Introduction

Depression is an increasingly common disease. At a global level, over 300 million people are estimated to suffer from depression, equivalent to 4.4% of the world’s population. In 2015, depressive disorders led to a global total of over 50 million years lived with disability. Depression can also lead to suicide, which is currently one of the leading causes of death in 15–29 years old [1].

A growing body of research strongly supports an interrelationship between depression and metabolic syndrome, obesity, type 2 diabetes, and cardiovascular disease [2,3,4,5,6], which, in turn, are widely recognized as diet-related diseases [7]. Several studies have found a positive association between westernized diets rich in fast food and added sugars and depressive symptoms [8,9,10]. However, the degree of food processing according to NOVA classification [11] and its association with the risk of depression has not yet been studied.

Ultra-processed foods (UPF), as defined by the NOVA food classification system [11, 12], are energy-dense, ready-to-consume industrial formulations, which typically contain little or even no intact food, and are depleted in dietary fiber, micronutrients, and other bioactive compounds. The overall purpose of ultra-processing is to create convenient, hyper-palatable, and highly profitable food products that are usually packaged attractively and marketed intensively, and are designed to displace all other food groups [13]. Consumption of UPF has exponentially increased worldwide and, over the last few years, epidemiological evidence has demonstrated that UPF consumption is associated with poorer diet quality [14], functional gastrointestinal disorders [15], obesity [16], and associated diseases [17,18,19].

Thus, the aim of our study was to prospectively examine the association between UPF consumption and the risk of depression in the SUN cohort.

Methods

Study population

The “Seguimiento Universidad de Navarra” (SUN) Project is a prospective, multipurpose, and dynamic Spanish cohort of university graduates focused on evaluating the impact of diet and lifestyle on the prevention of non-communicable diseases. The recruitment of participants started on December 1999, and is permanently ongoing. Graduates from the University of Navarra and other Spanish universities and professional unions are invited to participate on an annual basis. Once participants accept to enter the study, they receive a detailed questionnaire by ordinary mail or an email with a personal code to answer the questionnaire at the SUN website. Every 2 years, shorter follow-up questionnaires are sent by ordinary mail or e-mailed to track changes in lifestyle habits, diagnosis of new diseases, and overall well-being. Further details about the methodology and characteristics of participants have been previously reported [20]. The overall follow-up rate approaches 91%.

Up to December 2016, 22,561 participants had completed the baseline questionnaire of the SUN Project. Participants who were lost to follow-up, who had not completed at least one follow-up questionnaire, who were outside of predefined limits for baseline total energy intake [< 3347 kJ/day (< 800 kcal/day) or > 16,736 kJ/days (4000 kcal/day)] in men, and [< 2092 kJ/day (< 500 kcal/day) or > 14,644 kJ/days (3500 kcal/day) in women] [21] and participants with cancer, cardiovascular disease, or diabetes at baseline were excluded from the analysis. We also excluded participants with prevalent depression, which were those who reported life-time history of depression, diagnosis of depression, and/or antidepressant medication use at baseline or within the first 2 years upon entry to the cohort. Finally, 14,907 participants were included in the prospective analyses for the present study (Fig. 1).

Fig. 1
figure 1

Flowchart

Full size image

This study was conducted according to the Declaration of Helsinki and the protocol (including the informed consent process) was approved by the Institutional Review Board of the University of Navarra. The completion of the self-administered questionnaire was considered to imply informed consent.

Outcome assessment

Incident cases of depression were defined as participants who were free of depression at baseline and reported a diagnosis in the successive follow-up assessments conducted only after the first 2 years of follow-up. The case definition required a positive response to the following question “Have you ever been diagnosed of depression by a medical doctor?” or the reporting of a regular use of antidepressant drugs in any of the biennial follow-up questionnaires from the 4th year up to the 16th year of follow-up, i.e., diagnoses or treatments reported for the first time in the 2-year follow-up questionnaire were considered as “prevalent cases”. We excluded those participants with a depression diagnosis during the first 2 years of follow-up from the analyses to avoid a potential reverse causality bias. The rationale for this caution is that it is possible that participants with subclinical depression at the beginning of our study might have changed their food habits due to their latent mood disorder. Although antidepressants could have been prescribed for conditions other than depression, this situation is extremely unusual in Spain [22]. Therefore, we considered any of both, the use of antidepressants or the medically made diagnosis, as valid criteria to operationally define incident cases of depression.

