Trajectories of Maternal Postpartum Depressive Symptoms

The Upstate KIDS Study is a population-based birth cohort study (2008–2010) designed to evaluate the impact of infertility treatment on child growth and development through the age of 3 years.¹⁹  By using birth certificates from the 57 counties in New York state, excluding the 5 New York City boroughs, infants were oversampled on infertility treatment, with all twins and higher-order multiples eligible to participate regardless of conception mode. The study cohort comprised 5034 mothers, including 1498 who had used some form of infertility treatment and 1129 mothers of multiple births (45 of whom birthed triplets or higher-order multiples). Most mothers (n = 4866; 97%) provided at least 1 assessment of depressive symptoms. Human subjects research approval was obtained from all participating institutions (New York State Department of Health Institutional Review Board 07-097; University at Albany 08-179), and informed consent was obtained before data collection.

Depressive symptoms were assessed by using the abridged 5-item Edinburgh Postnatal Depression Scale (EPDS-5), when infants were 4, 12, 24, and 36 months of age. The EPDS-5 has 5 items assessing emotional experiences over the past 7 days. Three of the 5 items included in the EPDS-5 are similar in content to the 5 excluded items from the 10-item Edinburgh Postnatal Depression Scale (EPDS).²⁰  For example, the excluded item “I have felt scared or panicky for no very good reason” is nearly identical to included item “I have been anxious or worried for no good reason.” Unlike the 10-item version of the EPDS, the 5-item version does not include a question about suicidality, but this item has not been shown to correlate strongly with the 10-item EPDS.^21,22  Items in the EPDS-5 are scored from 0 (no, never) to 3 (yes, most of the time), and items are summed to form a total score of depressive symptoms (range: 0–15). The EPDS-5 has been used to assess depressive symptoms in pregnant and postpartum women in various populations^23–25  and correlates strongly with the full EPDS (Pearson’s r = 0.96),²²  and the brief 5-item format allows for rapid assessment. The internal consistency omega²⁶  of the 5 items ranged from 0.71 (95% confidence interval [CI]: 0.68 to 0.74) at 24 months to 0.79 (95% CI: 0.77 to 0.80) at 4 months.

A baseline questionnaire at 4 months was used to assess demographic and perinatal factors, and hospital discharge records were abstracted to supplement maternal reports of perinatal conditions. Mothers reported their previous history of mood disorders (by endorsement of “has a doctor ever diagnosed you with a mood disorder”), education, marital status, race and ethnicity, and history of infertility treatment. When maternal reports were incomplete, responses were supplemented with vital records. Mothers’ ages, prepregnancy BMI, parity, plurality, gestational age in weeks, gestational diabetes mellitus (GDM), gestational hypertension, and infant sex were primarily obtained from vital records and supplemented with maternal reports when necessary. For mothers of multiples, one infant was randomly selected to serve as the index child for infant sex. Education was dichotomized as less than college versus college degree or higher. Marital status was coded as married, civil union, or domestic partnership versus not. Race and/or ethnicity were coded as white and non-Hispanic versus other. Infertility treatment was coded as none versus any treatment, including ovulation induction, intrauterine insemination, and/or in vitro fertilization. Plurality was coded as singleton versus multiple.

First, a linear latent growth model was fit in Mplus, version 8.2,²⁷  to determine if there was significant variance around the intercept and slope, which would suggest the presence of multiple groups with empirically distinct trajectories of depressive symptoms (see Supplemental Information). Next, the number of distinct trajectories of depressive symptoms was explored by using Asparouhov and Muthén’s²⁸  3-step method for latent class and growth mixture modeling. The 3-step method is used to (1) estimate latent classes with no predictors (which is identical to the unconditional model with no predictors in a traditional growth mixture model), (2) assign each participant to their most likely latent class group by using the posterior distribution obtained during the first step, and (3) regress class membership onto predictor variables by using multinomial logistic regression, taking into account the probabilistic nature of assigning class membership in the second step. On the basis of the overall favorable fit statistics of both the 3-step and conditional models and the stability of the class solution (similar class trajectories in the 3-step and conditional models; see Supplemental Information), the 4-class solution was determined to best fit the data. For all models, full-information maximum likelihood estimation (FIML)²⁹  was used to incorporate all observed data on the depressive symptom scores. FIML has been shown to work as well as multiple imputation in handling missing data.³⁰ 

