Exploring Heterogeneity in perinatal depression: a comprehensive review

This systematic review is conducted per the Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) and the Meta-analysis of observational studies in epidemiology (MOOSE) guidelines for systematic reviews [16, 17]. The protocol for this systematic review was registered a priori in PROSPERO [18]. We searched PubMed and Web of Science, through 21^st February 2022, using a pretested search strategy encompassing terms about heterogeneity and perinatal depression (Table 1).

After searching academic databases, duplicate entries were removed using Endnote v. X9. After that, eligible studies were filtered in a two-phased screening process by two independent reviewers. First, the reviewers extracted study design and publication characteristics such as the scope of the study, country, setting, and sampling technique. Characteristics related to the study sample included inclusion and exclusion criteria, timepoints for data collection, psychometric tools, and diagnostic criteria used to assess PND. After that, statistical techniques used to analyze heterogeneous profiles of PND were enumerated. Findings were synthesized narratively according to their themes, such as the description of the study's heterogenous symptom profiles or trajectories, risk factors, child outcomes, and treatment considerations.

Quality assessment was done using an adapted version of The Newcastle–Ottawa Scale (NOS) for assessing the quality of nonrandomised studies in meta-analyses [22]. All studies were scored for quality across five domains, including appropriateness of study sample, comparability of the cohort based on design and analysis, adequacy of statistical analysis, attrition, and adequacy of outcome assessment. Each domain was rated as adequate, partly adequate, and inadequate.

We assessed the adequacy of the study sample by evaluating its representativeness of the population under study in each individual research. This evaluation was based on several factors, including the inclusion and exclusion criteria used in the studies and the reported characteristics of the participants. Specifically, we examined the recruitment methods used in each study and compared the demographic and clinical characteristics of the study participants with those of the broader population of interest.

The evaluation of exposure was deemed as either adequate or partially adequate based on the methods used for participant recruitment. Specifically, if diagnostic criteria were applied, the ascertainment of exposure was considered adequate. In contrast, if psychometric scales were used, it was rated as partially adequate. As for the assessment of outcome, it was deemed adequate if validated scales were employed to measure depressive symptoms, thereby facilitating the elucidation of symptom profiles or trajectories. Attrition rate was rated adequate if it was ≤ 20%. Statistical analysis was judged as adequate if established statistical methods were used for modelling heterogeneity in PND followed by use of established statistical criteria for retaining number of classes/subgroups.

We also utilized the Guidelines for Reporting on Latent Trajectory Studies (GRoLTS) to assess the quality of reporting in the studies reviewed [23]. The GRoLTS checklist standardizes the reporting of results from latent trajectory analyses. This checklist was particularly relevant for our review, given its focus on studies employing latent growth mixture modeling (LGMM) or latent class growth analysis (LCGA). We systematically evaluated each study against the GRoLTS key components, identifying potential weaknesses and gaps in reporting. This checklist includes 16 key components that are crucial for reporting results of trajectory studies. These components cover a range of aspects, including the metric of time used in the statistical model, handling of missing data, distribution of observed variables, software used, consideration of alternative specifications of within-class heterogeneity and between-class differences, use of covariates, number of random start values and final iterations, model comparison tools, total number of fitted models, number of cases per class, entropy, plots of estimated mean trajectories, characteristics of the final class solution, and availability of syntax files [23].

A narrative synthesis approach was used to synthesize evidence in this systematic review. Meta-analyses were not conducted due to heterogeneity in study designs and outcomes included in the review.

Generally, the quality of evidence presented by these studies was robust. Representativeness of the study sample was judged as adequate in 18 (40.9%) of the studies and partly adequate in 66.67% of the studies. Ascertainment of exposure was done using clinical diagnoses in only three studies (4.55%) and psychometric scales among the rest. Outcome assessment was done using psychometrically sound scales in all the studies. The extent of missingness and attrition was > 20% among 36 (54.55%) studies, while all but four studies (6%) used appropriate statistical analysis techniques and appropriate criteria for choosing the number of classes/trajectories of PND in their study samples (Fig. 2).

