Introduction
Depression is a common mental health problem among women of childbearing age, with prevalence rates ranging from 10% to 32% (Ertel et al. Reference Ertel, Rich-Edwards and Koenen2011; Wang et al. Reference Wang, Wu, Anderson and Florence2011). The impact of depression is significant not only for women's quality of life (Darcy et al. Reference Darcy, Grzywacz, Stephens, Leng, Clinch and Arcury2011), but also for their children's developmental outcomes (Ertel et al. Reference Ertel, Rich-Edwards and Koenen2011; Turney, Reference Turney2012). Most information on the occurrence of maternal depression comes from studies conducted during the perinatal period (pregnancy to 12 months after delivery) (Horwitz et al. Reference Horwitz, Briggs-Gowan, Storfer-Isser and Carter2009), although symptoms may persist or re-emerge later (Essex et al. Reference Essex, Klein, Miech and Smider2001; Goodman, Reference Goodman2007), setting women on a course of chronic depressive symptoms which have especially deleterious consequences for children (Fihrer et al. Reference Fihrer, McMahon and Taylor2009; Connelly et al. Reference Connelly, Baker-Ericzen, Hazen, Landsverk and Horwitz2010; Turney, Reference Turney2012).
Risk factors for maternal depression (such as socioeconomic disadvantage, a prior history of mental disorders, stressful life events, poor social support, domestic violence, and pregnancy/obstetric complications) have been well examined in the perinatal period (Koleva et al. Reference Koleva, Stuart, O'Hara and Bowman-Reif2011; O'Hara & McCabe, Reference O'Hara and McCabe2013; Seimyr et al. Reference Seimyr, Welles-Nyström and Nissen2013). Yet, little is known about predictors of persistent depression in mothers. With few exceptions (Seto et al. Reference Seto, Cornelius, Goldschmidt, Morimoto and Day2005; Horwitz et al. Reference Horwitz, Briggs-Gowan, Storfer-Isser and Carter2009; Skipstein et al. Reference Skipstein, Janson, Stoolmiller and Mathiesen2010, Reference Skipstein, Janson, Kjeldsen, Nilsen and Mathiesen2012; Wang et al. Reference Wang, Wu, Anderson and Florence2011; Giallo et al. Reference Giallo, Cooklin and Nicholson2014; Woolhouse et al. Reference Woolhouse, Gartland, Mensah and Brown2015), longitudinal studies examining chronic mental health difficulties in women beyond the second-year postpartum are scarce. These studies found that depressive symptoms are disproportionately likely to persist in women who are young, belong to an ethnic minority group, are unemployed, have low social support, experience high parenting stress, and stressful life events. However, most initial data collection was conducted 10–20 years ago and there is need for contemporary data on this topic. With the exception of the work by Woolhouse et al. (Reference Woolhouse, Gartland, Mensah and Brown2015), none of these studies included depressive symptoms occurring in pregnancy, even though these are prevalent (12.4%) and associated with risk of later depression (Banti et al. Reference Banti, Mauri, Oppo, Borri, Rambelli, Ramacciotti, Montagnani, Camilleri, Cortopassi, Rucci and Cassano2011; Gaillard et al. Reference Gaillard, Le Strat, Mandelbrot, Keita and Dubertret2014; Vliegen et al. Reference Vliegen, Casalin and Luyten2014).
In order to gain better understanding of women's longitudinal patterns of depression and associated early risk factors we focused on antenatal predictors for persistent maternal depression from pregnancy onwards. We used data from a community-based sample of mothers followed from pregnancy up to when the child was 5 years of age to (1) determine symptom trajectories; (2) identify sociodemographic, psychosocial and psychiatric predictors of maternal depression trajectories present before or during pregnancy.
Method
Participants
Data for this study come from the EDEN mother–child study, set up to assess the pre- and postnatal nutritional, social, and environmental determinants of infant and child development and health (Drouillet et al. Reference Drouillet, Forhan, De Lauzon-Guillain, Thiebaugeorges, Goua, Magnin, Schweitzer, Kaminski, Ducimetiere and Charles2009). Pregnant women were recruited before 24 weeks of gestation from two maternity wards (Poitiers and Nancy University hospitals in France) between September 2003 and January 2006. Exclusion criteria were multiple pregnancies, history of diabetes, inability to speak and read French or plans to move out of the study region within the next 3 years. Among eligible women, 55% agreed to participate. Of the 2002 women recruited during pregnancy, birth data were available for 1899 mother–infant pairs. From pregnancy onwards, mothers and children were followed nine times (pregnancy, birth, 4, 8, 12, 24 months, 3, 4 and 5 years) via face-to-face or self-completed questionnaires completed by the mothers. Data on the child's birth characteristics were collected directly from medical records.
