*2.3. Statistical Analysis*

Initially, the trends of depression and other characteristics in the six continuous cycles were estimated with the Cochran–Armitage trend test. Then, the baseline characteristics of different groups were compared using the weighted Scott–Rao chi-square test [30]. HEI scores were described with a median (P25, P75). Finally, a series of weighted steps forward (likelihood ratio) binary logistic regression models were fit to assess the relationship between diet quality and depression. Estimates were weighted to be representative of the general adult population. All *p* values refer to two-tailed tests. Statistical analyses were conducted using the SPSS statistical package (Version 23.0; SPSS Inc., Chicago, IL, USA).

#### **3. Results**

Figure 1 described the study design, sampling, and exclusion; and 18,006 participants were excluded because of missing data on any of the covariates. Among them, 2361, 12, 219, 12, 3594, and 11,808 individuals were excluded because of missing data on income, education, BMI, smoking, drinking, and diabetes, respectively. Our final sample included 10,349 NHANES participants, representing a population of about 167.8 million non-institutionalized U.S. adults, with 48.6% being female and 72.0% being non-Hispanic White.

Table 1 described the prevalence of depression and associated characteristics in six continuous NHANES cycles, in which the trend tests were also conducted. The prevalence of depression grew with time, from 4.8% in 2005–2006 to 7.4% in 2015–2016. In addition, there are other points worth noting. For example, the proportion of women and adults aged over fifty years old increased with time. The proportion of adults with normal or low weight decreased with time, indicating the urgency of body shape management. The prevalence of diabetes increased with time, which reminds adults of the significance of blood glucose control. The proportion of adults with inadequate HEI status decreased with time, showing the improving diet quality in the six cycles.


**Table 1.** US trends in characteristics among adults aged 20 years or older.

Values are survey-weighted percentages. FPL = family income to poverty. HEI = healthy eating index.

Table 2 describes the characteristics of participants with the weighted Scott–Rao chisquare test. Adults with depression were more likely to be female, non-Hispanic Black,

obese, over 50 years old, current smokers, diabetic, alcoholic, have less than high school education, have low household income, and have inadequate HEI scores.


**Table 2.** Characteristics among adults aged 20 years or older by depression.

Values are survey-weighted percentages. FPL = family income to poverty.

Table 3 shows the results of three weighted logistic regression models. Model 1 was adjusted for demographics characteristics (i.e., sex, age group, race, income, and education). Model 2 was adjusted for all Model 1 covariates and BMI, smoking, and drinking status. Moreover, Model 3 was adjusted for all Model 2 covariates and diabetes. After adjusting for demographic characteristics, optimal HEI status was associated with 0.378 times lower odds (95% CI, 0.377–0.379) of current depression, relative to inadequate HEI status. Additional adjustment for BMI, smoking, drinking, and diabetes status did not substantially attenuate these relationships. After multivariable adjustment, adults with average HEI status (OR: 0.848, 95% CI: 0.846–0.849) and optimal HEI status (OR: 0.455, 95% CI: 0.453–0.456) were associated with

reduced odds of depression. Adults with diabetes were more likely to suffer from depression, with the odds ratio being 1.637 (95% CI: 1.634–1.640).

**Table 3.** Relationship between HEI and Depression among Adults aged 20 years or older.


FPL = family income to poverty; CI = confidence interval; OR = odds ratio. Model 1 = adjusted for demographics characteristics (i.e., sex, age group, race, income, and education); Model 2 = Model 1 covariates + BMI, smoking, and drinking status; Model 3 = Model 2 covariates + diabetes.

> Figure 2 shows the trend of the HEI-2015 score in the form of a violin plot. HEI score increased from 47.77 (39.39, 56.51) in 2005–2006, to 50.74 (43.13, 59.13) in 2015–2016. The proportion of adults with inadequate HEI status decreased with time, and that of adults with optimal HEI status increased with time.

> Figure 3 shows the results of three weighted logistic regression models in the form of a forest plot. As is shown, average and optimal HEI status are both protective factors for depression, reducing the depression risk by 15.2% and 54.5%. Diabetes is the risk factor of depression, increasing the depression risk by 63.7%. Cigarette smoking and heavy drinking are both behavioral risk factors.

**Figure 2.** The trends of HEI-2015 score in the six cycles from 2005–2016 in NHANES.

**Figure 3.** The forest plot shows the odds ratios of analyzing variables in three weighted logistic regression models.

#### **4. Discussion**

The results revealed that a higher HEI score was significantly correlated to less elevated depression symptoms, so we concluded that higher diet quality was significantly correlated to a lower risk of depression. Our findings also suggest that Mexican Americans are less likely to suffer from depression, which needs further analyses of genetic factors.

We analyzed the depression status of participants from nine symptomatic questions from the PHQ in 2005–2016 NHANES data, and the results revealed a depression prevalence of 6.9%. Since the sample is generalizable to the non-institutionalized civilian U.S. population, we assume the prevalence to be credible, similar to the data published by the WHO in 2017.