We assessed the quality of reporting in the included studies (n = 53) using the GRoLTS checklist (Fig. 3, Supplementary file 2). Our findings indicate that all studies reported the metric of time used in the statistical model and the software used. The majority of studies also provided information on how missing data were dealt with (86.8%), the distribution of observed variables (92.5%), and included a plot with the mean trajectories for the final solution (84.9%). Furthermore, 81.1% of studies provided a numeric description of the final class solution, and 79.2% reported the number of cases per class.

However, there were several areas where reporting could be improved. Only 39.6% of studies presented information about the mean and variance of time within a wave, and less than half (47.2%) reported the missing data mechanism. Only 26.4% of studies provided a description of what variables were related to missing data. While 74.5% of studies ensured replicability if covariates were used, only 49.1% considered alternative explanations for within-class heterogeneity (LGCA vs LGMM), and 50.9% described alternative forms of trajectories.

Furthermore, only 32.1% of studies reported the total number of fitted models, and a mere 5.7% provided information about the number of random start values and final iterations. Only 54.7% of studies reported entropy, and just 1.9% included plots with the mean trajectories for each model. None of the studies included a plot of the combination of estimated means of the final model and the observed individual trajectories. Lastly, only 35.8% of studies made their syntax files available.

A total of 53 studies presented evidence of heterogeneity in longitudinal trajectories of PND. A higher proportion of the studies (14, 26.42%) were conducted during the postpartum period, followed by during pregnancy (38, 71.70%). Only one study recruited study participants during both the pregnancy and postpartum periods. Various statistical techniques were utilized to assess longitudinal heterogeneity in perinatal depression. The most frequently utilized statistical analysis technique was growth mixture modelling (n = 13), k-means cluster analysis (n = 1), latent class (n = 4) and latent class growth analysis (n = 6), group-based trajectory modelling (n = 4), and semi-parametric mixture modelling (n = 12) and latent profile analysis (n = 2). Timepoints for data collection ranged from a minimum of two to ten. EPDS (n = 31) and CES-D (n = 12) were the most frequently used scale in analyses.

In the following sections, we will provide a more nuanced understanding of the interconnected aspects of pattern, severity, chronicity, and onset of symptoms, and their collective influence on the trajectory of Perinatal Depression (PND). Recognizing the complexity of PND, our discussion is structured into two interconnected sub-themes. The first, 'The longitudinal trajectory of PND symptoms: An examination of severity and chronicity' , explores how the intensity and duration of symptoms can shape the course of PND. The second, 'The longitudinal trajectory of PND symptoms: A focus on onset during prenatal and postnatal periods', investigates how the timing of symptom onset, whether during the prenatal or postnatal period, can influence the progression of PND [13, 21].

While this categorization may seem artificial due to the intertwined nature of these aspects, it is guided by the variables used by authors of the primary studies to derive the latent trajectories of PND. This approach underscores the importance researchers place on these aspects when studying longitudinal trajectories of PND. Furthermore, the distinction between trajectories based on severity and chronicity versus onset timing underscores the complexity of PND and suggests that postnatal depression may often be a continuation of antenatal depression. This observation is crucial in understanding the differences between the DSM-V's perinatal specifier and the ICD-10's focus on postpartum depression alone [19, 20]. Through this approach, we aim to offer valuable insights into the heterogeneity of PND and inform more effective strategies for its screening, diagnosis, and treatment.

Risk factors for PND trajectories in 40 of the studies (Table 2). Among these studies, predictors of more debilitating trajectories included psychosocial, biological, and environmental factors. Among psychosocial risk factors, the most frequently reported risk factors included poorer marital relationships, poor social support, unwanted pregnancy, stressful life events, low income and financial difficulties and belonging to ethnic minorities. In addition, the presence of anxiety and stress during pregnancy and history of depression were among significant psychiatric risk factors.

