The results of this prospective study suggest that whereas moderate alcohol intake (5–15 g/day) may protect women against the development of depression, higher amounts of alcohol intake may not confer any protection against depression.
Many studies have investigated the relationship between exposure to alcohol and depressive status. However, most of them have focused on subjects with AUD more than in subjects with alcohol intake levels in a lower range. Some other studies [17, 18] have longitudinally investigated the relationship between alcohol dependence or alcohol abuse, but not categories of alcohol intake, and major depression. In fact, a recent review  reported a positive association between AUD and major depression.
However, longitudinal studies focusing on a normal range of alcohol intake even if they are follow-up studies, may still be affected by reverse causation bias because the follow-up is relatively short (only 1–2 years), or because they have not excluded prevalent cases of depression, or have assessed depression only on a narrow window of time during follow-up. Moreover no Mediterranean cohort, with a higher red wine consumption, has ever assessed this association. Therefore the differences in the assessment of exposure and/or events, or in the type of patients evaluated make the comparison with these studies difficult. A study from the Health and Retirement Study  found an association between problematic alcohol intake and depression in a cohort of men aged over 50, but no association was found with non-problematic alcohol intake. Biennially for 6 years, alcohol intake was assessed through questionnaires, and incidence of depression was assessed using a symptoms scale with a cut-point. However, 63% of the participants were classified as problematic alcohol drinkers even when they reported drinking less than 1 drink per week. In addition, some cases of depression might be underestimated in that study since they only evaluated self-reported cases of depression during the last 2 years of the study, but not those occurring in the first 4 years of follow-up. Finally, that study only included men since the exposure was problematic drinking, which is much more prevalent among men. Another study examined how alcohol use predicts changes in psychological symptoms among 393 adolescents followed up for 18 months  and found that initial levels of alcohol use did not predict changes in depression. Inclusion criteria for that study were to score one standard deviation above the school mean on one of the four subscales of the Substance use Risk Personality Scale: hopelessness, anxiety, impulsivity, and sensation seeking. In addition, depression was assessed through the Brief Symptom Inventory, not through a doctor made diagnosis. Finally, previous or prevalent cases of depression were not excluded since diagnosis of depression as such was not assessed. Paljärvi et al.  performed a two-wave 5-years follow-up study to determine which aspect of drinking pattern would be the best predictor for depressive symptoms. Self-reported depressive symptoms were measured with a questionnaire at baseline and at year-5. They found that binge-drinking and also higher categories of alcohol intake were directly associated with depressive symptoms. These results are in contrast with our findings, however Paljärvi el al. excluded abstainers from their analyses, and used the moderate alcohol consumers as the reference category. Skogen et al.  found an U-shaped association between alcohol intake and both anxiety and depression. They distinguished between the possible former drinkers and abstainers. However, these authors acknowledged that the direction of causality was not clear. Another study evaluated the influence of alcohol dependence on the first incident depressive episode in a 1-year follow-up . Both disorders were investigated using Diagnostic Interview Schedule, according to DSM-III. Prevalent and previous cases of depression were excluded. Multivariate logistic regression was conducted separately for men and women, obtaining a stronger positive association for women. These results are not comparable with ours because we did not assess alcohol dependence as exposure.
Some case–control studies have investigated the influence of alcohol intake on the development of depression [25–27]. For example, an study conducted by Armenian et al.  after an earthquake found a protective association for alcohol intake in cases of depression without comorbidity of other psychiatric disorders: OR (95% CI) =0.6 (0.3-0.9). The post-traumatic situation makes our results not comparable with the results of this study. Two other studies found contradictory results [25, 26]. However, these case–control studies may also be affected by reverse causality bias.
Finally, some cross-sectional studies have found high comorbidity between both conditions ; however, their cross-sectional nature is an obstacle to assess causality, especially when it is well known that depression is a risk factor for AUD [17, 18].
