Introduction
Depression is a prevalent mental health disorder in older ages, representing the largest contributor to ‘years lived with disability’ (World Health Organization (WHO), 2017a). It is associated with a decline in wellbeing (D'Alisa et al., Reference D'Alisa, Miscio, Baudo, Simone, Tesio and Mauro2006), increased morbidity and mortality (Nuyen et al., Reference Nuyen, Volkers, Verhaak and Schellevis2005; Farrokhi et al., Reference Farrokhi, Abedi, Beyene, Kurdyak and Jassal2014) and frequent use of health-care services (Andersson et al., Reference Andersson, Magnusson, Carstensen and Borgquist2011; Belloni et al., Reference Belloni, Morgan and Paris2016). Hence, identification of factors that are related to this disorder is of importance, especially among older persons, whose numbers are on the increase (WHO, 2017b).
Past analyses have pointed out the importance of early life conditions for later-life health (Arpino et al., Reference Arpino, Gumà and Julià2018); exposure to diseases, low socio-economic status (SES) and adverse life experiences may undermine physical and mental health both in adulthood and older ages (Galobardes et al., Reference Galobardes, Lynch and Davey2004; Case et al., Reference Case, Fertig and Paxson2005; Luo and Waite, Reference Luo and Waite2005; Haas, Reference Haas2007; Doblhammer et al., Reference Doblhammer, Van den Berg and Fritze2013; Pudrovska and Anikputa, Reference Pudrovska and Anikputa2014). Further, adverse adulthood and later-life circumstances, including bereavement, loneliness, disability, negative and stressful life events (Aziz and Steffens, Reference Aziz and Steffens2013), poor health and financial strain, are partly accountable for later-life mental health disorders (Butterworth et al., Reference Butterworth, Rodgers and Windsor2009; Gallagher et al., Reference Gallagher, Sawa, Kenny and Lawlor2013; Halmdienst and Winter-Ebmer, Reference Halmdienst and Winter-Ebmer2014; Crowe and Butterworth, Reference Crowe and Butterworth2016). By contrast, higher educational attainment is an important protective factor (Ladin, Reference Ladin2008) which may attenuate the effects of childhood financial adversity (Schaan, Reference Schaan2014).
A thorough analysis of later-life depression would involve the study of conditions pertaining to three major domains of life: health, SES and adverse experiences (Moore et al., Reference Moore, McDonald and McHugh-Dillon2014; WHO, 2014; Monnat and Chandler, Reference Monnat and Chandler2015; Montez et al., Reference Montez, Bromberger, Harlow, Kravitz and Matthews2016). However, associations are complex, as these three domains are interrelated; further, childhood circumstances may have an effect on adulthood and later-life conditions. For instance, lower childhood SES may affect health in later life both directly and indirectly (Luo and Waite, Reference Luo and Waite2005); on the one hand, it is related to lower adulthood SES (which itself is related to worse health both in adulthood and later life) and, on the other hand, to worse childhood health, which is associated with poorer health and lower SES in adulthood and, subsequently, in later life. Adverse experiences in childhood are also of interest in tracing the origins of depression in older ages (Dvir et al., Reference Dvir, Ford, Hill and Frazier2014), acting either independently or in conjunction with other adverse and stressful adulthood experiences (Nurius et al., Reference Nurius, Green, Logan-Greene and Borja2015).
To disentangle these multifaceted associations and assess causality, life trajectories and paths leading from childhood to later-life depression have been explored in past research. Potential mechanisms are outlined below.
Conceptual framework and mechanisms
There are four theories dealing with mechanisms leading from early life disadvantage to later-life poor health (Pudrovska and Anikputa, Reference Pudrovska and Anikputa2014). First, the critical period model stresses the importance of childhood conditions which may irrevocably affect biological processes, having thus a direct effect on later-life health, independently of other experiences over the lifecourse. Second, the accumulation of disadvantage model purports that exposure to risk factors accrues over the lifecourse. Hence, adversity experienced in different stages of life contributes equally to poor health later on. Third, the pathway model deals with trajectories over the lifecourse; adverse early life circumstances may lead to disadvantage in mid-life and, subsequently, to poor health. Hence, this model emphasises the role of mediators in the relationship between childhood conditions and later-life health. Fourth, the social mobility model claims that early life adversity is mitigated by improved adulthood circumstances while adulthood adversity may obliterate the beneficial effects of a more advantageous childhood.
An alternative approach describing how early life circumstances may affect later-life health categorises the relevant mechanisms into three groups (Hertzman and Boyce, Reference Hertzman and Boyce2010). The first group includes latent mechanisms or direct effects. These reflect the long-term consequences of biological processes, initiated by exposure to risk factors during gestation and early childhood, and are unrelated to intervening adulthood circumstances. In that sense, latent mechanisms conform with the critical period model. The second group involves pathway mechanisms or indirect effects; childhood adversity influences adulthood circumstances which, in turn, affect later-life health. These mechanisms are consistent with the pathway model. Finally, the third group includes conditional effects: the consequences of early life adversity on later-life health are conditioned on mid-life circumstances which can modify these associations. These mechanisms conform with the social mobility model.