By the year 5 follow-up, data on maternal depression were available for 1190 participants. Attrition rates were highest for mothers who were of non-French origin, young, had low educational level, and at baseline were single, unemployed, had financial difficulties or experienced other life events, had low social support, reported childhood adversity, used tobacco, had a history of mental health problems. The study was approved by the Comité Consultatif de Protection des Personnes dans la Recherche Biomedicale (Ethics Committee, Kremlin Bicêtre Hospital) and by the Commission Nationale de l'Informatique et des Libertés [National Committee for Processed Data and Freedom (CNIL)]. Written consent was obtained from the mother for herself at inclusion and for her newborn child after delivery.
Measures and procedures
Maternal symptoms of depression
Depressive symptoms during pregnancy and at 3 and 5 years follow-ups were assessed using the Center for Epidemiological Studies Depression (CES-D) questionnaire (Radloff, Reference Radloff1977). This previously validated 20-item questionnaire measures the number of depressive symptoms over the past week (score range 0–60) (Fuhrer & Rouillon, Reference Fuhrer and Rouillon1989; Joiner et al. Reference Joiner, Walker, Pettit, Perez and Cukrowicz2005). A threshold of ⩾16 is commonly used to identify individuals at risk for clinical depression.
Depressive symptoms during the first year after the child's birth (4, 8 and 12 months postpartum) were assessed using the Edinburgh Postnatal Depression Scale (EPDS), a 10-item questionnaire designed to detect postnatal depression (score range 0–30) (Cox et al. Reference Cox, Holden and Sagovsky1987). In this study we used the cut-off ⩾12, which has been recommended for the French translation and has demonstrated its validity for research purposes (Guedeney & Fermanian, Reference Guedeney and Fermanian1998; Teissedre & Chabrol, Reference Teissedre and Chabrol2004).
To identify trajectories of maternal depressive symptoms we needed to meaningfully combine the scores of both instruments, which have different possible symptom severity score ranges. Thus, the scores for each instrument were standardized to t scores (mean = 50, s.d. = 10), which allowed us to study them jointly as continuous measures.
Predictors
Based on the scientific literature we identified several sociodemographic, psychosocial and psychiatric characteristics potentially associated with maternal depression trajectories.
Sociodemographic characteristics were: study centre (Poitiers v. Nancy), age (dichotomized at the sample mean <30 v. ⩾30 years), country of birth (France v. other), family situation (single v. married/cohabitating), number of children, maternal educational level (<12 v. ⩾12 years), maternal employment at the beginning of pregnancy (yes v. no), partner's educational level (<12 v. ⩾12 years), partner's employment (yes v. no), financial difficulties (any of the following (yes v. no); difficulties to feed or clothe the family, difficulties to pay the rent, heating or electricity, regular overdraft of bank account, subsided health coverage). Psychosocial characteristics included negative life events that were assessed using a 17-item measure including major life events such as moving house, separation or divorce, death or serious illness of a close friend or family member, or legal troubles. Endorsed items were totalled to create an overall score (hypothetical range 0–17). As a large part of the sample (80%) did not experience any events, scores were dichotomized for any life event occurrence (yes v. no). Work-related stress and work-related over-commitment were measured with the Effort-Reward Imbalance (ERI) scale (Siegrist et al. Reference Siegrist, Starke, Chandola, Godin, Marmot, Niedhammer and Peter2004). The standard ERI questionnaire comprises 46 items measuring efforts (including work obligations and physical demands of the job) and rewards (including financial compensation, job security, personal and social recognition). Responses for each scale are summed and a categorical score for effort-reward imbalance is calculated (yes v. no). The over-commitment scale includes items such as ‘As soon as I get up in the morning I start thinking about work problems’, and ‘People close to me say that I sacrifice too much for my job’. These are scored on a four-point response scale, giving a range from 6 to 24; this scale is divided into the upper third of the distribution versus the rest (yes v. no). Additional psychosocial risks included at least one maternal childhood adversity experienced prior to age 18 years (material deprivation, parental conflict or violence, child abuse or neglect, placement out of home; yes v. no), partner's substance use (tobacco, alcohol or cannabis use; yes v. no), social support (yes v. no). Psychiatric characteristics included: childhood behaviour problems (yes v. no), previous mental health problems (yes v. no), pre-pregnancy mental health treatment (yes v. no), pre-pregnancy substance use (tobacco, alcohol or cannabis use; yes v. no), and anxiety in pregnancy [State-Trait Anxiety Inventory (STAI) score (Spielberger et al. Reference Spielberger, Gorsuch, Lushene, Vagg and Jacobs1983] dichotomized at the 80th centile, <17 v. ⩾18).