As is common in longitudinal studies, there was dropout over time,³¹  with 30% of the 4866 mothers completing all 4 waves and 71% completing ≥2 waves (see Supplemental Information). There were few missing data points (0.35%) on the predictors because they were all collected at baseline, and these data were missing completely at random (Little’s missing completely at random = 0.56, P = .756). Consequently, only cases with complete data on the predictors (n = 4693) were included in the multinomial logistic regression model.

Table 1 presents demographic and perinatal characteristics of the sample as well as the mean levels of depressive symptoms at 4, 12, 24, and 36 months after birth. Overall, mothers displayed relatively low levels of depressive symptoms, but the range was large. To illustrate, by using a cutoff score of ≥7 for the EPDS-5,²²  11% of women were considered to have moderate depressive symptoms at 4 months, 8% at 12 months, 6% at 24 months, and 7% at 36 months, and 21% of women had moderate depressive symptoms at ≥1 time points.

We identified 4 longitudinal trajectories (Fig 1). The largest group was termed “low-stable” (n = 3637; 74.7%). These mothers began with low levels of symptoms (b-intercept = 1.55 EDPS points; 95% CI: 1.45 to 1.65) and remained low through all waves (b-slope = −0.01 points per 4 months; 95% CI: −0.01 to 0.03). The “low-increasing” group (n = 398; 8.2%) began with some depressive symptoms at 4 months (b-intercept = 3.73; 95% CI: 2.85 to 4.62) that increased across waves (b-slope = 0.38; 95% CI: 0.15 to 0.61). The “medium-decreasing” group (n = 613; 12.6%) began with higher depressive symptoms (b-intercept = 6.22; 95% CI: 5.51 to 6.94) and abated in symptoms over time (b-slope = −0.44; 95% CI: −0.62 to −0.26). Finally, the “high-persistent” group (n = 218; 4.5%) had the highest levels of depressive symptoms at 4 months (b-intercept = 9.83; 95% CI: 9.17 to 10.49), and, although symptoms decreased across waves (b-slope = −0.30; 95% CI: −0.44 to −0.17), they remained higher than other groups through 36 months.

Next, predictors of group membership were identified for the different courses of depressive symptomatology. Compared with the low-stable group (Table 2), mothers with a history of mood disorder diagnosis were >3 times as likely to belong to the medium-decreasing group (odds ratio [OR]: 3.67), 4 times as likely to belong to the low-increasing group (OR: 4.12), and 15 times as likely to belong to the high-persistent group (OR: 15.38). The high-persistent group was also markedly different from the low-stable group in other ways, tending to have been younger, less than college educated, unmarried, multiracial or a person of color, multiparous, diagnosed with GDM, and delivered earlier. Lastly, nulliparous mothers and mothers who smoked during pregnancy were more likely to be in the medium-decreasing than low-stable group.

To identify the risk factors associated with unremitting depressive symptoms, comparisons were made with the high-persistent group set as the reference (Table 3). The high-persistent group separated from the others in sociodemographics (ie, age and education) and by medical history (ie, mood disorders and GDM). Older mothers (ORs: 1.08–1.10) and mothers with a college education (ORs: 2.29–2.52) were more likely, and mothers with a mood disorder (ORs: 0.07–0.27) or GDM diagnosis (ORs: 0.23–0.44) were less likely to be classified in any other depressive symptom group than the high-persistent group. In addition, compared with the high-persistent group, mothers who smoked during pregnancy were more likely to be in the medium-decreasing group, and mothers who were nulliparous were more likely to be in the low-increasing group. A follow-up analysis was computed by using contrasts for mothers with no previous live births versus 1, 2, and ≥3 live births in place of nulliparity. The results were similar, indicating that having any number of previous live births increased the risk of being in the high-persistent symptom group compared with the low-stable group (data not shown).