Although the above-mentioned theoretical models and mechanisms offer different explanations of how early life adversity may impact later-life health, in practice they may be viewed as complimentary. In fact, past analyses have found that mechanisms often co-exist. Pakpahan et al. (Reference Pakpahan, Hoffmann and Kröger2017a) suggest that the effect of low childhood SES on later-life self-rated health is nearly fully mediated by adulthood SES, a fact in line with the pathway model, whereas the effect of childhood health remains substantial, supporting the critical period model partly. Torres and Wong (Reference Torres and Wong2013), on the other hand, examining the effects of childhood poverty on later-life depression in Mexico, find both a strong direct effect and an indirect one, the latter partly mediated by adulthood SES, supporting thus both the latency and the pathway mechanisms. By contrast, Tani et al. (Reference Tani, Fujiwara, Kondo, Noma, Sasaki and Kondo2016) find that the effects of childhood SES on the onset of later-life depression are consistent mainly with latency mechanisms, as they remain significant in spite of adjusting for adulthood SES and current health status. The findings of Pudrovska and Anikputa (Reference Pudrovska and Anikputa2014) support, on the one hand, the pathway model, as parental SES seems related to adolescent and mid-life SES which, in turn, is related to lower mortality, and, on the other hand, the social mobility model, as downward social mobility seems to have a detrimental effect. Zimmer et al. (Reference Zimmer, Hanson and Smith2016) show that early life and adulthood SES have an independent effect on later-life health, which is consistent with the accumulation of disadvantage model. Finally, Nurius et al. (Reference Nurius, Green, Logan-Greene and Borja2015) also find evidence supporting the accumulation of disadvantage model regarding the effects of childhood adverse experiences on adult mental health.
A gender perspective
An issue of special interest is the study of disparities in depression between genders. Depression is more prevalent among girls and women both in adulthood and later life (Hankin, Reference Hankin2002; Van de Velde et al., Reference Van de Velde, Bracke and Levecque2010). In many cases, this is due to hormonal differences and personal traits (Albert, Reference Albert2015), as well as to the way males and females deal with stressful life events. Additionally, females experience more stressful events and adversity in childhood compared to males (St Clair et al., Reference St Clair, Croudace, Dunn, Jones, Herbert and Goodyer2015), while they also tend to perceive the cause of such events in a negative way (Hankin, Reference Hankin2002).
Findings regarding the importance of childhood circumstances on later-life health also imply the existence of gender differentials. Though Kendig et al. (Reference Kendig, Gong, Yiengprugsawan, Silverstein and Nazroo2017) find a strong direct and an indirect effect of childhood health on later-life self-rated health in China, consistent both with latency and pathway mechanisms, the indirect effect is mediated differently by gender; among men marital status and urban residence matter more whereas among women educational attainment is an important mediator. Angelini et al. (Reference Angelini, Howdon and Mierau2019) find that the association of childhood SES with later-life depression, which persists even after controlling for later-life socio-economic circumstances, is stronger for women. Arpino et al. (Reference Arpino, Gumà and Julià2018) suggest that poor childhood health has a direct effect on later-life health, which is in accordance with the critical period model, whereas the effects of low SES in early life are mediated by educational attainment, supporting thus the pathway model. Moreover, the mediating effects of educational attainment are more substantial among women.
Aims of the study
In this context, the present study attempts to shed light on these multilayered associations, considering in a comprehensive manner, first, the direct impact of socio-economic disadvantage, adverse experiences and poor health over the lifecourse on later-life depression and, second, the indirect effect of childhood circumstances and the extent that these are mediated by different adulthood and later-life predictors. Emphasis is given on disparities between genders as later-life depression is more prevalent among women, who also exhibit greater vulnerability to adversity compared to men. Hence, the analysis addresses separately the mechanisms that lead from childhood circumstances to later-life depression for men and women. It is anticipated that both mechanisms are relevant. Further, based on past analyses, we expect to observe differentials between men and women; more specifically, it seems possible that SES may have a stronger effect among men (Back and Lee, Reference Back and Lee2011) where adverse experiences may be more important among women (Flores and Kalwij, Reference Flores and Kalwij2014; Almuneef et al., Reference Almuneef, ElChoueiry, Saleheen and Al-Eissa2017).
The contribution of the study lies in the holistic approach, which deals simultaneously with adversity in three domains of life and in three different periods, i.e. childhood, adulthood and later life. Most similar analyses deal only with the effects of childhood SES while a few additionally include childhood health, but none, to the best of our knowledge, considers also adverse experiences. In the analysis, cross-sectional and retrospective data from the Survey of Health, Ageing and Retirement in Europe (SHARE) have been used while mediation is assessed on the basis of a decomposition technique.