Statistical analyses
Our aim was to identify trajectories of maternal depressive symptoms and associated predictors. We conducted a trajectory analysis, a semi-parametric group-based modelling strategy used to identify homogeneous latent trajectory classes based on longitudinal data (Nagin, Reference Nagin1999, Reference Nagin2005).
Using the six scores of maternal depression available to us, we modelled maternal trajectories of depression using growth trajectory models (proc traj in SAS v. 9.3; SAS Institute Inc., USA) under the censored normal distribution (Jones et al. Reference Jones, Nagin and Roeder2001). Missing data were handled by proc traj under the missing-at-random assumption, where individuals with missing information were assigned to their most likely group (Jones & Nagin, Reference Jones and Nagin2007). The number and shapes of latent trajectories were determined using the Bayesian Information Criterion (BIC) and refined by setting the trajectories’ order (i.e. linear, quadratic, cubic). Using backward elimination of higher-order trajectories, a more parsimonious model was maintained whenever a higher-order growth coefficient did not reach statistical significance and the BIC value decreased. To define a good model, average posterior probabilities of trajectory membership should be at least equal to 0.7 for all groups (Nagin, Reference Nagin2005).
In the second stage of analysis, we used multinomial logistic regression analyses to test associations between predictors and trajectory groups (using the never depressed group as the reference). Variables were included in the final multivariate model if they were associated with class membership at p < 0.10 in bivariate logistic regression models. Our antenatal predictors of maternal depression trajectories were assessed at study baseline, resulting in very few covariates with missing data levels of > 2% (overall: 2.23%). Exceptions were partner's substance abuse (9.5% of data missing), work-related stress and work-related over-commitment (4.1% and 4.0% missing, respectively). Analyses were conducted with SPSS v. 20 (IBM SPSS Statistics, USA).
Ethical standards
The authors assert that all procedures contributing to this work comply with the ethical standards of the relevant national and institutional committees on human experimentation and with the Helsinki Declaration of 1975, as revised in 2008.
Results
Maternal depression trajectories
The study sample included 1807 mothers with sufficient data to estimate depression trajectories.
The optimal number of maternal depression trajectories in our study was five. Average posterior probabilities of group membership were above 0.80 for all groups. The five trajectories of maternal depressive symptoms were as follows (Fig. 1): 60.2% (n = 1087) of mothers had no symptoms; 4.7% (n = 85) had high symptoms in pregnancy only; 4.9% (n = 89) had high symptoms in the child's preschool period only; 25.2% (n = 456) had persistent intermediate-level symptoms and 5.0% (n = 90) of mothers had persistent high-level symptoms. As expected, participants’ symptoms at each wave of measurement were associated with the trajectory of depressive symptoms. Scores for persistent high-level symptoms were well above clinical cut-off scores for both depression measures (Table 1).
Fig. 1. Trajectories of maternal symptoms of depression by child's age in months in the EDEN cohort study (N = 1807, 2003–2011, France).
Table 1. Unstandardized depressive symptoms by trajectories of maternal depressive symptoms in the EDEN cohort study [N = 1807, 2003–2011, France, mean (s.d.)]
CES-D, Center for Epidemiological Studies Depression scale. Clinical cut-off score ⩾16.
EPDS, Edinburgh Postnatal Depression Scale. Clinical cut-off score ⩾12.
Sample descriptives
Table 2 presents participants’ sociodemographic, psychosocial and psychiatric characteristics. Mothers were on average 30 years old at the time of the child's birth, mostly of French origin (95.7%), 92.7% were married or cohabitating and had on average one child; 54.2% had at least 2 years of higher education, 83.9% were in paid employment at the beginning of pregnancy, 12.7% reported financial difficulties, 19.8% experienced at least one negative life event during pregnancy and 29.3% reported childhood adversities. Work stress was reported by 14.6% and work overinvestment by 32.5%. Fifty-three per cent had a partner who used a psychoactive substance (mostly tobacco), while 3.8% experienced a lack of social support. Seven percent reported childhood behaviour problems and 10.7% previous mental health problems; 9.6% received pre-pregnancy mental health treatment, 36.9% reported pre-pregnancy substance use and 21.1% of mothers had symptoms of anxiety.