The self-reported medically diagnosed depression was validated in a subsample of our cohort applying the Structured Clinical Interview for DSM-IV (SCID-I) as ‘gold standard’ by experienced psychiatrists blinded to the answers of the questionnaires [23]. The percentages of confirmed cases of depression and non-depression were 74.2% (95% CI 63.3, 85.1) and 81.1% (95% CI 69.1, 92.9), respectively.

Exposure assessment: UPF consumption

Dietary intake was assessed at baseline using a self-administered 136-item semi-quantitative food-frequency questionnaire (FFQ) that has been repeatedly validated in Spain [24, 25]. The full-length validated FFQ was repeated after 10 years of follow-up.

To minimize the potential effect of diet variation during follow‐up, we considered repeated measurements of the diet (i.e., we used dietary data from both FFQs in the repeated-measure analyses).

The FFQ recorded a typical Spanish portion size for each item, and consumption frequencies were divided into nine categories ranging from never/almost never to > 6 servings/day. Daily food consumption was estimated by multiplying the portion size (g) by the consumption frequency for each food item.

UPF were defined according to the NOVA classification [11], which classifies foods into four groups based on the extent and purpose of industrial processing: (i) unprocessed or minimally processed foods (fruits and vegetables, grains, nuts and seeds, fresh and pasteurized milk, and natural yogurt with no added sugar or artificial sweeteners); (ii) processed culinary ingredients (salt, sugar, honey, vegetable oils, butter, lard, and vinegar); (iii) processed foods (canned or bottled vegetables and legumes, fruits in syrup, canned fish, unpackaged cheeses, freshly made bread, and salted or sugared nuts and seeds); and (iv) UPF and drink products [carbonated drinks, processed meat, biscuits (cookies), candy (confectionery), ‘instant’ packaged soups and noodles, sweet or savory packaged snacks, and sugared milk and fruit drinks].

The frequency of UPF consumption per person (g/day) was estimated using the sum of the UPF items shown in Supplementary Table 1. Total UPF consumption was adjusted for total energy intake using the residual method separately for men and women [21]. The sample was divided into quartiles according to total consumption.

Table 1 Age- and sex-adjusted* baseline characteristics [mean and standard deviation (SD), or  %] of participants according to their ultra-processed food (UPF) consumption in the SUN Project
Full size table

Assessment of covariates

In addition to dietary data, the baseline assessment gathered information regarding sociodemographic characteristics (e.g. sex, age, marital status, and employment status), anthropometric variables (e.g., weight and height), lifestyle and health-related habits (e.g., smoking status, sleeping hours, television viewing, and eating attitudes such as following a special diet), and medical history (e.g., chronic diseases and medication use). It also included a physical activity questionnaire that collects information about 17 activities and has demonstrated fair validity against a triaxial accelerometer [26].

Participants also answered questions about personality and behavior. Self-perception of competitiveness, anxiety, and psychological dependence levels were ascertained by means of the following Likert-type questions previously used in this cohort [27] and associated with depression: (a) Do you consider yourself a competitive, nonconformist, fighter person, who demands everything of yourself at work and sometimes even more of what you can afford?, (b) Do you consider yourself a tense, aggressive, usually feeling overloaded, highly strung person or do you think of yourself as a relaxed and calm person?, and (c) Do you think you have enough resources, preparation, and autonomy to solve any problems at work, or do you exclusively depend on others to do it?. For each question, 11 possible answers could be chosen by the participant ranging from 0 (more conformist, relaxed, or autonomous) to 10 (more competitive, tense, or dependent).

Total energy and nutrient intake were estimated from the FFQ administered at baseline using Spanish food composition tables [28]. Adherence to the Mediterranean diet was assessed using the Mediterranean Diet Score (MDS) proposed by Trichopoulou [29].

Body Mass Index (BMI) was calculated from the weight (in kilograms) and height (in meters) values self-reported at baseline, which were previously validated in a subsample of the SUN cohort obtaining successful results [30].

Statistical analyses

Age- and sex-adjusted baseline characteristics of participants according to quartiles of UPF consumption were evaluated using inverse probability weighting.