Data and methods
Sample
In the analysis SHARE data have been used. SHARE is a multi-disciplinary and cross-national database of micro data on health, demographic and socio-economic characteristics of persons aged 50 or higher resident in several European countries (Börsch-Supan et al., Reference Börsch-Supan, Brandt, Hunkler, Kneip, Korbmacher, Malter, Schaan, Stuck and Zuber2013). The sample studied here includes cross-sectional data from the second wave of the survey, carried out in 2006–2007, and retrospective material from SHARELIFE (Wave 3), carried out in 2008–2009. There were 25,052 respondents who participated at both the second and third waves; out of these, 23,768 (10,489 males and 13,279 females) were retained in the analysis, due to the requirement of complete data in the variables of interest. As the percentage of missing cases is around 5 per cent, application of imputation techniques was deemed unnecessary (Jakobsen et al., Reference Jakobsen, Gluud, Wetterslev and Winkel2017). Respondents originate in 13 countries, covering geographically most of Europe: Greece, Germany, Sweden, Netherlands, Spain, Italy, France, Switzerland, Denmark, Austria, Belgium, Czech Republic and Poland.
Measures
Dependent variable
Depression in SHARE is represented by the EURO-D scale (Beekman et al., Reference Beekman, Copeland and Prince1999; Prince et al., Reference Prince, Beekman, Deeq, Fuhrer, Kivela, Lawlor, Lobo, Magnusson, Meller, van Oyen, Reischies, Roelands, Skoog, Turrina and Copeland1999a, Reference Prince, Reischies, Beekman, Fuhrer, Jonker, Kivela, Lawlor, Lobo, Magnusson, Fichter, van Oyen, Roelands, Skoog, Turrina and Copeland1999b; Börsch-Supan and Jurges, Reference Börsch-Supan and Jurges2005), comprising 12 symptoms (depression, pessimism, suicidality, guilt, sleep, interest, irritability, appetite, fatigue, concentration, enjoyment and tearfulness). In the present study, a binary indicator based on that scale has been employed, comparing respondents reporting four or more depressive symptoms (having depression) to those reporting zero to three symptoms (no depression: reference category). This cut-off point has been provided by the SHARE team and has been validated psychometrically against several clinically relevant indicators at the EURODEP study while several analyses have also shown that it is appropriate and measures depression caseness (Dewey and Prince, Reference Dewey, Prince, Borsch-Supan, Brugiavini, Jurges, Mackenbach, Siegrist and Weber2005; Castro-Costa et al., Reference Castro-Costa, Dewey, Stewart, Banerjee, Hupper, Mendonca-Lima, Bula, Reisches, Wancata, Ritchie, Tsolaki, Mateos and Prince2007, Reference Castro-Costa, Dewey, Stewart, Banerjee, Huppert, Mendonca-Lima, Bula, Reisches, Wancata, Ritchie, Tsolaki, Mateos and Prince2008).
Independent variables
The independent variables cover circumstances over the lifecourse and correspond to three distinct periods: (a) childhood (age 10), (b) adulthood (events occurring after age 15 but well before the interview), and (c) later life (the year preceding the interview). Further, the events/circumstances considered pertain to three domains of life: health, socio-economic status (SES) and adverse experiences and are described in detail below.
Childhood predictors
Health status is represented by childhood self-perceived health (CSPH) at age 10 in binary form (Angelini et al., Reference Angelini, Klijs, Smidt and Mierau2016; Pakpahan et al., Reference Pakpahan, Hoffmann and Kröger2017a, Reference Pakpahan, Hoffmann and Krögerb; Arpino et al., Reference Arpino, Gumà and Julià2018); respondents having poor, fair and good CSPH (= 1) are compared to those having very good or excellent (= 0) CSPH. SES is based on the number of books the respondent had access to at age 10; this measure reflects the educational attainment of the parents (Cavapozzi et al., Reference Cavapozzi, Garrouste, Paccagnela, Börsch-Supan, Brandt, Hank and Schröder2011; Pakpahan et al., Reference Pakpahan, Hoffmann and Kröger2017a, Reference Pakpahan, Hoffmann and Krögerb; Van Bergen et al. Reference Van Bergen, van Zuijen, Bishop and de Jong2017; Arpino et al., Reference Arpino, Gumà and Julià2018) and contrasts individuals who had access to none or very few books to those who had access to at least ten books. Cognitive function is represented by the respondent's relative position in mathematics compared to his/her classmates at age 10 (Verropoulou and Zakynthinou, Reference Verropoulou and Zakynthinou2016): better, much better or about the same (= 0) versus worse and much worse (= 1). Finally, adverse experiences in childhood are represented by two binary variables reflecting circumstances when the respondent was 10 years old; the first variable denotes whether the respondent's parents drank heavily while the second one denotes whether they had mental health problems (Angelini et al., Reference Angelini, Klijs, Smidt and Mierau2016; Verropoulou and Zakynthinou, Reference Verropoulou and Zakynthinou2016). These two variables are used as proxies since other information on child abuse or neglect is unavailable in SHARE; further, there are analyses indicating that childhood maltreatment (physical or psychological) and other adverse experiences are often related to parental alcohol abuse or parental mental health problems (Dube et al., Reference Dube, Anda, Felitti, Croft, Edwards and Giles2001; Angelini et al., Reference Angelini, Klijs, Smidt and Mierau2016).