Table 2. Sociodemographic, psychosocial and psychiatric characteristics of EDEN cohort study participants, N = 1807, 2003–2011, France [n (%) or mean (s.d.) for continuous variables]
STAI, State-Trait Anxiety Inventory.
Predictors of maternal depression trajectories
Most sociodemographic, psychosocial and psychiatric characteristics were associated with depression trajectories, with the exception of maternal age and family situation (Table 2).
Table 3 presents the results [adjusted odds ratios (aORs)] of multivariate regression analyses examining predictors of maternal depression trajectory. Compared to the no symptoms group, variables that were associated with depressive symptoms in pregnancy only included non-French origin [aOR 4.94, 95% confidence interval (CI) 1.35–18.07], experience of childhood adversities (aOR 2.49, 95% CI 1.33–4.68), lack of social support (aOR 4.31, 95% CI 1.52–12.25), history of mental health problems (aOR 2.31, 95% CI 1.02–5.22), pre-pregnancy substance use (aOR 2.27, 95% CI 1.17–4.39) and anxiety during pregnancy (aOR 11.97, 95% CI 6.42–22.31). Depressive symptoms in the child's preschool period only were associated with partner's low educational level (aOR 1.99, 95% CI 1.09–3.65), pre-pregnancy mental health treatment (aOR 2.58, 95% CI 1.18–5.65) and anxiety during pregnancy (aOR 3.89, 95% CI 2.17–6.99). Persistent intermediate-level depressive symptoms were associated with the experience of life events during pregnancy (aOR 1.47, 95% CI 1.03–2.09), work overinvestment (aOR 2.04, 95% CI 1.49–2.78), pre-pregnancy mental health treatment (aOR 1.78, 95% CI 1.09–2.93) and anxiety during pregnancy (aOR 3.33, 95% CI 2.35–4.71). Finally, persistent high depressive symptoms were associated with non-French origin (aOR 5.92, 95% CI 1.86–18.86), childhood adversities (aOR 4.09, 95% CI 2.16–7.75), work overinvestment (aOR 4.78, 95% CI 2.23–10.23), history of mental health problems (aOR 3.78, 95% CI 1.82–7.85) and anxiety during pregnancy (aOR 7.94, 95% CI 4.29–14.69).
Table 3. Sociodemographic, psychosocial and psychiatric predictors of maternal depression trajectories: EDEN cohort study, N = 1334, 2003–2011, France adjusted OR (95% CI), multivariate polynomial logistic regression analysis
aOR, Adjusted odds ratio; CI, confidence interval.
In additional analyses we tested whether the aORs of predictors associated with several trajectories significantly differed. This was the case for work overinvestment (aOR for persistent high-level symptoms significantly higher than that for persistent intermediate-level symptoms, p = 0.03), and for anxiety [aORs for symptoms in pregnancy only significantly higher than for symptoms in preschool only (p = 0.005) and persistent intermediate-level symptoms (p < 0.0001), but not for persistent high-level symptoms (p = 0.304)].
Discussion
Main findings
Using data from a large community-based birth cohort study, we identified five distinct trajectory groups of maternal symptoms of depression from pregnancy to the child's 5th birthday (no symptoms: 60.2%; high symptoms in pregnancy: 4.7%; high symptoms in the child's preschool period: 4.9%; persistent intermediate-level symptoms: 25.2%; and persistent high-level symptoms: 5.0%).
Depression trajectories, particularly persistent symptoms, appeared especially strongly associated with women's psychosocial and psychiatric characteristics. Particularly, women who are anxious or who experience stressful work conditions during pregnancy may have an increased probability of persistent depressive symptoms over time. Women who present these characteristics may require specific monitoring during pregnancy and afterwards.