For each participant, we computed person-years of follow-up from the date of returning the baseline questionnaire to the date of returning the questionnaire in which depression was reported, the date of death, or the date of returning the last follow-up questionnaire, whichever came first. To take advantage of the repeated measurements of the dietary exposure, we fitted Cox proportional-hazard regression models to assess the relationship between baseline and 10-year UPF consumption and the incidence of depression during follow-up. Hazard ratios (HRs) and their 95% CIs were calculated considering the lowest quartile as the reference category. Tests of linear trend across increasing categories were conducted by assigning the medians to each category; this variable was treated as continuous.

Cox regression models included age as the underlying time variable. Multivariable models were adjusted for sex, baseline BMI (tertiles: kg/m2), physical activity (quartiles: METS-h/week), smoking status (never, current, former), marital status (single, married, other), living alone (yes/no), employment status (employed, unemployed, retired), working hours per week (continuous), health-related career (yes/no), years of education (continuous), total energy intake (continuous, kcal/day), adherence to the MDS (low, moderate, high: 0–9 scale), and baseline self-perception of competitiveness, anxiety, and dependence levels (continuous: 0–10 scale). Analyses were stratified by age groups (10-year periods) and calendar year of recruitment (1999–2001, 2002–2004, 2005–2007, 2008–2010, and from 2011 onwards). Robust standard errors (SEs) were used.

As recommended [31], the selection of potential confounders in the multivariate model was based on a priori subject-matter knowledge (taking into account previously published scientific literature, including results from other studies in this cohort regarding risk factors for depression), and not from statistical associations detected in the data nor from stepwise procedures or changes in point estimates after adjusting for potential confounders.

To minimize the potential effect of diet variation during follow‐up, UPF consumption was updated with dietary data collected after 10 years of follow-up and time-varying Cox proportional-hazard models were fitted with repeated measures of diet.

The potential non-parametrical non-linear association between cumulative average consumption of UPF and incident depression was calculated with restricted cubic splines [32]. The results were adjusted for the same potential confounding factors as the Cox model regression analysis.

Multiplicative interactions between quartiles of UPF baseline consumption and different sociodemographic and lifestyle-related variables regarding depression risk were assessed. Possible effect modification of UPF intake by sex, age (< 40/≥ 40 years), BMI [below/above the median (23 kg/m2)], and physical activity [below/above the median (21 METS-h/week)] was tested using likelihood ratio tests comparing the fully adjusted model and the same model with an interaction product term.

Finally, sensitivity analyses were used to assess the robustness of our results under different scenarios. We repeated all analyses including participants with chronic diseases at baseline, excluding prevalent pregnant women (their diet could vary), excluding participants with prevalent obesity (BMI ≥ 30 kg/m2), excluding participants with high baseline levels of depression-related personality traits in addition to those with prevalent depression, exclusively considering participants that reported a medical diagnosis of depression as incident cases without any use of antidepressants, additionally adjusting for incident events of cardiovascular disease, additionally adjusting for sleeping hours and following a special diet, and additionally adjusting for use of aspirin and non-steroidal anti-inflammatory drugs.

All p values presented are two-tailed; p < 0.05 was considered statistically significant. Analyses were performed using STATA/SE version 12.0 (StataCorp, College Station, TX, USA).

Results

The current study included a total of 14,907 participants (6031 men and 8876 women). After a median of 10.3 years of follow-up (148,240 person-years), we recorded 774 incident cases of depression.

Table 1 shows the age- and sex-adjusted sociodemographic and lifestyle characteristics at baseline of participants according to quartiles of consumption of UPF. Those participants who reported higher consumption of UPF were, on average, more likely to be unmarried, live alone, consume less fruits and vegetables, have a lower dietary fiber intake, lower adherence to the Mediterranean Diet, and to be less physically active. They were also more likely to be current smokers and use analgesics more frequently.

Table 2 shows that a higher consumption of UPF was directly associated with the risk of developing depression during the follow-up (p for trend = 0.004), with the fourth quartile having a significantly increased risk (HR, 95% CI 1.33, 1.07–1.64) compared to the lowest quartile after adjusting for a wide array of potential confounders.