Adulthood predictors
Health is measured by an indicator of whether the respondent had experienced in the past a period of poor health as information on SPH is unavailable for that period. SES is based on educational attainment divided into two categories (0–12 years versus at least 13 years) and an indicator of whether the respondent had experienced a period of financial hardship; the latter has been used in past studies as a measure of SES (Conklin et al., Reference Conklin, Forouhi, Suhrcke, Surtees, Wareham and Monsivais2013; Amlaev, Reference Amlaev2015). Finally, adverse experiences are also represented by two indicators, a binary variable denoting whether the respondent had experienced a period of stress and another one showing whether he or she had experienced a period of hunger (Halmdienst and Winter-Ebmer, Reference Halmdienst and Winter-Ebmer2014).
Later-life predictors
Health status is represented by SPH (ranging from 1 to 5, i.e. from excellent to poor). Though SPH is a subjective indicator of health which might be influenced by mental health status, it exhibits strong associations with several aspects of physical health and has been found to be a strong predictor of mortality, even when controlling for morbidity and depression (Verropoulou, Reference Verropoulou2014). Further, it has been shown to be a strong predictor of concurrent depression and mental health (Aziz and Steffens, Reference Aziz and Steffens2013; Padayachey et al., Reference Padayachey, Ramlall and Chipps2017), while it also mediates the association between physical illness and depressive symptoms (Segel-Karpas, Reference Segel-Karpas2015). Health indicators also include mobility limitations in binary form (0 = less than three limitations, 1 = at least three limitations). SES is represented by whether the household was able to make ends meet the year preceding the survey with great or some difficulty (= 1) versus fairly easily or easily (= 0).
Control variables
All models control for age of the respondent at the time of the interview and welfare system. Regarding the latter variable, four welfare models are distinguished: the Nordic model which includes Sweden, Denmark and the Netherlands; the Continental model including Austria, France, Germany, Belgium and Switzerland; the Southern model including Italy, Spain and Greece; and the Central/Eastern model including Poland and the Czech Republic. In this way the estimates are adjusted for differentials related to the standards of living, social assistance, benefits and health care (Sengoku, Reference Sengoku2003; Norden, 2013; Popova and Kozhevnikova, Reference Popova and Kozhevnikova2013).
Statistical analysis
For the purposes of the analysis, three binary logistic regression models were applied separately by sex. The first model includes childhood and control variables only; the second model is an extension of the first one, additionally including adulthood predictors; and the third model is built on the second one and also incorporates later-life predictors. The sequential additive nature of these models allows an assessment of the relevant importance of circumstances pertaining to different stages of the lifecourse. More specifically, it allows an evaluation of how addition of adulthood conditions affects the predictive power of childhood circumstances on later-life depression and, second, of whether childhood and adulthood circumstances retain their importance after controlling for concurrent factors.
Subsequently, a decomposition technique was implemented via use of the KHB module, developed for the Stata 13 software (Kohler et al., Reference Kohler, Karlson and Holm2011). The method, apart from the direct effect of each childhood factor, provides estimates of their relative (or indirect) effect, which is mediated by adulthood and later-life circumstances. More specifically, the method involves the estimation of the direct effect of a variable of interest, based on a comprehensive model including all predictors (full or adjusted model), as well as the estimation of the total effect of that variable in a reduced or unadjusted model, which omits the influence of the specific mediator. The indirect effect is derived as the difference between the total and the direct effect. As the method allows for simultaneous use of multiple mediators, it can provide estimates of the indirect effect due to each of them. The overall confounding percentage, i.e. the proportion of the total effect of a childhood variable expressed through mediators, is computed as the overall indirect effect due to all mediators divided by the coefficient of the total effect. The percentage contribution of each mediator to the overall confounding percentage for a specific childhood variable is estimated as the ratio of the indirect effect of the respective mediator divided by the sum of all indirect effects of the mediators for that variable. These estimates are provided by the KHB module.
In fact, the decomposition method compares the estimated coefficients of two nested nonlinear probability models, thus allowing assessment of the degree that a variable mediates or explains the relationship of a predictor with the outcome (Kohler et al., Reference Kohler, Karlson and Holm2011; Breen et al., Reference Breen, Karlson and Holm2013), extending the decomposition properties of linear models to nonlinear probability models under the sequential ignorability assumption (Imai et al., Reference Imai, Keele and Tingley2010a, Reference Imai, Keele and Yamamoto2010b), according to which, given baseline covariates, mediators are selected randomly (Small, Reference Small2013). The decomposition estimates provided by the method are unbiased, as effects are rescaled, and more robust compared to logit models (Kendig et al., Reference Kendig, Gong, Yiengprugsawan, Silverstein and Nazroo2017; Berg et al., Reference Berg, Kalmijn and Leopold2018). Another advantage of the method is the simultaneous use of multiple mediators which allows disentangling the relative contribution of each of them to the indirect effect of a variable of interest (Arpino et al., Reference Arpino, Gumà and Julià2018). Further, the method also allows for the inclusion of variables that control for confounding influences on the decomposition.