Strengths and limitations
The current study has several strengths, e.g. (a) a large community sample, (b) longitudinal and repeated assessments of multiple demographic, psychosocial and psychiatric characteristics, and (c) the use of validated measures of maternal mental health. However, we also acknowledge limitations. First, maternal depressive symptoms were self-reported, using two different measures. However, both the CES-D and the EPDS, which we standardized to ensure comparability, have previously been validated and we were able to study the entire depression spectrum rather than diagnoses, which may be a closer estimate of variations in symptomatology at the population level and show higher reliability and validity (Markon et al. Reference Markon, Chmielewski and Miller2011). Second, we only included baseline predictors, possibly underestimating the influence of factors that can change over time (e.g. financial difficulties increased from 12.7% at baseline to 22.2% at 5-year follow-up, 92.7% of women were married/cohabitating compared to 88.5% after 5 years). Nevertheless, this approach allowed us to identify early risk factors that may set women on depressive symptom trajectories from pregnancy onwards. In addition, including baseline predictors might have led to inflation of the associations found between the high symptoms in pregnancy trajectory and possible risk factors due to shared method variance. However, most of the associated sociodemographic, psychosocial and psychiatric risk factors preceded the occurrence of depressive symptoms during pregnancy and are thus unlikely to be biased in this way. Finally, the present study sample is not nationally representative. Compared with a national perinatal survey performed in 2003 on a representative sample of French women (Blondel et al. Reference Blondel, Supernant, Du Mazaubrun and Breart2006) the EDEN study was similar in age and proportion of unmarried couples but included a larger proportion of women with university-level education (53% v. 43%) (Drouillet et al. Reference Drouillet, Kaminski, De Lauzon-Guillain, Forhan, Ducimetiere, Schweitzer, Magnin, Goua, Thiebaugeorges and Charles2008). This is further increased by selective attrition among socially disadvantaged women. As it is likely that this group has a higher proportion of maternal depression (Ertel et al. Reference Ertel, Rich-Edwards and Koenen2011), our results are plausibly an under- rather than an over-estimate of the relationship between socioeconomic predictors and maternal depression trajectories. Replication of our findings in a population including a higher proportion of disadvantaged, high-risk families is warranted.
Maternal depressive symptom trajectories
Exploring maternal depressive symptoms and associated predictors during the early childhood period is important to improve our understanding of the variation in symptom severity and persistence and can clarify why the trajectories of some individuals differ from those of others. Using a trajectory modelling approach we were able to gain insight into five distinct patterns of depressive symptoms from pregnancy onwards. The majority (60.2%) of mothers reported no or very few depressive symptoms across follow-up. This finding is consistent with previous studies, suggesting that most mothers do not experience elevated levels of depressive symptoms during and after the perinatal period (Mora et al. Reference Mora, Bennett, Elo, Mathew, Coyne and Culhane2009; Sutter-Dallay et al. Reference Sutter-Dallay, Cosnefroy, Glatigny-Dallay, Verdoux and Rascle2012).
We identified two groups of women with a marked increase in symptom scores in specific periods. As previously described, a small group of mothers (4.7%) appear to experience depressive symptoms only during pregnancy (Mora et al. Reference Mora, Bennett, Elo, Mathew, Coyne and Culhane2009; Vänskä et al. Reference Vänskä, Punamäki, Tolvanen, Lindblom, Flykt, Unkila-Kallio, Tiitinen, Repokari, Sinkkonen and Tulppala2011). This may partly be due to measurement issues, as it is sometimes difficult to distinguish depressive symptoms from normal somatic symptoms experienced in pregnancy (e.g. changes in appetite and weight, sleep and energy). It has been suggested that, particularly during pregnancy, depression is best assessed by studying cognitive-affective symptoms (Pereira et al. Reference Pereira, Marques, Soares, Maia, Bos, Valente, Nogueira, Roque, Madeira and Macedo2014). The CES-D, used to ascertain depressive symptoms in pregnancy in the EDEN study, mostly covers cognitive symptoms, therefore our measure is probably only moderately affected by such bias. It may be that a unique subgroup of women predominantly suffers from prenatal depression, which resolves postnatally.
A third group of women (4.9%) appear to have high symptoms in the child's preschool period only. Although the prevailing view is that women are especially vulnerable to depression in the perinatal period, maternal depression may actually be more common 4–5 years after a birth than at any time in the first 12 months postpartum (Najman et al. Reference Najman, Andersen, Bor, O'Callaghan and Williams2000; Woolhouse et al. Reference Woolhouse, Gartland, Mensah and Brown2015). This may be because child-rearing makes greater demands on the mother's mental health than the perinatal period itself (Betts et al. Reference Betts, Williams, Najman and Alati2014). Contrary to some previous studies, we did not identify a group of women with symptoms exclusively in the postpartum period. It may be that a significant portion of women with postpartum-onset depression remain depressed beyond the child's first year, which would have lead them to them being assigned to one of the groups with persistent symptoms.