Among biological variables, a smoking history, alcohol consumption, and inadequate BMI were essential. Child developmental difficulties, poor baby weight, preterm birth, and delivery complications were also significant risk factors. Younger and poorly educated women having high parity were also at risk of developing more severe forms of PND.

Four studies provided risk factors for a worse prognosis while accounting for the onset period [77, 79, 82, 97]. As per these, those at high risk of developing postpartum depression had higher anxiety scores during the third trimester of pregnancy, high levels of stress in family, low education, minority ethnic groups and reported lower weight among their offspring [79]. Oh et al. reported no association of depression trajectories with age at childbirth, education level, delivery mode, breastfeeding, alcohol consumption, or smoking [82]. Van der Waerden et al. reported that French women who developed severe depression during pregnancy belonged to ethnic minority groups, lacked social support, had a history of mental health problems, pre-pregnancy substance use, and reported comorbid anxiety during pregnancy [85‐87, 98]. Women with depressive symptoms persistent during their child’s preschool period only report their partner’s low educational level, pre-pregnancy mental, health treatment, and anxiety during pregnancy. While those with persistent depression reported experience of life events during pregnancy, work over-investment, pre-pregnancy mental health history/ treatment, and anxiety during pregnancy [85‐87, 98].

Child health outcomes were reported in 29 of these studies (Table 3). These studies reported several health problems among offspring born to mothers with complex depression trajectories: developmental problems, behavioural problems, poor health at delivery, and poorer physical health.

Four studies found significant associations between delivery-related complications and physical illnesses among women with severe PND. These included preterm birth [36], low birth weight [52], low BMI, and low-fat mass [65] among children and adolescents. Among these studies, Farías-Antúnez et al. utilized Pelotas 2004 Birth Cohort [65], following perinatal women up to 11 years postpartum. After adjusting for confounders, children raised by mothers with chronic high PND have adequate body composition indices (BMI and fat mass). In contrast, Surkan et al. found that mothers with moderate to severe PND symptoms had children with shorter stature, and this persisted throughout the child’s first six years of life [33]. In addition, Puosi et al. reported a high risk of wheezing and eczema among children born to mothers with high (persistent or otherwise) depressive symptoms [84].

Lee and Park [38] reported that children born to mothers with aggravated depression reported significantly poorer cognitive (Cohen’s d = 0.43) and motor development (Cohen’s d = 0.34) according to Bayley’s Scale. Aoyagi et al. also found lower expressive language scores among children with mothers reporting late-onset PPD with a monotonic decline between 10 to 40 months postpartum [26].

A higher risk of developing conduct [41, 42, 51, 63] and emotional problems [52] was associated with severe forms of PND. Netsi et al. utilized the ALSPAC birth cohort data and found that children born to mothers with persistent PND were significantly at higher odds of developing behavioural problems (OR 4.84; 95%CI, 2.94–7.98); lower GCSE mathematics grades (OR = 2.65, 95% CI, 1.26–5.57) and higher depression rate (OR = 7.44, 95% CI, 2.89–19.11). Flouri & Loakeimidi utilized data from the UK Millenium Cohort Study to model the effect of trajectories of maternal depressive symptoms on antisocial behaviour and delinquency among children (aged three to 11 years, N = 12,494) [51]. They found that the boys born to mothers with chronically high or accelerating maternal depressive symptoms were more likely to report engaging in loud and rowdy behaviour, alcohol use, and bullying. Females exposed to chronically high maternal depressive symptoms were more likely to support the view that alcohol use is harmless [51]. Hammerton et al. reported that compared to offspring of mothers with minimal symptoms, the most significant risk of suicidal ideation was found for offspring of mothers with chronic-severe symptoms [95]. Only one study found no association of PND trajectories with alcohol use, abuse, and involvement in fights among postpartum Brazilian mothers in Pelotas, Brazil [64]. Fransson et al. found that postpartum bonding mediated most of the adverse effects of postpartum and persistent depression on child behaviour [27].