Results
Descriptive analysis
Table 1 shows descriptive statistics for the total sample and by sex. The mean age of males in the sample is 66.4 years while for females it is 65.8 years. The percentage of women suffering from depression is double that for men (30.3% versus 14.9%). Regarding childhood, females report slightly worse CSPH and more perform worse in mathematics at age 10 compared to men. On the other hand, slightly more women than men had, at age 10, access to at least ten books. Concerning adverse experiences, percentages of parents drinking heavily are nearly equal for both genders while among women it is somewhat more frequent to have parents with mental health problems. In adulthood, women have experienced a period of stress (54.5%), a period of poor health (42.5%) and a period of financial hardship (34.5%) in higher proportions compared to men (46.9, 39.2 and 30.7%, respectively), whereas fewer have experienced a period of hunger; further, they have, on average, fewer educational qualifications. Concerning later life, women have worse SPH than men and a much higher proportion reports at least three mobility difficulties (27.0% compared to 14.4%). Finally, females experience greater financial hardship.
Table 1. Descriptive statistics for the total sample and by gender
Note: 11 = excellent, 5 = poor.
Logistic regression analysis
Tables 2 and 3 present the results of the three logistic regression models separately for males and females. In the first model for males (Table 2), which includes only childhood circumstances and controls, poor parental mental health is the factor having the greatest adverse effect on later-life depression. Nevertheless, all childhood circumstances are significant predictors and the associations point to the expected direction. In the second model, which additionally includes adulthood conditions, childhood circumstances remain very significant, though their effect is somewhat reduced. Adulthood conditions are also significant predictors of later-life depression; most important among them seems to be having experienced a period of poor health. Addition of later-life factors in the third model renders childhood health, educational attainment and having experienced a period of hunger in adulthood non-significant. Nevertheless, all other childhood and adulthood conditions remain important, especially poor parental mental health and having experienced poor health in adulthood. Later-life poor health and financial hardship exhibit a strong association with concurrent depression. It is worth noting that the substantial differentials between all welfare systems and the Nordic one are nearly wholly explained by later-life covariates.
Table 2. Odds ratios and 95 per cent confidence intervals (CI) for males
Notes: 11 = excellent, 5 = poor. Pseudo-R 2, showing the amount of variance of the outcome explained by the independent variables, indicates an increase in the explanatory power of Model 3 (R 2 = 0.241) compared to Model 1 (R 2 = 0.056) and Model 2 (R 2 = 0.107). Classification tables of the models show the percentage of cases of the dependent variable predicted correctly for Models 1, 2 and 3 and hence they are very high (85.10, 85.10 and 86.30%, respectively), this is an indication of a very good fit for all models. Ref.: reference category.
Significance levels: * p < 0.10, ** p < 0.05, *** p < 0.01.
Table 3. Odds ratios and 95 per cent confidence intervals (CI) for females
Notes: 11 = excellent, 5 = poor. Pseudo-R 2, showing the amount of variance of the outcome explained by the independent variables, indicates an increase in the explanatory power of Model 3 (R 2 = 0.260) compared to Model 1 (R 2 = 0.078) and Model 2 (R 2 = 0.131). Classification tables of the models show the percentage of cases of the dependent variable predicted correctly for Models 1, 2 and 3 and hence they are quite high (70.30, 71.40 and 75.60%, respectively) this is an indication of a quite good fit for all models.
Significance levels: * p < 0.10, ** p < 0.05, *** p < 0.01.
Considering females (Table 3), it seems that all childhood factors are very significant in the first model, though poor parental mental health is not as important. Addition of adulthood circumstances (Model 2) renders poor parental mental health non-significant; the remaining childhood conditions, however, maintain their importance as predictors of depression while all adulthood factors are significant. Inclusion of later-life health and SES do not change appreciably the existing relationships, though the effects of childhood and adulthood circumstances are reduced somewhat in most cases while the association with educational attainment, significant only at the 10 per cent level, points now to the opposite direction. Later-life covariates have a significant association with depression. It is worth noting that the significant differentiations observed between the Nordic and the other welfare systems, which are very pronounced in Model 1, especially regarding the Central/Eastern model, are reduced in Model 3 but remain substantial for women.
Comparing the findings for men and women there are some marked differences. Among men, childhood poor health has no significant impact in the third model whereas poor parental mental health maintains a strong effect. Βy contrast, among women the first factor maintains a significant impact while the latter predictor has a minor effect even in the first model. Further, financial hardship in adulthood and later life are more important among men whereas among women adulthood adverse experiences such as stress and hunger have a greater impact. Finally, differentials across welfare systems become non-significant for men in Model 3 but for women they remain important.