Indeed, in both groups of chronic depressive symptoms we noted a steady increase in symptom severity between 4 and 12 months postpartum. Thus, we identified a substantial group of mothers (25.2%) with intermediate-level depressive symptoms throughout follow-up. Finally, consistent with previous research, a small group (5.0%) of women had persistently high depressive symptoms, which were probably above a clinically significant threshold (Campbell et al. Reference Campbell, Matestic, von Stauffenberg, Mohan and Kirchner2007; Mora et al. Reference Mora, Bennett, Elo, Mathew, Coyne and Culhane2009; Vänskä et al. Reference Vänskä, Punamäki, Tolvanen, Lindblom, Flykt, Unkila-Kallio, Tiitinen, Repokari, Sinkkonen and Tulppala2011; Cents et al. Reference Cents, Diamantopoulou, Hudziak, Jaddoe, Hofman, Verhulst, Lambregtse-van den Berg and Tiemeier2013). Although we could not determine how many of these women had diagnosable major depression, their CES-D and EPDS scores were well above established cut-off scores, indicating that their symptoms fall within a clinical range. Our results indicate that these chronic depression trajectories may already start in pregnancy and persist over time.
Predictors of maternal depressive symptom trajectories
Maternal depressive symptoms often occur within a context of personal and environmental risk, and the sociodemographic, psychosocial and psychiatric predictors we studied were shared across various trajectories. The only predictor associated with an increased likelihood of depressive symptoms at any moment during follow-up was anxiety during pregnancy, with especially high associations with depression in pregnancy (aOR 11.97) and persistently high symptoms (aOR 7.94). Anxiety symptoms are frequently reported by pregnant women and often considered by midwives and obstetricians as part of the normal psychic experiences of pregnancy (Goodman & Tyer-Viola, Reference Goodman and Tyer-Viola2010; Dunkel Schetter & Tanner, Reference Dunkel Schetter and Tanner2012). As depression and anxiety during pregnancy are moderately to highly co-morbid (Heron et al. Reference Heron, O'Connor, Evans, Golding and Glover2004; Wynter et al. Reference Wynter, Rowe and Fisher2013), it may be difficult to clearly attribute symptoms to one or the other condition (Skouteris et al. Reference Skouteris, Wertheim, Rallis, Milgrom and Paxton2009). Yet, there is increasing evidence to suggest that antenatal anxiety symptoms form a clinical entity of their own, possible forming a cycle of co-morbidity, whereby initial levels of depressive symptoms lead to higher levels of anxiety, which in turn may then predict higher depressive symptoms (Sutter-Dallay et al. Reference Sutter-Dallay, Giaconne-Marcesche, Glatigny-Dallay and Verdoux2004; Skouteris et al. Reference Skouteris, Wertheim, Rallis, Milgrom and Paxton2009). Recent studies have suggested that high levels of maternal anxiety during pregnancy could, independently from prenatal depression, predict an elevated risk of long-term depressive symptoms (Horwitz et al. Reference Horwitz, Briggs-Gowan, Storfer-Isser and Carter2009; Vänskä et al. Reference Vänskä, Punamäki, Tolvanen, Lindblom, Flykt, Unkila-Kallio, Tiitinen, Repokari, Sinkkonen and Tulppala2011; Sutter-Dallay et al. Reference Sutter-Dallay, Cosnefroy, Glatigny-Dallay, Verdoux and Rascle2012). However, to our knowledge previous studies had a shorter follow-up than our investigation. We are not aware of studies indicating that the impact of prenatal anxiety lasts even up to 5 years after birth.
Other variables were more uniquely associated with particular trajectory classes, with women with depressive symptoms in pregnancy only or persistently high depressive symptoms presenting the largest number of risk factors. This suggests that the cumulative effect of multiple risk factors may well increase the chronicity of maternal depressive symptoms, especially since overall there were very few predictor variables that were significantly more strongly related to one trajectory than any other. A specific predictor for depression in pregnancy was a lack of social support. Psychosocial and emotional resources are needed to adapt to the significant life changes associated with pregnancy and the early parenting period (Manuel et al. Reference Manuel, Martinson, Bledsoe-Mansori and Bellamy2012). Insufficient support is negatively associated with maternal psychological well-being during pregnancy and increases the risk of depression, as well as worse health and pregnancy outcomes (Robertson et al. Reference Robertson, Grace, Wallington and Stewart2004; Orr, Reference Orr2004; Elsenbruch et al. Reference Elsenbruch, Benson, Rucke, Rose, Dudenhausen, Pincus Knackstedt, Klapp and Arck2007). Women reporting any substance use (tobacco, alcohol or cannabis) before pregnancy also had an elevated likelihood (aOR 2.27) for depressive symptoms during pregnancy. Many expectant mothers follow recommendations to maintain a healthy lifestyle during pregnancy, and they may have stopped or reduced their consummation (Viau et al. Reference Viau, Padula and Eddy2002). However, reducing or eliminating substance use by pregnant women has only inconsistently been found to be related to prenatal depressive symptoms (Zhu & Valbo, Reference Zhu and Valbo2002; Solomon et al. Reference Solomon, Higgins, Heil, Badger, Mongeon and Bernstein2006). It is likely that women who continued their substance use in pregnancy have an increased vulnerability to mental health problems (Farrell et al. Reference Farrell, Howes, Bebbington, Brugha, Jenkins, Lewis, Marsden, Taylor and Meltzer2003; Le Strat et al. Reference Le Strat, Dubertret and Le Foll2011).
