Network Structure of Perinatal Depressive Symptoms Among Women in Rural India

Abstract

Background:

Perinatal depression (PND) is a common mental disorder affecting women worldwide. Despite receiving considerable attention as a unitary construct, PND’s inter-symptom interactions have received limited attention. The present study examined the network structure of perinatal depressive symptoms, their relative importance, and interconnectedness among women experiencing PND from rural Bihar.

Methods:

A community-based cross-sectional study screened women for PND using the Edinburgh Postnatal Depression Scale (EPDS) through door-to-door surveys by trained workers. Women scoring above the cutoff were recruited. Using R 4.0.3 and the qgraph package, the network structure was estimated to visualize symptom relationships. Centrality indices highlighted the connectivity of each symptom within the network. Accuracy and stability analyses assessed the robustness of the results.

Results:

Data from 132 participants with elevated depressive symptom scores on the EPDS were analyzed. The most central item was “I have been so unhappy that I have been crying”. The strongest association was between “laughing/seeing the funny side of things” and “looking forward with interest and pleasure.” Results suggest that mood and interest/pleasure symptoms, such as crying, anticipating enjoyable activities, and feeling sad showed relatively higher strength centrality in this exploratory analysis.

Conclusion:

The present study advances efforts to uncover symptom–symptom interactions within PND networks in India, particularly in rural Bihar. Future research to establish the clinical and prognostic utility of these symptoms can pave the way for evidence-based interventions in rural Indian settings.

Keywords

Depressive symptoms network analysis perinatal depression symptom-level interaction

Key Messages:

Question: What does examining perinatal depressive symptoms as a network reveal?

Finding: Mood and anhedonia symptoms were the most central symptoms in the network, with the strongest association found between humor and pleasure items.

Meaning: The findings shed light on the network structure of perinatal depressive symptoms in rural Bihar and can inform the development of targeted interventions.

Perinatal depression (PND) is a common condition affecting women and families worldwide, and a serious public health issue.¹ Studies have reported a nearly 12% prevalence of PND, with higher rates among women from low- and middle-income countries.² For example, in a recent study, the prevalence of PND was estimated to be nearly 24% in the Indian state of Bihar.³

Depression during or after pregnancy is associated with adverse neonatal outcomes and can also impact maternal physical and mental health as well as caregiving practices, including breastfeeding.^4–6 Some longitudinal studies show that outcomes may extend well beyond infancy, as children of depressed mothers have been found to show impairments in emotion recognition and social collaboration even in late childhood.⁷ PND, with its associated detrimental outcomes, reflects a pressing cause for concern, making the need to mitigate its far-reaching impacts crucial.

A notable gap in current PND literature is the relative lack of focus on its specific symptoms in favor of disease-level conceptualizations.⁸ Traditionally, symptoms of mental disorders have been presumed to stem from underlying disease.⁹ Emerging research, however, offers alternative frameworks such as the network theory of mental disorders, which have been gaining ground in psychopathology literature.¹⁰ Rather than viewing disorders as stemming from a latent entity, complex network approaches conceptualize them as systems of causally connected symptoms.¹¹ According to Borsboom and Cramer, “causal, meaningful relations between symptoms not only exist and should be acknowledged, but in fact are the very stuff of which mental disorders are made (p. 96).”¹¹

According to the network approach to psychopathology, symptoms of a mental disorder (represented as nodes) and their interactions can be understood as a network, in which the connections between nodes represent causal interactions among the symptoms. Symptoms that directly activate one another are seen as interconnected nodes, while the corresponding nodes of symptoms that do not directly activate one another are not connected in the network.¹⁰

A growing body of literature in recent years has examined the network structure of mental illnesses.^12,13 A core interest of such studies tends to be the overall connectivity of a symptom in the network. This metric, called centrality, represents how closely a symptom is interconnected with all other symptoms in the network.^14,15 In the case of depression, for example, if a highly central symptom is activated, the impulses from this symptom will spread through the network. Numerous other symptoms will be activated, but this will not be the case if a more peripheral symptom is activated, as it will have relatively fewer ways to exert such an effect on the network.¹⁵ A systematic review by Malgaroli and colleagues found fatigue and depressed mood to be frequently identified as central depression symptoms in network analyses.¹⁶

