We did not find a significant decrease in depression risk among participants at high risk of CVD assigned to MD supplemented with either nuts or EVOO in this randomized controlled primary prevention trial. However, when the analysis was restricted to subjects with DM2, participants assigned to MD-nuts had a 40% reduction in depression risk compared with the control group, which was significant.
To our knowledge, this is the first randomized field trial that has ascertained the effect of an intervention with an overall dietary pattern on depression risk in adults. Only a few prospective observational studies have inversely related healthy dietary patterns to the risk of developing adult depression [7–15]. Although some of these studies were based only on cross-sectional assessments [8, 9, 11–13], their results were consistent with those obtained after several years of follow-up in prospective cohorts [7, 10].
Regarding the MD, a very recent cohort study, the Australian Longitudinal Study on Women’s Health , found that women in the highest quintile of adherence to a ‘Mediterranean-style’ diet were 37% less likely to report depressive symptoms after 3 years of follow-up, (adjusted odds ratio (OR) = 0.63, 95% CI 0.47 to 0.85) compared with those in the lowest quintile of adherence. Similar results were previously obtained in the SUN (Seguimiento Universidad de Navarra) Project  in which participants with the highest adherence to the MD had a significant reduction in the risk of developing depression (comparison of fifth versus first quintile: adjusted HR = 0.58; 95% CI 0.44 to 0.77).
Nevertheless, the available evidence is still sparse and not definitive. Moreover, interpretation of findings resulting from observational studies demand caution . Most of the studies had a cross-sectional design [8, 9, 11–13], which is a weak design for inferring cause and effect relationships that can only be suggested. In such studies, exposure is ascertained simultaneously with disease and, therefore, an alternative interpretation of the results could be made as a consequence of reverse causation bias; for example, that depression may lead to poorer dietary habits . These large studies generally need the use of food frequency questionnaires to collect information on dietary factors, and although such questionnaires have been customarily used and generally have been validated, it is known that they have some potential for misclassification bias. Finally, it is necessary to take into account the possibility of residual confounding.
By contrast, findings from observational studies can be supported by some biochemical and physiological mechanisms that may be implicated in depression risk and that are also intimately related to dietary factors. Examples include low-grade systemic inflammation and endothelial and metabolic disturbances, which may be present in patients with depression [28–31]. Indeed, a large number of clinical trials including the PREDIMED study and some observational studies have reported an inverse association between adherence to a MedDiet pattern and the levels of inflammatory, metabolic, or endothelial biomarkers [32–35]. The presence of inflammatory processes and endothelial dysfunction compromise the production and secretion of brain-derived neurotrophic factor (BDNF), a peptide implicated in synaptic plasticity and neuronal survival, and whose levels are decreased in patients with depression . In a previous sub-analysis of the PREDIMED trial, conducted by our group in the PREDIMED-NAVARRA center, significantly higher plasma BDNF levels were seen for patients with depression assigned to the MD-nuts compared with those assigned to a control diet .
Our results indicate that adherence to a Mediterranean dietary pattern supplemented with nuts could be particularly important to prevent depression among participants with DM2. The association between obesity, DM2, metabolic syndrome (MetS), and depression has been suggested in several studies [38–40]. Metabolic disturbances/dysregulation of markers such as insulin, leptin, glucose [30, 31], or tryptophan/serotonin [41, 42] could explain the link between obesity, DM2, and depression. In fact, in the Whitehall II cohort study, low insulin secretion was associated with an increased risk of developing depressive symptoms . Recent studies have also reported an increased risk of depressive symptoms associated with higher levels of leptin , especially in the presence of abdominal obesity [31, 44]. The association of leptin with depression could be explained not only by its metabolic properties but also by its neurobiological activity, as leptin is able to affect neuroprotection, cognition, and mood in the hippocampus, the cortex, and other brain areas . Moreover, in several studies, hyperleptinemia and insulin resistance, which can be present in obesity, MetS, and DM2 have been also linked to endothelial dysfunction or inflammation processes [46, 47], conditions also present in depression. Not only metabolic disturbances but also inflammatory markers have been recently associated with depressive symptoms in participants with diabetes from the SEARCH for Diabetes in Youth cohort study .
Interestingly, within the PREDIMED trial, several sub-analyses support this hypothesis. After 3 months of follow-up, participants assigned to the MD + nuts exhibited significant reductions in fasting glucose and insulin levels and in Homeostasis Model Assessment index compared to those assigned to the control diet . Moreover, a significant reversion of MetS was seen in the MD-nuts group after 1 year of intervention; in this group, the OR for reversion of MetS was 1.7 (95% CI=1.1 to 2.6) compared with the control diet group .
Moreover, in a recent clinical trial based on a sample of patients with MetS assigned to receive a control diet or a diet enriched in mixed nuts for 12 weeks, two intermediate urinary metabolites of the tryptophan pathway (which leads to biosynthesis of serotonin and melatonin from tryptophan) were identified: N-acetylserotonin sulfate and hydroxyindoleacetic acid . These metabolites are considered urinary markers of nut intake, as walnuts are one of the most important dietary sources of serotonin . However, the authors pointed out that the presence of these urinary metabolites could be due both to a high intake of walnuts and to a high endogenous serotonin turnover following the intake of these food items. Thus, such metabolites could be markers of the effect of the global dietary intervention, instead of being markers of intake .