Concerning the predictors of persistently elevated depressive symptoms in mothers from pregnancy onwards, previous studies have reported that maternal depression is associated with maternal age, income, relationship and employment status and educational attainment (e.g. Seto et al. Reference Seto, Cornelius, Goldschmidt, Morimoto and Day2005; Sutter-Dallay et al. Reference Sutter-Dallay, Murray, Dequae-Merchadou, Glatigny-Dallay, Bourgeois and Verdoux2011; Seimyr et al. Reference Seimyr, Welles-Nyström and Nissen2013). Contrary to these reports, we found that sociodemographic characteristics were not associated with depressive symptoms, with the exception of non-French origin. Migrant women had a high level of symptoms in pregnancy (aOR 4.94) or persistently high symptoms (aOR 5.92). Pregnant migrant women may go unidentified and untreated due to under-utilization of health services, and therefore, are at increased risk of persistent mental health difficulties (Lindert et al. Reference Lindert, Schouler-Ocak, Heinz and Priebe2008; Giallo et al. Reference Giallo, Cooklin and Nicholson2014). However, as previously indicated, the role of socioeconomic predictors might be underestimated due to the relatively small proportion of socially disadvantaged women in our sample.
Psychosocial risk factors for persistent depression were the experience of childhood adversity. Our finding extends previous research that mostly focused on child abuse (Leigh & Milgrom, Reference Leigh and Milgrom2008; Clark et al. Reference Clark, Caldwell, Power and Stansfeld2010). Additionally, we found an increased likelihood of depression in women who encountered life stressors during pregnancy. While pregnancy and birth are often regarded as stressful in their own right, many women have additional stressful experiences during pregnancy (Robertson et al. Reference Robertson, Grace, Wallington and Stewart2004; Giallo et al. Reference Giallo, Cooklin and Nicholson2014), which are consistently shown to be associated with women's increased vulnerability to postpartum depression (Rubertsson et al. Reference Rubertsson, Wickberg, Gustavsson and Radestad2005). Our results indicate that stressors in pregnancy are also associated with chronic maternal depression. Importantly, it appears that depression is not only associated with a high but also a moderate number of stressors (Liu & Tronick, Reference Liu and Tronick2013). Finally, we found that work stress, and particularly work overinvestment, is associated with persistent depression, of both intermediate and high intensity. Prospective epidemiological studies show that overinvestment, i.e. an individual's need for approval and recognition at work, is associated with significantly elevated risks of depression (Siegrist, Reference Siegrist2008; Clark et al. Reference Clark, Pike, McManus, Harris, Bebbington, Brugha, Jenkins, Meltzer, Weich and Stansfeld2012). Pregnancy represents a dynamic period when both work conditions and worker assessment of occupational psychosocial stressors may change. Previous studies found that unskilled and low-waged occupations are often related to adverse employment conditions in pregnancy, including a lack of access to paid and unpaid parental leave. Women experiencing these working conditions are likely to experience greater work-related psychological distress than those who are more advantaged (Cooklin et al. Reference Cooklin, Rowe and Fisher2007; Miyake et al. Reference Miyake, Tanaka, Sasaki and Hirota2011). Yet, paid maternity leave is generally available in France (Ruhm & Teague, Reference Ruhm, Teague and Ferber1998) and χ2 comparisons (data not shown) revealed that work overinvestment was not associated with women's sociodemographic characteristics, the type (manual or white collar) or duration (full or part-time) of work they exercised, nor their partner's employment status. It was, however, more common in women who reported financial difficulties. In addition to work stress, overinvestment may also capture an individuals’ personality, characterized by low self-esteem and elements of Type A behaviour (van Vegchel et al. Reference van Vegchel, de Jonge, Bosma and Schaufeli2005; Clark et al. Reference Clark, Pike, McManus, Harris, Bebbington, Brugha, Jenkins, Meltzer, Weich and Stansfeld2012). It could well be that work overinvestment partly taps into these psychological constructs that have been found to be related to depression aetiology. Our finding suggests that work-related stressors should more frequently be studied in the context of pre- and postnatal depression, as work–life imbalance may contribute to mental health difficulties in certain women. In our sample, previous mental health problems and use of mental health services were independent psychiatric predictors of chronic depressive symptom trajectories, confirming earlier results (Seimyr et al. Reference Seimyr, Welles-Nyström and Nissen2013; Giallo et al. Reference Giallo, Cooklin and Nicholson2014).
