with BMI. FA is a mediator with an indirect effect size of 25.18% between perceived stress affecting BMI, and 25.48% for the group with a perceived stress score of 31~40. We concluded that among people seeking to lose weight, proper stress management and screening for FA in order to apply related therapies may be an effective way for weight management.