Decomposition analysis
Table 4 shows the decomposition of the total effects of selected childhood predictors on later-life depression into their direct and indirect parts separately by gender. Αs mediators all adulthood and later-life factors included in Model 3 have been considered, with the exception of educational attainment and concurrent SPH, which are included only as control variables. Preliminary analysis indicated that educational attainment does not mediate the effects of childhood circumstances. Moreover, concurrent SPH, in spite of having a strong mediating role (see Figures 1–4) was excluded from Table 4 on the grounds that, as it represents a subjective measure of health, it may reflect to some extent the outcome variable of depression. Finally, as the number of books at age 10 had only a non-significant indirect effect for both genders, it was included in these models only as control, along with age and welfare systems.
Figure 1. Percentage contribution of adulthood and later-life mediators to the indirect effect of ‘childhood self-perceived health’ for males and females, with and without later-life self-perceived health (SPH) as mediator.
Figure 2. Percentage contribution of adulthood and later-life mediators to the indirect effect of ‘relative position to others in mathematics at age 10’ for females, with and without later-life self-perceived health (SPH) as mediator.
Figure 3. Percentage contribution of adulthood and later-life mediators to the indirect effect of ‘parental excessive alcohol consumption’ for males and females, with and without self-perceived health (SPH) as mediator.
Figure 4. Percentage contribution of adulthood and later-life mediators to the indirect effect of ‘parents had mental health problems’ for males and females, with and without self-perceived health (SPH) as mediator.
Table 4. Coefficients of total, direct and indirect effects of childhood predictors, confounding percentages and percentage contribution of mediators by gender, based on the decomposition method
Notes: Bold values denote the mediators with the highest contribution for each childhood predictor. Confounding percentage is defined through the quantity ((βR − βF)/βR) × 100 = (βI/βR) × 100 where βR is the coefficient of the reduced model (total effect), βF is the coefficient of the full model (direct effect) and βI is the coefficient of the difference model (indirect effect). Alternatively, confounding percentage is the percentage of the total effect due to adulthood and later-life mediators. OR: odds ratio.
Significance levels: * p < 0.10, ** p < 0.05, *** p < 0.01.
In Table 4, apart from the indirect effect (i.e. the effect of a childhood factor mediated by adulthood and later-life conditions), coefficients for the reduced and the full models are shown, along with the respective confounding percentages. For instance, for males, fair or poor CSPH increases the log odds of later-life depression by 0.250 (reduced model coefficient); controlling for all mediators included in the model, the effect of CSPH declines to 0.149 (full model coefficient), indicating thus an indirect effect of 0.101 (their difference). In other words, the probability of an individual suffering from later-life depression increases by 25 percentage points for a standard deviation change in CSPH (reduced model); controlling for all mediators, this average increase is reduced to 14.9 percentage points (full model coefficient). Hence, poor, fair and good CSPH leads to higher values of the mediating variables, which is translated into an increase of the probability of depression in later life by 10.1 percentage points (the difference between the reduced and full model coefficients).
All indirect effects for both sexes are significant, with the exception of relative position in mathematics at age 10 for males. Among men, the childhood variable with the most significant indirect effect is poor parental mental health, which is mediated by adulthood and later-life factors by about 24 per cent; CSPH is mediated by 40 per cent and parental excessive alcohol consumption by 32 per cent. Among women, CSPH and poor parental mental health have the most significant indirect effects; these are mediated by post-childhood conditions by 42 and 58 per cent, respectively. Parental excessive alcohol consumption and relative position in mathematics at age 10 are mediated by about 22 per cent.
The percentage contribution of each mediator to the indirect effect of childhood predictors for men and women is presented in Figures 1–4. All figures include two panels; in the left panel the relative contribution is depicted for the models including later-life SPH only as control variable whereas in the right panel SPH is included as mediator. Regarding CSPH (Figure 1, left panel), it is mediated to a greater extent by adulthood rather than later-life circumstances, especially among men. The more important mediators for both genders are stress, which contributes about 30 per cent to the indirect effect, and poor health, which contributes roughly another 20 per cent. Financial hardship in adulthood is more important for men whereas concurrent conditions are fairly significant for women, especially mobility limitations. Inclusion of later-life SPH in the model as mediator shows that over 60 per cent of the indirect effect of CSPH may be attributable to that factor; the importance of all other mediators, especially of adulthood factors, is reduced drastically.
Relative position in mathematics at age 10 (Figure 2) has a significant indirect effect only for women. Τhis is mediated mainly by later-life circumstances; financial hardship contributes 33 per cent to the indirect effect while mobility limitations another 29 per cent. Of the adulthood conditions, poor health is the least important. Inclusion of SPH again reduces drastically the relative importance of all mediators, especially of adulthood conditions, while SPH has now the highest contribution to the indirect effect (44%).