Given its multifaceted presentation, researchers have underscored the importance of employing a network analysis approach to untangle complex symptom interactions and elucidate the heterogeneity of PND.¹⁷ Increasingly, network analysis has been employed in studies with perinatal women, exploring connections between symptoms associated with a multitude of mental health issues, such as depression and anxiety.^8,18–21 Among depressive symptoms, frequently cited central symptoms include sad mood/feeling sad or miserable,^8,18,20,21 depressed mood,⁸ and crying.¹⁹ Other central symptoms include anhedonia and feeling guilty,²¹ feeling anxious/worried^19,20 or feeling scared/panicked,^18,20 and feeling happiness.⁸ Strong symptom–symptom associations have also been found between items relating to being able to laugh and looking forward to things with enjoyment,^18–20 feeling happy and enjoying life, feeling disliked by people, and feeling that people were unfriendly,⁸ feeling sad and crying,^8,19,20 sad mood and guilt,²¹ feeling anxious/worried and feeling scared/panicked.^18,20

Very few studies on Indian samples have explored the network structures of common mental disorders such as depression and anxiety, with no available literature on PND to the best of our knowledge. For instance, Wasil et al., in studies with Indian adolescents, found that feeling like a failure, sad mood,²² and negative feelings, worry, and feeling nervous²³ emerged as highly central symptoms. Among adult primary health care attendees in Goa, India, depressed mood, panic, fatigue, concentration problems, as well as somatic symptoms have also emerged as central symptoms.²⁴ Despite the growing evidence from international studies, India’s PND picture needs to be viewed within its unique context. While some factors associated with PND may be common among countries, such as age and marital conflict, some factors may be particularly salient in India.²⁵ For instance, the role of lower socioeconomic status in PND is more influential in India compared to Japan.²⁵ Further underscoring cross-country differences, Takegata et al. reported that in India, interpersonal factors relating to family characteristics played a major role in the development of PND, whereas in Japan, conflict at work and conflicts with family members were more prominent.²⁵ Given these differences in sociocultural considerations, the scenario may appear differently in India, and we risk losing valuable context and nuance by assuming that international findings apply as such to India, making this inquiry a much-needed addition to PND literature. The present study’s focus on a rural Indian setting is a strong and unique contribution, particularly since findings from rural Indian settings have identified a host of PND risk factors such as poor treatment by in-laws, poverty, or weak financial status, history of abortion, maternal physical/medical illness, complications during pregnancy, and the birth of a female baby.^3,26 Studying PND in an Indian population also offers additional advantages, such as helping strengthen the cross-cultural understanding of the disorder. While directly comparing networks stemming from other cultures may be difficult, symptom–symptom interaction in diverse settings, cultures, or sample characteristics may be studied to obtain a clearer picture of differences in the structure of psychopathology.²²

Another key consideration of the network theory of mental disorders is that factors outside the network (such as losing one’s partner)—the external field of symptoms—may activate symptoms in the network.¹⁰ Relatedly, network properties may differ across contexts. For instance, the connections among and between depression and anxiety symptoms have been shown to differ between low- and moderate-to-high stress groups, including higher levels of global network strength in networks with higher perceived stress levels.¹² Thus, shifts in the external field—such as the poor treatment by in-laws or the birth of a female baby (both among salient risk factors in PND in India)^3,26—may activate symptoms in the network, such as depressed mood, which, in turn, prompts neighboring symptoms to align themselves with this depression symptom.¹⁰ Given that causal relations among symptoms may rely on mechanisms in the environment, residing external to the person themselves,²⁷ such contextual factors may also influence the structure and connectivity of the PND network.

The present study aimed to elucidate the network structure of PND among women in the rural districts of Bihar, India. This study sought to address two primary objectives: (a) To identify the most central symptoms in the PND network among perinatal mothers, and (b) to identify the strongest edges or associations among the perinatal depressive symptoms within the network. This analysis was conducted using data from a previously published larger research project.³ The present network analysis was not pre-specified in the original study protocol and should therefore be considered an exploratory, secondary analysis.