This systematic review presents evidence on heterogeneous clinical presentations of PND. The reviewed studies employed different analytical approaches to visualize heterogeneous longitudinal trajectories, the pattern of symptom profiles, and symptom networks. Studies delineating trajectories reported both stable linear as well as unstable quadratic patterns. These trajectories ranged from minimal PND symptoms to stable linear and clinically severe, chronic, and persistent trajectories. Important symptom profiles challenged the notion of PND being considered a homogeneous latent construct. Most studies reported severe and persistent PND symptoms among perinatal women facing sociocultural stressors, less stable social networks, poverty, and displacement. At the same time, these were poor and persistent neurodevelopmental, socioemotional, and physical health outcomes among their children. The research evidence generally was robust, primarily obtained from extensive birth cohort studies or secondary analyses of RCTs.

This systematic review highlights an important consideration. Most studies reported a subset of perinatal women with either linear and worsening trajectories or persistent and severe symptom trajectories. It also highlighted the subset of women who are always at a high risk of relapse and recurrence. Moreover, this review demonstrates that this subset of perinatal women suffers from a double burden of severe PND symptoms and social, economic, and cultural diversities. This is an important finding because most mental health systems recommend utilizing cost-efficient stepped-care approaches [100, 101]. This is detrimental to the health of women who are not the right candidates for these stepped-care approaches and require high-intensity treatment at the outset to ease their suffering and offset poor infant and child outcomes.

This systematic review presents insights into the nature of PND. First, it demonstrates the heterogeneous presentations of PND and the associated differential risk factors. This challenges the prevailing understanding of PND as a homogenous latent construct based on the principles of essentialism and reductionism in psychiatry and psychometric research [102, 103]. There is, however, currently a lack of pathophysiological research presenting different mechanisms underpinning different heterogeneous profiles of PND [35]. Furthermore, more basic and clinical research is needed to understand this, accounting for this heterogeneity. This is vital because treating PND as a homogenous construct has hindered meaningful empirical research into the aetiology of PND [14]. Furthermore, although PND is considered a discrete diagnostic category, it has significant overlaps with symptoms of other disorders, including somatic and hypochondriac illnesses and anxiety [25], which usually bear poor prognosis.

Building on the evidence from this systematic review, it is crucial that interventions for PND be provided as early as possible. Although the current WHO recommendations [104, 105] currently emphasize acting during the early antenatal period, distinct trajectories of PND have also been shown with onset in early or postpartum. Therefore, we opine that screening and prevention should start earlier when the pregnancy is planned and during the early postpartum period. For instance, such screening and prevention strategies should be planned for those with prodromal symptoms (high distress pre-pregnancy) or those considered high-risk due to socioeconomic factors and pre-existing psychiatric and clinical pathologies [87], as this subgroup is at the highest risk for developing severe PND trajectories. This notion is also supported by Phua et al. who observed that the symptom networks among postpartum women are highly interconnected as they keep reinforcing each other over time and thus, become harder to treat [34].

This review has several clinical implications. Firstly, we opine that these findings can be used to develop cost-effective screening and prevention and treatment referral pathways. These pathways can be envisaged from a population health paradigm from promotion and prevention to treatment. From a health promotion perspective, policy makers must recognize the psychosocial risk factors that predispose perinatal women to severe forms of PND. As noted in this review, women with severe PND experience racism as migrant women, poverty, gendered oppression, and the lack of a social safety net. There is a need to address the intersecting effects of these social risks on perinatal mental health. Therefore, stakeholders must consider the perinatal experience's psychosocial underpinnings and design relevant policies [106]. Interdisciplinary approaches are required to bring about sustained health effects among the perinatal populations. For instance, using developmental economics approaches to tackle inequalities among perinatal women facing adversities can positively affect overall well-being [107]. These approaches have shown great promise in uplifting the general population's quality of life and mental health in recent trials [107].