Regarding adverse experiences in childhood, parental excessive alcohol consumption for males (Figure 3) is mediated mostly by later-life conditions whereas for female's adulthood circumstances are more important, especially having experienced a period of stress (42%). Mobility difficulties are also important for women while adulthood financial hardship is of consequence for men. Inclusion of SPH again indicates that it is the most substantial mediator.
Poor parental mental health (Figure 4) is mediated to a great extent by adulthood circumstances for both sexes: having experienced a period of stress and financial hardship contribute by 32 and 27 per cent, respectively, to the indirect effect for males, while stress and poor health contribute by 58 and 15 per cent, respectively, for females. Concurrent circumstances are fairly important for men only. Inclusion of SPH again reveals this mediator as the most significant for men (a contribution of 36%) but, for women, having experienced a period of stress still contributes the highest percentage to the indirect effect.
Discussion
The present study aimed to fill in a gap in the literature not only by considering in a comprehensive manner the direct and the indirect effects of childhood SES, health and adverse experiences on later-life depression, but also by estimating how the latter are mediated by adulthood and later-life events and conditions, for men and women separately, while also discussing potential mechanisms. To achieve these aims cross-sectional and retrospective data were combined from the SHARE study (Waves 2 and 3) and a decomposition technique was used (KHB module).
The descriptive analysis indicates that there is a greater prevalence of depression among females, a fact which is in accordance with past research (Hankin, Reference Hankin2002; Van de Velde et al., Reference Van de Velde, Bracke and Levecque2010; WHO, 2017a). Further, women experience more adversity than men in all three domains of life and in all three periods under consideration, with few exceptions (parental excessive alcohol consumption in childhood and having experienced a period of hunger in adulthood). Similar findings have been reported before (Zender and Olshansky, Reference Zender and Olshansky2009; McLean et al., Reference McLean, Asnaani, Litz and Hofmann2011; St Clair et al., Reference St Clair, Croudace, Dunn, Jones, Herbert and Goodyer2015).
The findings based on the consecutive addition of predictors, first those pertaining to childhood, then to adulthood and finally to later life (comprehensive model), indicate that: (a) though childhood circumstances have a significant effect on later-life depression, some predictors are mediated partly or fully by adulthood and later-life conditions; (b) several early, mid- and later-life factors have a strong and independent effect; (c) there are differentials by gender regarding the importance of different childhood circumstances as well as the way these are mediated.
Among men, childhood health is nearly fully mediated by adulthood and later-life circumstances, a fact consistent with the pathway model. By contrast, childhood SES and cognitive function remain significant predictors in the comprehensive model while their indirect effects are non-significant; hence, their effects are consistent with the critical period model. Finally, childhood adverse experiences retain a significant direct effect on later-life depression while they are, also, partly mediated by adulthood and later-life covariates, a fact in line with both models. Among women, poor parental mental health is nearly wholly mediated by adulthood and later-life covariates and is, thus, consistent with the pathway model. Childhood SES, just as for males, retains a significant direct effect in the comprehensive model while its indirect effect is insignificant, a fact in line with the critical period model. Finally, childhood health, cognitive function and excessive parental alcohol consumption retain a significant effect in the comprehensive model but are, also, partly mediated by adulthood and later-life conditions. These effects are thus consistent with both the critical period and the pathway models. Hence, our hypothesis that both mechanisms are relevant has been confirmed.
Regarding the direct effects of childhood circumstances, health is of greater significance for women whereas poor parental mental health is very important for men. Childhood SES status, on the other hand, seems equally important for both genders though adulthood and later-life SES has a greater effect among men. Adulthood adverse experiences (stress and hunger) are of greater consequence for women. With respect to the mediators, childhood health among men is mediated to a great extent by stress, poor health and financial hardship in adulthood while poor parental mental health is mediated by stress and financial hardship in adulthood as well as by mobility difficulties in later life. By contrast, parental excessive alcohol consumption is mediated mainly by later-life health and SES. Among women, childhood health is mediated largely by health in adulthood and later life as well as by stress in adulthood. Cognitive function is mediated mainly by later-life health and SES. Adverse experiences, on the other hand, are mediated among women to a great extent by stress in adulthood. Overall, regarding mediators, it seems that stress in adulthood is more important for women, whereas SES is of greater consequence among men. Hence, our hypothesis about gender differentials in the relative importance of childhood circumstances and how these are mediated has been confirmed. It has been suggested that women are more susceptible to stress due to hormonal changes and biological factors (Verma et al., Reference Verma, Balhara and Gupta2011; Albert, Reference Albert2015). Further, a greater effect of SES on later-life health among men has been observed before (Back and Lee, Reference Back and Lee2011; Verropoulou and Zakynthinou, Reference Verropoulou and Zakynthinou2016).