Methods

Ethical Considerations

The study protocol was approved by the institutional review board and ethics committee before the commencement of the study. Written informed consent, including both consent to participate and consent to publish, was obtained from the study participants in the local language (Hindi) before recruitment. Additionally, women who met or exceeded the cutoff score on the PND screening tool were provided with the appropriate management, including care under a trained psychiatrist, and pharmacological and/or non-pharmacological treatments, as required.

To ensure transparency and methodological rigor, study findings are reported according to the Strengthening the Repor-ting of Observational Studies in Epidemiology (STROBE) cross-sectional reporting guidelines.²⁸ The STROBE checklist has been uploaded as supplementary online material (Supplementary File 1).

Study Site

This cross-sectional study was condu-cted in Dalsinghsarai Taluk, Samastipur District, Bihar, during June 2019–December 2019. With a total population of 10,38,05,000,²⁹ Bihar is an underdeveloped and heavily populated state in the Indian subcontinent. In 2025–2026, Bihar has allocated 6.6% of its total budget expenditure to health,³⁰ which is below the National Health Policy’s (2017) recommended target for states to allocate over 8% of their budgets to health.³¹

Participants

The target population consisted of women who were pregnant or postpartum (within one year of giving birth) at the time of the survey, and participants were selected through a two-stage case selection strategy. A door-to-door survey was carried out in the Dalsinghsarai Taluk by community-level health workers, who were trained before data collection in relevant mental health concepts and in administering the screening tool to identify potential participants. The health workers conducted systematic visits to households in the area. Eligible women (pregnant or one year postpartum) were identified through direct inquiry and available health records, such as infants’ antenatal cards or immunization records. They were screened using the selected screening tool. Those who screened positive were approached for informed consent, and if they provided it, they were formally enrolled in the study. Based on this criterion, women with elevated PND symptoms were included in the final sample.

Sample Size

This study is part of a larger research project on PND that recruited 564 participants, all of whom were screened using the Edinburgh Postnatal Depression Scale (EPDS).³² For the present analysis, our focus was on estimating the network structure of PND. In line with established guidelines for case identification, we included participants who met or exceeded the cutoff score to identify women at risk of depression.

Instruments

A semi-structured proforma was developed to collect a basic demographic profile of the study participants. The EPDS was used to screen and assess symptoms of depression among the participants.³² The self-report scale comprises 10 items and asks participants to rate their feelings across several areas over the past one-week period. The maximum possible score is 30. Based on the recommended cutoff for research use and in accordance with previous research in an Indian population, a cutoff score of 10 and above was used as an indicator of possible depression.^32,33 Though originally intended to detect postnatal depression, the EPDS may be used to screen for PND.³⁴ The present study employed a validated Hindi translation of the tool.³⁵

Statistical Analysis

Descriptive statistics were used to describe the socio-demographic variables. The mean and standard deviation were used to describe continuous variables, while frequency and percentage were used for categorical variables.

As the study analyzed item-level responses from the EPDS, which are ordinal Likert-scale data, traditional normality testing was not applicable. Polychoric correlations were therefore used, as they are suitable for ordinal data and do not require normality assumptions. There were no missing responses in the dataset; hence, all 132 participants with complete EPDS data were included in the analysis.

Network analysis involved the following steps:

Network Estimation

To estimate the network structure of the ten depression symptoms, data were analyzed using R (version 4.0.3),³⁶ and the qgraph package.³⁷ Visually, the network is represented graphically based on the Fruchterman–Reingold algorithm, and the nodes that have stronger connections and/or a greater number of connections are placed closer together.³⁸ The network was estimated using a Gaussian graphical model,³⁹ where edges represent partial correlation coefficients between variables and nodes represent observed variables. The analysis used polychoric correlations as a base. The “least absolute shrinkage and selection operator (LASSO),”⁴⁰ which shrinks partial correlations and sets others to zero, has been used in estimating psychological networks as a form of regularization and gives a conservative or sparse network.⁴¹ Network selection involves finding an optimal fit between the network and data by minimizing some information criterion.⁴² In this case, the extended Bayesian Information Criterion (EBIC) was used.⁴³ The qgraph package combines a variant of the LASSO, the graphical LASSO, with EBIC model selection, and was used to estimate the present EPDS network.^37,44 The default value of 0.5 was used as the hyperparameter value gamma.