Table 2 Hazard ratios and 95% CI of incident depression according to consumption of ultra-processed food (UPF)
Full size table

When analyses with repeated measures were conducted after updating UPF consumption with dietary data obtained from a new FFQ after 10 years of follow-up, we obtained comparable estimates (HR, 95% CI for the highest vs. lowest quartile: 1.35, 1.09–1.67) (p for trend: < 0.001).

Restricted cubic spline analysis was used to assess the shape of the dose–response association (Fig. 2), and we found that increasing UPF consumption from null or low to moderate was enough to increase the risk of depression. There was an apparent threshold effect, and we did not find subsequent further increases in risk beyond a very high consumption (> 400 g/day).

Fig. 2
figure 2

Dose–response association between ultra-processed food (UPF) consumption and depression risk in the SUN Project. Hazard ratios (95% confidence intervals)

Full size image

Likelihood ratio tests determined that no statistical interaction existed between UPF consumption and sex (p for interaction 0.27), BMI (p for interaction 0.83), or age (p for interaction 0.87). However, the p value for multiplicative interaction between UPF consumption and physical activity was statistically significant (p = 0.008). We performed a stratified analysis and, in fully adjusted models, the association between UPF consumption and the risk of depression was stronger when levels of physical activity were under the median (21 METS-h/week) (HR, 95% CI 1.47, 1.10–1.97 for highest vs. lowest quartile) than when levels of physical activity were above the median (HR, 95% CI 1.13, 0.81–1.57).

As shown in Table 3, all sensitivity analyses confirmed the robustness of our analysis, as the main results did not substantially change in any of the different alternative scenarios.

Table 3 Sensitivity analyses
Full size table

Discussion

In this study conducted in a young middle-aged Mediterranean population, participants with the highest UPF consumption (fourth quartile) showed a 31% higher risk of developing depression during follow-up than those with the lowest consumption (first quartile). This result was expected in view of the available studies regarding the risks associated with UPF consumption [14,15,16,17,18,19], and is in line with the previous findings in the other studies evaluating the role of diet in depression [10, 33,34,35,36,37,37].

In other analyses of our cohort, we found that micronutrient intake inadequacy, which is characteristic of a Western diet high in UPF, could exert a moderate role in the development of depression [34]. We have also observed that participants who consumed more fast food and commercial bakery products [36], or more transfatty acids [37], had a higher risk of depression compared to participants who had a lower intake. Nevertheless, the degree of food processing should also be taken into account.

The health risks associated with food processing vary according to the nature, extent, and purpose of processing. Almost all food consumed in industrialized settings is somehow processed, and this is not necessarily negative as many processes are useful to preserve food or make it safe. However, ultra-processing is different as it is used to make products from combinations of many ingredients (hence ‘ultra-processed’), usually with little or even no intact whole foods. As a result, UPF are energy-dense, high in unhealthy types of fat, refined starches, free sugars and salt, and poor sources of protein, dietary fiber, and micronutrients [13].

UPF are made to be hyper-palatable (extremely tasty) and attractive to consumers due to their convenience (long shelf-life, and able to be consumed anywhere, any time). Their formulation, presentation, and marketing often promote overconsumption, and the evidence so far shows that displacement of minimally processed food and freshly prepared dishes by UPF is associated with unhealthy dietary nutrient profiles [14, 38,38,40] and several diet-related non-communicable diseases [41]. Thus, in addition to the harms directly linked to the consumption of UPF due to their negative nutritional attributes, UPF consumption also indirectly harms the overall diet quality given their ability to displace and interfere with the consumption of healthy food. Numerous prospective observational and experimental epidemiological studies have reported inverse associations between diet quality or adherence to traditional and prudent diets, such as the Mediterranean Diet, and the risk of depression with consistent results across different countries [42,43,44,45]. However, no studies assessing the relationship between UPF and the incidence of mental disorders can be found in the literature.