We identified empirically distinct trajectories of postpartum depressive symptoms in a population-based sample of mothers. Like other studies, we found a large low-stable group and a small persistently depressed group.^10–16  In addition, we found 2 other groups of mothers with initially low-level but increasing symptoms and initially medium-level but decreasing symptoms. Approximately one-quarter of women were assigned to groups with elevated symptoms at some point in the 3 years after birth. These findings may, if replicated, have significant implications for identifying mothers at risk for persistent symptomatology. For example, a small group of mothers with elevated symptoms at 4 months may be at risk for sustained, clinically significant symptoms that endure up to 3 years, but another larger group of mothers may screen relatively high at 4 months, and then remit over time. Likewise, a mother who initially appears to have low-level depressive symptoms at 4 months may remain at low risk or have increasing symptoms over 3 years. Elevated symptoms can last for years after birth, and at least 2 assessments separated by at least several months are needed to determine the trajectory of an individual mother’s symptoms.

In this study, we also explored demographic and perinatal predictors of latent class trajectory membership. Understanding the factors that may identify women at risk for high-persistent depressive symptoms is important because women with depression experience comorbid medical problems^32,33  and struggle more with parenting and bonding with their children.^34,35  In this study, 2 factors distinguished mothers in the low-stable group versus groups with higher depressive symptoms initially or over time. Consistent with other studies, mothers with no history of mood disorder diagnosis were between 3 and 15 times more likely to be in the low-stable group than other groups.^13,36–38  Mothers who were having their first infant were also more likely to be in the low-stable group than the medium-decreasing and high-persistent groups. In 2 previous studies, researchers found that women with more previous live births were more likely to have high chronic levels of depressive symptoms than continuously low symptoms of depression,^12,14  but, in other studies, researchers have found no relations between parity and depressive symptoms³⁹  or, conversely, found that nulliparity increased the risk of depression.⁴⁰  In these studies, researchers defined parity differently (>2 previous pregnancies versus any previous live birth) and, in each study, adjusted for different factors (eg, only one adjusted for partner support and stressful life events³⁹ ), which may explain the different findings.

One novel finding in this study was the particular risk associated with GDM on membership in the high-persistent symptom group. Women with GDM were more than twice as likely to be classified in the high-persistent group than all other trajectory groups. Previous studies also revealed that women who experience GDM may be at increased risk for postpartum depression,^41,42  even after prepregnancy weight is controlled for analytically,^43,44  but this is the first study to link GDM to persistent depressive symptoms up to 3 years postpartum. Some common-cause environmental factors for depression and diabetes include low socioeconomic status, poor sleep and diet, and inactivity, pointing to stress as a potential mechanism of this association.⁴⁵ 

Infertility treatment, multiple births, prepregnancy BMI, gestational hypertension, and infant sex were not associated with depressive symptom trajectories. The existing literature on these factors is mixed. In most studies, researchers find no association between infertility treatment and postpartum depression,^39,46–49  although infertility before pregnancy is associated with depression.⁵⁰  Postpartum mothers conceiving with infertility treatment tend to be older and better-educated, which may correspond to better support and lower the likelihood of postpartum depression. However, multiple births occur more often in mothers who used infertility treatment than those conceiving naturally, and that may increase the likelihood of depression.⁵¹  In a systematic review of 7 studies, researchers concluded that multiple births is a risk factor for postpartum depressive symptoms,⁵¹  but 6 of the 7 studies included <200 women with multiple births, and, in 3 of the 7, researchers found no association between multiple births and depression. The null findings for infertility treatment and multiple births in this study are important given the large samples and controls for other relevant perinatal factors.