Centrality

An important component of network research, centrality, which refers to the extent of connectivity of a symptom overall, was also assessed.^14,15 Highlighting the importance of centrality in network research, researchers have suggested that highly central symptoms may represent ripe targets for clinical interventions.⁴⁵ Strength centrality was calculated, which is defined as the sum of absolute weights of all edges that contain a given node.^14,46 Betweenness centrality (how many times a given node lies on the shortest path between two other nodes), and closeness centrality (the inverse of the sum of the shortest path lengths from a node to other nodes) were also calculated.¹⁴

Network Accuracy and Stability

The accuracy and stability of network edges and centrality indices were assessed using the R package bootnet⁴¹ to check the robustness of the findings. Following established recommendations,⁴¹ the following procedures were applied: the non-parametric bootstrap (1,000 iterations) was used to estimate the accuracy of the edge weights based on calculating 95% confidence intervals (CIs) around the edges, and a case-dropping subset bootstrap (2,500 iterations) was used to assess the stability of centrality indices, that is, how the order of centralities holds up after observing only a subset of the data. To assess this stability quantitatively, the correlation stability coefficient (CS-coefficient) was used.⁴¹

Additionally, a confirmatory factor analysis (CFA) was conducted to validate the scale’s underlying latent structure. This analysis was performed on the larger dataset from the parent project, as restricting it to only the subgroup with elevated depressive symptoms would reduce score variance and yield unstable estimates. Therefore, using the full sample is consistent with recommended practice.⁴⁷

Results

Demographic Characteristics of the Study Participants

The sample consisted of 132 women from Dalsinghsarai Taluk, Samastipur District, Bihar. Participants included women who were currently pregnant or within one year of delivery. These were first pregnancies for 17% of the participants. Around 16% of participants had also undergone a previous abortion. The average age of the participants was 17.9 years. The mean EPDS score was 15.0 (SD = 4.39, range = 16). Table 1 provides item-wise EPDS score distributions for the study participants. Mean education attained (in years) was 6.2 (± 4.8). The monthly household income was less than Rs. 5,000 for 43% of the sample, and only 2.3% had household incomes higher than Rs. 15,000 per month. Furthermore, 28% of women reported that their husbands worked away from home in major cities, and 22% of participants reported living alone.

Table 1.

Frequency of the 10-item Edinburgh Postnatal Depression Scale Responses (N = 132).

Item	Score
	0	1	2	3
	N (%)	N (%)	N (%)	N (%)
I have been able to laugh and see the funny side of things	41 (31.1)	68 (51.5)	14 (10.6)	9 (6.8)
I have looked forward with enjoyment to things	30 (22.7)	69 (52.3)	15 (11.4)	18 (13.6)
I have blamed myself unnecessarily when things went wrong	16 (12.1)	18 (13.6)	73 (55.3)	25 (18.9)
I have been anxious or worried for no good reason	6 (4.5)	20 (15.2)	81 (61.4)	25 (18.9)
I have felt scared or panicky for no very good reason	26 (19.7)	26 (19.7)	69 (52.3)	11 (8.3)
Things have been getting on top of me	15 (11.4)	34 (25.8)	40 (30.3)	43 (32.6)
I have been so unhappy that I have had difficulty sleeping	20 (15.2)	18 (13.6)	70 (53)	24 (18.2)
I have felt sad or miserable	14 (10.6)	44 (33.3)	55 (41.7)	19 (14.4)
I have been so unhappy that I have been crying	12 (9.1)	74 (56.1)	28 (21.2)	18 (13.6)
The thought of harming myself has occurred to me	58 (43.9)	13 (9.8)	53 (40.2)	8 (6.1)

Perinatal Depression Network

The network of the 10 EPDS symptoms is shown in Figure 1. The network was sparsely connected with no negative edges (red lines) observed among nodes. Strong associations were observed between the two positively-worded items on the EPDS—item one (I have been able to laugh and see the funny side of things) and item two (I have looked forward with enjoyment to things)—as well as between item eight (I have felt sad or miserable) and item nine (I have been so unhappy that I have been crying), and between item seven (I have been so unhappy that I have had difficulty sleeping) and item eight (I have felt sad or miserable). Other items showed relatively weak connections, and item three (I have blamed myself unnecessarily when things went wrong) had the fewest connections with other nodes in the network.