The findings of the present study stand out compared to other similar analyses as most of them focus solely on the effects of socio-economic adversity on later-life health or examine the effects of SES in conjunction with childhood health. By contrast, the present study additionally considers adverse childhood experiences. Our findings regarding the effects of childhood health on later-life depression among women are roughly in agreement with Kendig et al. (Reference Kendig, Gong, Yiengprugsawan, Silverstein and Nazroo2017) and Pakpahan et al. (Reference Pakpahan, Hoffmann and Kröger2017a), who also find that both the critical period and the pathway models apply regarding effects on later-life SPH. However, our results contrast with those of Arpino et al. (Reference Arpino, Gumà and Julià2018), who suggest that these effects are solely direct and are not mediated. Regarding childhood SES, our findings are fairly consistent with Tani et al. (Reference Tani, Fujiwara, Kondo, Noma, Sasaki and Kondo2016), who find latency mechanisms in operation, and with Zimmer et al. (Reference Zimmer, Hanson and Smith2016), who find a strong and independent effect of both early and adult SES on later-life depression; however, they contrast with analyses suggesting pathway mechanisms regarding later-life SPH or depression (Torres and Wong, Reference Torres and Wong2013; Pakpahan et al., Reference Pakpahan, Hoffmann and Kröger2017a; Angelini et al., Reference Angelini, Howdon and Mierau2019). Finally, a strong and independent effect of early life and adulthood adverse experiences on adult poor mental health has been noted before (Nurius et al., Reference Nurius, Green, Logan-Greene and Borja2015).
Some limitations of the study should be considered when interpreting the findings. First, depressive symptoms and all independent variables included in the analysis are self-reported and, thus, they may be subject to misreporting. Additionally, retrospective material may be affected by recall errors. Hence, it would be of great interest to confirm these findings using longitudinal material from a cohort study, based on interviews carried out in childhood, adulthood and later life. Unfortunately, such studies are scarce and there is currently none covering several European countries. Second, information on adverse childhood circumstances is based on proxy variables (excessive parental alcohol consumption and poor parental mental health). Other measures of child abuse/neglect (unavailable in SHARE) would have provided more reliable conclusions. Third, there is no available information on the onset of depression; early onset may exhibit a stronger association with early life factors. Fourth, the present analysis does not control for biological and hereditary factors predisposing to depression, as the relevant information is not included in SHARE. Further analyses should attempt to control for such factors, thus enabling more accurate assessment of the relative effects of childhood disadvantage on depression in later life. Finally, around 5 per cent of the respondents were excluded from the analysis due to missing information in the variables of interest. These persons are on average slightly older, include more females, have somewhat worse health and are slightly more disadvantaged compared to the sample used in the study. Hence, their exclusion from the analysis may have led to a slight underestimation of the impact of adversity on later-life depression.
Conclusions
Despite the limitations, the present study provides evidence of the long-term impact of childhood health, SES and adverse experiences on later-life depression for both men and women. Most childhood factors have a significant direct and indirect effect, the latter mediated by adulthood and later-life conditions; SES for both genders and cognitive function for men exhibit a direct effect only, while childhood health for males and poor parental mental health for females have only an indirect effect. Hence, the findings indicate that both the critical period and the pathway models are relevant. Regarding gender differentials in mediation, adulthood and later-life SES mediate early life health and adverse experiences more strongly for men, while stress seems to mediate early life adverse experiences to a greater extent among women.
Intervening policies aiming to reduce the incidence of depression should first seek to address childhood adversity, considering at the same time the differential vulnerability of men and women. Putting emphasis on social and psychological support for abused and maltreated children as well as for those experiencing material deprivation might help to prevent depression both in adulthood and in later life.
Future research could seek to validate these findings based on longitudinal data, additionally distinguishing between early and late onset of depression. Further, it would be worth examining interactions between the three domains of life (health, SES and adverse experiences) and to construct trajectories over the lifecourse. Finally, it would be of interest to examine whether associations differentiate across European welfare systems and countries.
Data
This paper uses data from SHARE Waves 2 and 3 (SHARELIFE) (DOIs: 10.6103/SHARE.w2.600, 10.6103/SHARE.w3.600), see Börsch-Supan et al. (Reference Börsch-Supan, Brandt, Hunkler, Kneip, Korbmacher, Malter, Schaan, Stuck and Zuber2013) for methodological details. The SHARE data collection has been primarily funded by the European Commission through FP5 (QLK6-CT-2001-00360), FP6 (SHARE-I3: RII-CT-2006-062193, COMPARE: CIT5-CT-2005-028857, SHARELIFE: CIT4-CT-2006-028812) and FP7 (SHARE-PREP: No. 211909, SHARE-LEAP: No. 227822, SHARE M4: No. 261982). Additional funding from the German Ministry of Education and Research, the Max Planck Society for the Advancement of Science, the US National Institute on Aging (U01_AG09740-13S2, P01_AG005842, P01_AG08291, P30_AG12815, R21_AG025169, Y1-AG-4553-01, IAG_BSR06-11, OGHA_04-064, HHSN271201300071C) and from various national funding sources is gratefully acknowledged (see www.share-project.org).
Financial support
The second author has been supported by the General Secretariat for Research and Technology (ES) and the Hellenic Foundation for Research and Innovation (ES, Scholarship Code 991).
Conflict of interest
The authors declare no conflicts of interest.