To our knowledge, many factors are associated with the occurrence of depression, for instance, alcohol consumption and diabetes. Moreover, it has been found by many researchers that women are more susceptible to depression than men [31–33], in accordance with our results (odds ratio, 1.889). Moreover, three logistic regression models were adopted in this analysis to explore a more appropriate model.

Some results have been widely recognized by many researchers. For example, the results of the weighted logistic models revealed that the risk of adults aged over 50 years old suffering from depression was 1.827 times that of adults aged 20–24 years old. The report from the WHO also concluded that the prevalence varies by age, peaking in older adulthood, similar to our results [3]. After multivariable adjustment, the odds ratio for other races changed from less than one, to more than one, which needs more specific classification. Our results revealed higher education level reduced the depression risk, similar to other studies [34–36]. In addition, our results revealed a negative correlation between household income and depression, consistent with other research [37–39]. It is recognized by many experts that a positive association exists between smoking and depression, as in our results [40–42]. Among the results, we found some interesting facts. Compared with normal and low weight participants, overweight adults were less likely to suffer from depression, contrary to our original thoughts. However, we found that Z Ul-Haq got a similar result from a cross-sectional study consisting of 37,272 participants, which revealed that overweight participants had better mental health than the normal-weight group [43]. Generally, overweight and obese adults suffer more ridicule and gossip than normal and low weight adults, which may be a reason for depression. After further analyses, we found that overweight adults accounted for about one third of American adults, so were obese adults. This meant high BMI is typical among American adults, and discrimination in the US is not as high as in China, leading to less psychological pressure in overweight adults. At the same time, overweight adults relieve pressure through diet, and are less susceptible to depression.

Another fact is that light and moderate drinking is a protective factor for depression. A meta-analysis in 2013 concluded that light drinking increased the risk of cancer of the oral cavity and pharynx, esophagus, and female breast [44]. However, many researchers have found that light and moderate drinking also have some health benefits, including reducing the risk of dementia [45], heart failure [46], ischemic stroke [47], type 2 diabetes [48], and allcause mortality [49]. In a cohort study with ten years of follow up, a J-shaped association was found with increased psychological distress among abstainers and heavy drinkers compared to light or moderate drinkers [50]. Our analysis believes that light and moderate drinking helps people deal with emotional issues and refresh themselves, thus reducing the risk of depression.

Physical activity has been found to be correlated to depression in a substantial number of studies [51–54]. However, three versions of the physical activity questionnaire were adopted in 2005–2016, and no identical and detailed information could be used to conduct an analysis. Therefore, physical activity was not included as a covariant in this analysis.

Diet quality was reported to be correlated to diabetes in many studies, not only in randomized controlled trials, but also in large population-based cohorts [55,56]. The relationship between diabetes and depression was recognized by R J Anderson in 2001 [57], yet whether there is a causal relationship is still under debate.

Thus, we conducted a mediation analysis to figure out whether diet quality influenced depression through diabetes. The HEI-2015 score was the independent variable, and the PHQ-9 score was the dependent variable. The result showed a mediating effect, but the mediating effect's proportion was only 0.14%. For further mechanism analysis, we need to explore more possible factors causing the mediation effect.

The present study has several limitations. The cross-sectional design of the study is the primary limitation, and no causation should be inferred from this study. Second, using self-reported 24-h dietary recall data and the PHQ-9 questionnaire is a limitation, as they are subject to over- or under-reporting. Finally, 18,006 participants were excluded because of missing data on any of the covariates, which may have affected the results. Despite these limitations, our study has several strengths. Using a large, nationally representative database to estimate diet quality is a major strength of the present study. Adopting the latest edition of HEI is another strength. Moreover, data in six cycles were combined to increase the sample size.

Since we found that a higher HEI-2015 score is associated with a lower risk of depression, the next step of our plan is to figure out the pathway by which HEI-2015 influences depression, with structural equation models.

By virtue of this article, we would like to make the public aware of the significance of better diet quality on depression. Since better diet quality is associated with less depression risk, why do we not improve our diet quality to reduce the risk of depression?

#### **5. Conclusions**

This study's primary finding is that depression is rapidly growing in prevalence among American adults, from 4.8% in 2005–2006, to 7.4% in 2015–2016. Poor diet quality is significantly associated with elevated depressive symptoms. An optimal HEI-2015 score reduces the risk of suffering from elevated depressive symptoms by 54.5% compared to an inadequate HEI-2015 score.

**Author Contributions:** Conceptualization, K.W.; methodology, K.W. and C.Y.; software, K.W. and S.W.; validation, K.W. and S.W.; formal analysis, K.W.; investigation, K.W.; resources, S.W.; data curation, K.W.; writing—original draft preparation, K.W.; writing—review and editing, Y.Z.; visualization, K.W. and C.Y.; supervision, J.N.; project administration, H.X. All authors have read and agreed to the published version of the manuscript.

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

**Institutional Review Board Statement:** Not applicable.

**Informed Consent Statement:** Not applicable.

**Data Availability Statement:** Publicly available datasets were analyzed in this study. This data can be found here: [https://wwwn.cdc.gov/nchs/nhanes/ContinuousNhanes/Default.aspx?BeginYear= 2005].

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

#### **References**