We did not find a significant effect on depression risk for adherence to the MD supplemented with EVOO in the overall sample. However, when the analyses were restricted to those participants with DM2. the relationship was strengthened. The effect on depression of the intervention with MD supplemented with EVOO for patients with DM2, although non-significant, was similar to that obtained for patients with DM2 in the analysis of the effect of MD on risk of CVD in the PREDIMED framework . As mentioned above, several physiological processes may be responsible for the link between depression and cardiovascular and metabolic disorders [28–31]. In fact, a beneficial role of supplementation with a Mediterranean diet enriched with olive oil in various processes such as oxidative stress, inflammation, lipid metabolism, or weight regulation for patients with DM2 or multiple sclerosis has been described in several epidemiological studies [52–54].
Although our results indicate a possible beneficial role of MD supplemented with nuts (compared with a low-fat diet) in the prevention of depression, with a risk reduction of around 25%, the results were not statistically significant. There are several reasons that may explain the lack of statistical significance. First, the number of new cases was not large. In the report from the SUN cohort relating the Mediterranean diet with depression, 480 new cases of depression were identified , whereas in the present study, the number of new cases was less than half of that the SUN cohort. Second, observational studies generally compare extreme quintiles or quartiles of adherence to a dietary pattern, therefore, high between-subject variability in adherence allows for comparison of extremes in exposure and guarantees that a strong effect can be detected. By contrast, in the present study, the variability in adherence to the MD between the Mediterranean and control groups was relatively small. In fact, after 6 years of follow-up, the participants assigned to the control group showed a mean adherence to the MD of 9 points (up to 14 points maximum), whereas the mean adherence within the groups assigned to both MDs was 10.5 points .
Moreover, in intervention trials, the degree of change in the dietary habits of participants is always suboptimal, because of the lack of compliance with the intended intervention of some participants . In fact, the control group in the PREDIMED trial was assigned a healthy dietary pattern recommended by the American Heart Association to prevent CVD , and also they tended at baseline to be on a dietary pattern that was similar to the Mediterranean diet. This reality suggests that there would be even a potentially greater benefit of the Mediterranean diet if it were compared to a typical (and unhealthy) Western diet . Third, the PREDIMED trial was not designed to study depression as a primary end-point; the primary end-point was in fact a composite of cardiovascular clinical events. Therefore, it is likely that some degree of misclassification in the ascertainment of depression cases may have happened. However, we took care to ensure that all included cases had received a medical diagnosis of depression (or were under antidepressant treatment), and that they were not prevalent cases at baseline. In fact, we excluded all participants with a short follow-up period (the first 3 years) in order to ensure that reverse causation could not explain the reported results.
Although in the PREDIMED trial, follow-up data were available for 97% of the sample, it is important to note that the time each participant remained in the trial was variable because centers started to recruit participants from 2004 to 2006, many centers continued to recruit participants until 2009, and the trial ended in 2010. Thus, to obtain a more homogeneous period of follow-up in all participants, we carried out our analyses only with those participants with at least 3 years of follow-up. Usually, to minimize the influence of the presence of undiagnosed disease at baseline, a methodological approach in cohort studies has been to exclude cases occurring during the first 3  or even the first 5 years of follow-up . We chose a 3-year period because we believed it likely that participants with undetected (sub-clinical) depression at baseline might receive a delayed diagnosis of depression during the first, second, or even third year of follow-up. However, we considered that in participants with true but undiagnosed depression at baseline, it was unlikely that their depression would remain undetected during the first 3 years of follow-up and that they might yet receive a delayed diagnosis only during the fourth year of follow-up or later.
One of the main problems of a nutritional intervention trial is the variability in the compliance of participants with the intended dietary intervention. To address this issue, we carried out a per protocol approach as an ancillary analysis to compare the results, taking into account what the participants really did. The category of highest adherence to the MD after 3 years of intervention exhibited the lowest risk of depression. This is consistent with the results obtained in the ITT analyses. However, in both case,s the lack of statistical significance is probably attributable to the small number of cases included.
Finally, the highest retention rates were seen in the MD groups, and the lowest retention rate was seen in the control (low-fat diet) group. This high retention rate in the two MD group by be partly attributable to the free provision of specific food items (EVOO and nuts). In additon, the palatability of the Mediterranean diet has been identified as a key factor in its higher compliance . In general, the research group was able to obtain nearly complete follow-up for the main outcomes because participants represented a stable and well-defined population regularly attending their general practitioners. In addition, a comprehensive search for events was performed yearly through review of all the medical records of participants in all the university hospitals of the area in which the respective recruiting center was located. Nevertheless, 59 participants without a diagnosis of incident depression could not be contacted for at least 2 years (22 participants assigned to the low-fat group, 19 to the MD-nuts and 18 to the MD+EVOO). Thus, differential misclassification bias in the outcome is not a very likely possibility. Moreover, when the analyses were repeated imputing the missing values for subjects lost to follow-up, the results did not change substantially.