Although depression results from a complex interaction of social, psychological, and biological factors, several mechanisms could explain the relationship between UPF and depression. An extensive body of research has described the bidirectional relationship (the presence of one increases the risk for developing the other) between depression and other diet-related chronic diseases such as obesity or metabolic syndrome [2,3,4], which have also been associated with UPF consumption [16,17,18,19], suggesting that they may share common pathophysiological mechanisms. In fact, metabolic disturbances and inflammatory processes are present in both cardiometabolic diseases and depressive disorders. These include alterations in systems shown to be involved in both mood and homeostatic regulation: the hypothalamic–pituitary–adrenal (HPA) axis (hyperactivity of the HPA axis and elevated cortisol secretion) [46], immuno-inflammatory signals (a pro-inflammatory diet—typically rich in UPF—has been previously associated with depression in our cohort, especially among subjects with cardiometabolic diseases, as a result of responses to chronic low-grade inflammation that can influence brain activity) [35, 47], neuroendocrine hormones responsible for energy metabolism such as leptin and insulin (consumption of added sugars, which is also related with leptin and insulin resistance, has been associated with an increased risk of depression) [10, 48], and the microbiota–gut–brain axis (a greater density of detrimental microbiota may lead to hampered synthesis of brain-derived neurotrophic factor (BDNF) from the hippocampus) [49]. Furthermore, there is convincing evidence that nutritional content of processed foods is not accurately conveyed to the brain, which raises the possibility that the degree of food processing, beyond nutritional quality, affects physiology in unanticipated ways that could promote metabolic dysfunction and favor the development of depression (due to the described bidirectional relationship) [50].

A remarkable result obtained in our analysis suggests that the effect of a high UPF consumption on depression could be particularly detrimental among individuals with low levels of physical activity during leisure time, which may be due to the anti-inflammatory and beneficial neurochemical effects of physical activity [51, 52].

Some potential limitations of our study need to be mentioned. First, the data on food intake and clinical diagnosis of depression or the use of medication were self-reported, and the FFQ used to evaluate UPF consumption was not specifically designed for this purpose. Although the FFQ and the self-reported diagnosis of depression have been previously validated [23,24,25], some degree of information bias may exist. However, the use of a prospective cohort design mitigates this to some extent, as misclassification is more likely to be non-differential, and, therefore, would bias the results towards the null. Moreover, FFQs have been demonstrated to be the most appropriate method in assessing habitual diet in large cohorts [21]. Regarding the assessment of personality traits, dependence trait in our study refers specifically to the workplace, so we acknowledge that excluding other daily life environments is a limitation. However, the assessment of self-perception of competitiveness, anxiety, and dependence through three short and simple Likert-type questions has previously shown long-term prospective associations with the future risk of depression in our cohort [27]. Although we adjusted for major depression risk factors and other potential confounders to account for the fact that those participants who reported higher consumption of UPF also had an unhealthier lifestyle in general, residual confounding cannot be ruled out in our study.

Regarding the generalizability (external validity) of the study results, the SUN cohort only includes university graduates, so our participants are not representative of the Spanish population. Nevertheless, having a homogeneous cohort with highly educated participants improves the retention rate, reduces the likelihood of misclassification bias increasing internal validity, and minimizes potential confounding by educational level and, therefore, by socio-economic status [53]. Moreover, highly educated participants in this cohort might be more aware of the negative effects of detrimental foods like UPF and, thus, social desirability bias cannot be excluded. The presence of this non-differential misclassification bias would lead the associations towards the null. Finally, we also acknowledge not taking into account different subtypes or levels of depression, and the possibility that antidepressant medications were prescribed for other pathologies different from depression. However, when we considered physician diagnosis as the only criteria to define incident cases of depression in the sensitivity analyses, the association remained unchanged.

Despite these limitations, our study had many strong aspects, including the relatively large sample size, its prospective design, the high retention rate, the use of updated nutritional data, the ability to control for a variety of major potential confounders, the existence of published validation studies of our assessments, and the restriction to highly educated participants, who may be able to provide more reliable information. In addition, we performed multiple sensitivity analyses that confirm the robustness of our results and the exclusion of incident depression during the first 2 years of follow-up decreases the possibility of reverse causality.

In conclusion, a higher consumption of UPF was associated with an increased risk of developing depression among young middle-aged adult university graduates from a Spanish cohort. This effect could be even more important among individuals that are insufficiently physically active. To the best of our knowledge, this is the first study to assess the association between UPF consumption and depression. Further studies are needed to confirm our findings, and deepen the understanding about the relationship between UPF consumption and mental health.