Our results suggest an U-shaped association between alcohol intake and the development of depression, only for women. The non-significant association for heavy alcohol intake may be due to the very low average consumption and the scarcity of heavy drinkers in our cohort. This is consistent with Perreira et al.  who did not find any association for men. However, our results are not consistent with Mackie et al.  who did not find any association for men or women. These studies, and all cited above, are not generally comparable because of the differences in the alcohol and depression assessments . Moreover, the different results between these studies may be due to the differences in the distribution of alcohol intake between populations. When the population has moderate to low average alcohol intake, the association tends to be inverse or null [20, 21] and when the population has heavy alcohol intake, it tends to be positive [20, 22]. When assessing the relationship between alcohol intake and the development of depression a careful consideration of the temporal sequence is needed. In order to reduce the risk of falling into reverse causation bias, early incident cases of depression were excluded. These cases may be sub-clinical cases of depression that are likely to increase their alcohol intake as a consequence of their recent depression. On the contrary, alcohol intake information was collected at baseline and some cases of depression were developed after 10 years of follow-up. Considering on the other hand that this late incident cases might not be influenced by alcohol intake at baseline, we excluded these later cases and the magnitude of the association did not change considerably.
The development of depression may be influenced by the presence of other psychiatric disorders . In order to control for this possible bias we conducted the analysis after the exclusion of prevalent and lifetime cases of psychiatric disorders. The association did not change substantially. Other important diseases like cancer [25, 30] or cardiovascular disease may lead to the development of depression and also to changes in alcohol intake. After the exclusion of prevalent cases of both kinds of disorders, the association remained similar. Other potentially important issue is the comparison group. Abstainers might be a heterogeneous group. Those who do not drink may have different reasons not to do so. Some of them may be abstainer due to medical advice. Participants declaring being abstainers could have been former drinkers. These facts could bias the results towards a more protective association. However, after excluding former drinkers and those abstainers who did not drink due to medical advice, the association between alcohol intake and the development of depression did not change.
We conducted the analysis separately for men and women although the interaction product-term was not significant. However, previous studies [31, 32] found differences between both sexes, therefore we considered convenient to split our sample. Women seem to be more likely to the development of depression  and the same alcohol intake affects differently to men and women  due to their different body composition, affecting the volume of distribution, or different metabolism.
Some mechanisms have been proposed to explain the deleterious effect of heavy or chronic alcohol drinking on brain function ; however, moderate alcohol drinking have a GABAergic effect, acting on GABAA receptors , that may prevent or counteract the effects of depression on this system .
Our study has some important strengths, such as its prospective design (avoiding reverse causation bias), data collection and data analysis. Other strengths are its large sample size, its high retention rate, the good adjustment for potential confounders, the existence of published validation studies for our methods, the high correlation observed between the FFQ and the food records for alcohol intake in the validation study, and the high educational level of the participants, achieving high quality information and high internal validity. In our analysis, we have carefully considered the reverse causation bias, used validated diagnosis of depression, taken into account other psychiatric disorders, analysed sex-alcohol interaction, excluded other important diseases, and depurated the abstainers group.
A limitation of our study is the use of a self-reported clinical diagnosis of depression. We assume that a low proportion of participants could misreport the diagnosis. Moreover, participants who reported habitual use of antidepressant drugs were also considered incident cases. Research suggests first treatment for depression typically occurs several years following the onset of depression. The median time from onset of depression to first treatment has been reported to be 8 years in the U.S. population . Long lags in first contact for depression treatment have also been reported in other countries . Thus, identifying incident cases of depression by having received a diagnosis or prescribed medication by a medical doctor may result in missed cases. This is especially true for cases recruited later in the study who were followed up for a shorter period of time than case recruited earlier in the study. These missing cases can translate in low sensitivity of our case ascertainment definition. However, our validation study found very high specificity (0.96) for the self-reported diagnosis of depression and theoretically with perfect specificity, non-differential misclassification of disease, due to low sensitivity, will not bias the relative risk estimate .
A potential limitation is that few participants were heavy alcohol consumers, which may lead to a lack of statistical power to assess the relationship with depression for high levels of alcohol intake. Another limitation is that we do not have information about illicit drug use. However the SUN cohort participants are highly motivated, responsible and health-conscious who voluntarily agreed to complete long and complicated questionnaires. We expect that the use of illicit drugs will be very low or even non-existent among them.