For prevention, community-based surveys employing PND screening tools and social risk questionnaires [104, 108] can be utilized to identify women with either high social risk or prodromal depressive symptoms. Those found to be at the highest risk can be offered low-intensity interventions as recommended by the WHO and the US Preventive Services Taskforce [105, 109]. The same surveys could also be used to screen women who have already developed PND. In this context, these social risk questionnaire and depression screening tools can help identify heterogeneous trajectories of PND. This is an invaluable strategy and can lead to the development of next-generation of prognostic tools [110], which can match patients with PND to either low-intensity or high-intensity care depending on the severity and psychosocial risk; thus, helping improve the prognosis of women suffering from PND.

Another critical consideration in the prognosis of PND is the presence of overlapping anxiety symptoms [25]. There is considerable evidence that comorbid anxiety symptoms among women with PND worsen their prognosis [110]. Therefore, screening, prevention, and treatment strategies for PND must be considered. Using such transdiagnostic approaches has also been recommended in the Lancet’s Commission for Global Mental Health and sustainable development [111]. Furthermore, in this context, assessment of PND among women calls for either use of elaborate psychometric scales including symptoms of anxiety [112] or pairing depression-focused scales (such as the Patient Health Questionnaire) with scales for assessment of anxiety [113, 114].

The majority of evidence presented in this review pertains to the longitudinal trajectories of PND, with a focus on the severity of symptoms over time. While the timing of PND onset is a critical aspect of this discussion, it's important to note that different trajectories may not necessarily indicate a fundamental heterogeneity in the condition itself. We propose that variations in trajectory patterns could be influenced by factors such as life events, socioeconomic conditions, and access to treatment services (Table 2). These differences in baseline characteristics could contribute to the diverse trajectories of perinatal depressive symptoms observed across studies. This further supports our proposition that variations in trajectory patterns may be partly explained by psychosocial differences among women.

However, it's also possible that part of this variation could be attributed to differences in the timing of symptom onset or the specific symptom profiles. For instance, Waqas & Rahman [25] demonstrated a steep decrease in PND severity scores among perinatal women with somatic PND. This connection between diverse symptom profiles and varying longitudinal trajectories has not been extensively explored, and we recommend that future research investigates PND trajectories among perinatal women with different symptom profiles.

This systematic review presents updated evidence on heterogeneous profiles of PND. It provides a comprehensive overview of different types of empirical research in this domain and summarizes the risk factors associated with subtypes of PND with the worst prognosis. Moreover, evidence is also presented for the intergenerational effects of PND. The studies included in this review were generally of high quality and utilized datasets from large cohorts. However, large disparities in research evidence were noted, with only a small subset from LMICs. Such research delineating longitudinal trajectories and symptom profiles requires extensive infrastructure such as that of established birth cohorts in the UK [92]. Brazil contributed the most significant number of studies among the LMICs, where almost all utilized the dataset from the Pelotas birth cohort [62‐67, 69, 70, 75, 115, 116]. Therefore, we recommend that investigators based in LMICs build birth cohort infrastructure to delineate the nature of PND, associated risk factors, and inequities in their communities. This will help formulate data-backed policies in perinatal mental health and help intelligent channelling of resources to screen, prevent, and treat perinatal mental disorders [117].

Another limitation of this review is that while we have synthesized the heterogeneous symptom profiles and longitudinal trajectories of perinatal depression, we have not conducted a detailed meta-synthesis of the risk factors associated with trajectories of worse prognosis and long-term sequelae. This task would involve the meta-aggregation of heterogeneous effect sizes, as well as a quality review of studies using scales for epidemiological studies of risk factors and long-term consequences. Given the complexity and scope of this task, it was beyond the purview of our current review. However, we recognize the importance of this endeavour for a comprehensive understanding of PND and suggest it as a direction for future research. This limitation does not diminish the value of our current findings but rather highlights an area where further work is needed to enhance our understanding of PND.