Conclusion
In the present study we found evidence that depressive symptom trajectories in mothers of young children are heterogeneous and vary with a number of pre-existing characteristics. While most mothers do not have depressive symptoms, for some women maternal depression reaches its peak during pregnancy, in some during the preschool years while others follow a chronic course of symptoms, either of intermediate or high intensity. Women's trajectories of depressive symptoms may be changed when prevention or evidence-based treatments are implemented early on, during pregnancy or in the immediate postpartum. A recent review reported that psychosocial and psychological interventions, compared to usual postpartum care, were associated with a reduction in depressive symptomatology within the first 12 months postpartum (Dennis & Dowswell, Reference Dennis and Dowswell2013). Promising interventions include the provision of intensive, professionally based postpartum home visits, telephone-based peer support, and interpersonal psychotherapy. All of the potential risk factors of long-term depressive symptoms can be identified during routine prenatal care and could contribute to early diagnosis. We recommend that clinicians be specifically attentive to the well-being of women who experience prenatal anxiety symptoms or work-related stress. Nevertheless, proactive screening of psychosocial risks still remains a small part of prenatal care and less than 15% of pregnant and postpartum women receive adequate help (Coleman et al. Reference Coleman, Carter, Morgan and Schulkin2008; Austin et al. Reference Austin, Colton, Priest, Reilly and Hadzi-Pavlovic2013). Several psychosocial risk assessment tools, such as the Pregnancy Risk Questionnaire (Austin et al. Reference Austin, Hadzi-Pavlovic, Saint and Parker2005) or the Antenatal Psychosocial Health Assessment (Carroll et al. Reference Carroll, Reid, Biringer, Midmer, Glazier, Wilson, Permaul, Pugh, Chalmers, Seddon and Stewart2005), have been developed to assess pregnant women for the presence of psychosocial risk factors known to be associated with the onset of perinatal depression and anxiety. While some concerns remain about their psychometric properties (Johnson et al. Reference Johnson, Schmeid, Lupton, Austin, Matthey, Kemp, Meade and Yeo2012), it has been demonstrated that antenatal care providers (midwives, obstetricians, GPs) were more likely to detect concerns and identify a number of risk factors than providers who gave standard care (Austin et al. Reference Austin, Priest and Sullivan2008). Identifying women at risk for persistent depressive symptoms, through screening for multiple psychosocial risk indicators during pregnancy, may facilitate interventions targeting risks for maternal mental health problems and associated problematic early child development.
Appendix
The EDEN Mother–Child Cohort Study Group includes: I. Annesi-Maesano, J. Botton, M.A. Charles, P. Dargent-Molina, B. de Lauzon-Guillain, P. Ducimetière, M. de Agostini, B. Foliguet, A. Forhan, X. Fritel, A. Germa, V. Goua, R. Hankard, B. Heude, M. Kaminski, B. Larroque, N. Lelong, J. Lepeule, G. Magnin, L. Marchand, C. Nabet, R. Slama, M. J. Saurel-Cubizolles, M. Schweitzer, O. Thiebaugeorge
Declaration of Interest
None.
Acknowledgements
We acknowledge all funding sources for the EDEN study: Fondation pour la Recherche Médicale (FRM), French Ministry of Research: IFR Programme, INSERM Human Nutrition National Research Programme, and Diabetes National Research Programme [through a collaboration with the French Association of Diabetic Patients (AFD)], French Ministry of Health, French Agency for Environment Security (AFSSET), French National Institute for Population Health Surveillance (InVS), Paris-Sud University, French National Institute for Health Education (INPES), Nestlé, Mutuelle Générale de l'Education Nationale (MGEN), French-speaking association for the study of diabetes and metabolism (ALFEDIAM), National Agency for Research (ANR non-thematic programme), National Institute for Research in Public Health (IRESP: TGIR Cohorte Santé 2008 programme). This study was funded by the French National Research Agency (ANR, Programme on Social Determinants of Health). The funding agencies had no role in the design and conduct of the study; in the collection, management, analysis, and interpretation of the data; and preparation, review, or approval of the manuscript.
