Summary
The present study is the first to assess the association between depressive symptoms and all-cause mortality in the Chilean population. The mortality risk remained elevated for those with depressive symptoms after adjustment for demographic, SES, behavioural and chronic conditions in the 8.5-year period analysed. There was consistent evidence of a 38% increased risk of all-cause mortality for those with depressive symptoms compared to those without such symptoms in both cohorts.
Relationship with literature
The results from our work agree with the hypothesis of a higher risk of all-cause mortality for depressed individuals compared to those without. This hypothesis has been previously tested with both positive and null findings [2, 5]. Our fully adjusted estimates are lower than Cuijpers’ review estimate [2] but the latter was attenuated after removing outliers and adjusting for publication bias (RR 1.58 95% CI 1.51–1.65). Compared to literature from LMIC, the estimates from our work also tended to be smaller. After adjusting for outliers and publication bias, the review of the topic from LMIC reported a pooled RR of 1.60 (95% CI 1.37–1.86). Nevertheless, when only studies of high-quality are considered, the estimate decreased to 1.48 (95% CI 1.32–1.67).
Although pooled estimates from the literature tend to be larger in size than that estimated from our work, when critically assessed and adjusted for relevant factors, there is a much better agreement with our results. Most estimates from the literature are only adjusted by age and sex [4] and do not address the limitations described in the background, such as measurement of exposure, sample size, type of population, follow-up length, publication bias, outliers and adjustment for confounders [5, 11]. When only age and sex adjustments are considered, model 1, our estimate shows good agreement with the estimate from Cuijper’s review after adjustments for publication bias and outliers. The estimates of the association between depression and mortality in the literature are relatively similar when elements such as quality of studies and variables adjusted for are considered [2, 8]. This strengthens the generalisability of our results and suggests that, rather than specific mechanisms that influence all-cause mortality, more generic mechanisms drive the association between depression and mortality [2].
The hypothesized biological dysregulation as mechanism posits that depression causes changes at the biological level, such as changes in inflammatory responses [32], in the hypothalamic-pituitary-adrenal (HPA) axis [33, 34], cortisol levels [35] and noradrenaline levels [36]. These biological changes caused by depression increase the risk of mortality. On the other hand, unhealthy behaviours as mechanism range from lack of physical activity [37, 38], smoking [39, 40] to alcohol and drug abuse [41, 42]. In turn, these behaviours associated with depression, increase the risk of mortality. In this paper, the effect of certain unhealthy behaviours —smoking and physical activity— and chronic diseases strongly associated with unhealthy behaviours —diabetes and HBP— was accounted for. Thus, the estimated effect of depression on all-cause mortality in our model was independent of these factors. However, if unhealthy behaviours, such as lack of exercise, mediate the association between depression and mortality, adjusting for chronic conditions also associated with those unhealthy behaviours could potentially lead to overadjustment of our results [43]. The comparison of models with and without adjustments for unhealthy behaviours shows an inconsequential attenuation of the estimated effect, suggesting that, after accounting for depressive symptoms, demographic and SES variables, these unhealthy behaviours have little effect on all-cause mortality in our data. On the other hand, adjustment for chronic diseases strongly linked to unhealthy behaviours showed a small attenuation of the estimated effect between depression and mortality. Therefore, a potential overadjustment due to consideration of potential mediators in the model was considered to be negligible.
Despite some consistency in the positive association between depression and mortality reported in the literature [4, 44], Machado et al. emphasized that quality of studies could influence the size of this association [5]. This is true to some extent, as evidence from Machado’s and other reviews suggest a larger effect of studies of low-quality compared to high-quality studies. Nevertheless, in sensitivity analysis adjusted for comorbidities in community samples, the evidence from the umbrella review [5] still provides evidence for a positive association between depression and mortality, with a reported effect size very similar to our work (RR 1.38, 95% CI 1.29–1.47). It has also been posited that the increased mortality risk among depressed individuals is caused by other physical disorders that in turn cause depression [7]. However, both our estimate for the effect of depressive symptoms on mortality was robust to adjustment for chronic diseases and our sensitivity analysis excluding those with poorer health at baseline to address reverse causality do not support this idea.
Strengths
There were several strengths in this study. This is the first Latin American study on the topic to be conducted on two nationally representative samples, enhancing the generalisability of its results. Moreover, both cohorts had a large sample size and there were enough deaths to ensure an appropriate power for this study. The use of two cohorts allowed to examine secular trends in the estimated effect. Our main exposure showed a remarkable stable effect on mortality. The decision to restrict the follow-up up to 8.5 years allowed us to make a reasonable comparison between cohorts, as there is some evidence of changes in the estimated effect according to the length of the follow-up period [2]. Attrition has been consistently identified as a limitation in prospective studies [45]. As the mortality data of ENS’ participants were linked using administrative data and not by contacting all participants at follow-up, attrition was not a limitation. Furthermore, this study used reliable data. Diabetes and HBP were ascertained through physiological measurements made by a nurse and not through self-reporting. One of the most consistent limitations highlighted in the literature was the lack of adjustment for relevant confounders. In this paper, a wide variety of confounders was included from different domains: demographic, socioeconomic, lifestyle variables and chronic conditions. The attenuation between model 1 and the fully adjusted model suggests that these adjustments —especially the SES variables— were relevant to assess the association between depression and mortality.
Limitations
The results from this study need to be examined cautiously as there were some limitations. Some pertinent variables were unavailable or had high level of missingness and could not be used in the analysis. These variables relate to diet, other chronic conditions, and other mental health conditions. Therefore, residual confounding cannot be discarded. The instrument for measuring our exposure was not a structured psychiatric interview, instead, we relied on a screening instrument. This has the potential limitation of misclassification of subjects compared to the gold standard. However, there is evidence that the used instrument shows a high level of agreement with structured psychiatric interviews [27]. The measurement of depressive symptoms at baseline could reflect either an acute state at the moment of interview or the presence of a more chronic condition. The trajectories of depressive symptoms cannot be ascertained by our design, consequently, it was not possible to assess how much of the estimated effect of depressive symptoms on mortality is due to an acute or repeated depressive episodes. Public health policies could focus either on prevention —if the mere presence of depression increases risk of mortality— or on the management of depressive episodes if repeated episodes are more relevant for mortality. Overadjustment based on including potential mediators in the association between depressive symptoms and mortality could underestimate the true association. However, as previously discussed, the potential overadjustment due to the inclusion of potential mediators of unhealthy behaviours and chronic conditions in the models 3 and 4, respectively, was judged to be negligible based on the small attenuation of the estimates in our analyses The potential bias due to incompleteness of mortality records was addressed by examining the quality of the data in terms of cause and number of deaths. Overall, there was a relatively constant number of deaths over time and a similar proportion in causes of death by cohort. The small differences in these aspects were considered irrelevant. Lastly, there is some evidence of a difference in the effect of depression on mortality depending on the cause of mortality and gender [44]. Our analysis did not consider these aspects solely based on a lack of power to detect significant differences for a more specific outcome, such as cardiovascular mortality, or by gender.