No characteristics predicted membership in the low-increasing group relative to the low-stable group apart from history of a mood disorder. Risk factors that occur or worsen after birth may be more important for predicting increasing symptoms. For example, it is possible that women who have infants with health complications or developmental delays or women who experience marital strain or other stressors after childbirth are at a greater risk for increasing depressive symptoms. In 2 other studies, researchers that identified groups of women with increasing depressive symptoms over time found only self-rated poor maternal health⁵²  and preterm birth¹²  to predict membership in these groups. Future studies should include more postnatal child and family factors to explore how these conditions predict depression trajectories.

This study’s strengths include the large sample, long-term follow-up, and recruitment of women of varied sociodemographic backgrounds with and without infertility treatment and multiple births. The limitations of this study should be acknowledged. First, this study was not designed to diagnose postpartum depression. Hence, the measure of depressive symptoms was not a clinical measure and was brief (5 questions). It is possible that longer or more clinical measures of depression might produce different trajectory groups and predictors of those groups. Still, this brief measure discriminated 4 trajectories of symptoms, and pediatricians are likely to use brief screeners in clinical care.⁵³  Second, the sample was predominantly non-Hispanic white, and, to have adequate variance, all other racial and ethnic groups were combined in the analysis. Third, in this study, we did not collect data before 4 months postpartum and, therefore, may have missed some important information about early postpartum symptoms. Fourth, we cannot rule out that choosing to participate in a study may itself have selected for women with fewer depressive symptoms, but the prevalence of at least moderate depressive symptoms at 4 months postpartum in this sample (EPDS-5 ≥7; 11%) was similar to the population prevalence of a major depressive episode in the postpartum period (12%).^8,54  Finally, we do not know which women received treatment of depressive symptoms, which may have altered their trajectories.

Pediatric primary care clinicians should be aware of the different trajectories of postpartum depressive symptoms and the possibility that elevated symptoms can persist for at least 3 years postpartum. This study reveals that assessing depression only once or too early in the postpartum period (ie, at 6 months or earlier, as recommended by the AAP¹ ) may make it difficult to discern the future course. Evidence suggests that a large percentage of mothers with depression do not seek treatment.^55,56  Many women hope that symptoms will abate over time without treatment.⁵⁷  Assessing mothers multiple times early and late in the postpartum period and extending the postpartum period to at least 2 years after birth would provide a clearer picture of mothers whose symptoms are persisting or increasing, and mothers who had not already sought treatment could be connected with resources.

Pediatricians should also pay particular attention to women who exhibit risk factors for depression. Women with a history of mood disorder are at increased risk for initially high and persistent depressive symptoms, medium-level symptoms that abate, and increasing symptoms up to 3 years after the birth of their child, relative to women who have low-stable symptoms. Young mothers and those without college education as well as women who had GDM may also be at particular risk for high-persistent depressive symptoms, compared with all other trajectories. It is important to note that pediatricians may not have easy access to the mother’s perinatal history because this information is generally stored in the mother’s and not the child’s medical record. Consequently, pediatricians may have to ask the mother more questions to assess these factors. In a national survey of pediatricians’ perceived responsibilities for identifying and treating maternal depression (conducted before the release of the AAP screening recommendations), >70% of pediatricians stated that there was insufficient time to obtain the mother’s history and provide counseling and education about maternal depression.⁵⁸  Routine well-child visits may need to be extended for pediatricians to have the time to assess the mother’s mental health and risk factors, in addition to the child’s health and development. Mothers’ mental health is critical to children’s well-being and development.

AAP

American Academy of Pediatrics

CI

confidence interval

EPDS

Edinburgh Postnatal Depression Scale

EPDS-5

5-item Edinburgh Postnatal Depression Scale

FIML

full-information maximum likelihood estimation

GDM

gestational diabetes mellitus

OR

odds ratio