Figure 1.

Nodes represent items and edges represent partial correlations among them. The size and density of the edges between the nodes represent the strength of connectedness among items. B1 = I have been able to laugh and see the funny side of things, B2 = I have looked forward with enjoyment to things, B3 = I have blamed myself unnecessarily when things went wrong, B4 = I have been anxious or worried for no good reason, B5 = I have felt scared or panicky for no very good reason, B6 = Things have been getting on top of me, B7 = I have been so unhappy that I have had difficulty sleeping, B8 = I have felt sad or miserable, B9 = I have been so unhappy that I have been crying, B10 = The thought of harming myself has occurred to me.

Centrality indices (betweenness, closeness, and strength centrality) for the network were computed and are depicted in Figure 2. Since researchers have questioned the applicability of betweenness and closeness indices in psychological networks,⁴⁶ the present discussion will focus on strength centrality. The three symptoms—item nine (I have been so unhappy that I have been crying), item two (I have looked forward to things with enjoyment), and item eight (I have felt sad or miserable)—ranked comparatively higher on strength centrality.

Figure 2.

Values on the horizontal axis are z-scores. B1 = I have been able to laugh and see the funny side of things, B2 = I have looked forward with enjoyment to things, B3 = I have blamed myself unnecessarily when things went wrong, B4 = I have been anxious or worried for no good reason, B5 = I have felt scared or panicky for no very good reason, B6 = Things have been getting on top of me, B7 = I have been so unhappy that I have had difficulty sleeping, B8 = I have felt sad or miserable, B9 = I have been so unhappy that I have been crying, B10 = The thought of harming myself has occurred to me.

Accuracy and Stability Analysis

Figure 3 presents the estimated edge weights of the EPDS network, and their corresponding bootstrapped 95% CIs. The large CIs (wide gray areas in the figure) around the estimated edge weights indicate that many of the edge weights are likely not significantly different from one another. Thus, interpreting the order of edges in the network should be done with caution.

Figure 3.

The horizontal lines represent edges in the network, and are ordered from highest to lowest edge weight. The red line in the figure shows the original sample values, the black dots depict the bootstrap means, and the surrounding gray areas indicate the bootstrapped CIs for the edge weights. B1 = I have been able to laugh and see the funny side of things, B2 = I have looked forward with enjoyment to things, B3 = I have blamed myself unnecessarily when things went wrong, B4 = I have been anxious or worried for no good reason, B5 = I have felt scared or panicky for no very good reason, B6 = Things have been getting on top of me, B7 = I have been so unhappy that I have had difficulty sleeping, B8 = I have felt sad or miserable, B9 = I have been so unhappy that I have been crying, B10 = The thought of harming myself has occurred to me.

Figure 4 displays the stability of strength centrality assessed using the case-dropping subset bootstrap. This plot shows the average correlation of the strength measure with the original sample across different sampling proportions. The shaded area indicates the variability (confidence interval) across resampled subsets.

Figure 4.

Values on the x-axis show the proportion of the original sample retained at each step, while the y-axis represents the average correlation between the original and re-estimated strength centrality values after progressively excluding portions of the sample.

Visual inspection of the centrality-stability plot, together with the computed CS-coefficient for strength (0.205), which falls below the recommended minimum of 0.25,⁴¹ indicates that the estimated strength centrality values are unstable. Therefore, the centrality of symptoms in the present network should be interpreted with caution.

CFA supported the underlying factor structure of the EPDS in the full dataset. The two-factor model showed good fit (χ²(34) = 76.96, p < .001; RMSEA = 0.048; CFI = 0.981; TLI = 0.975), and the three-factor model also demonstrated good fit (χ²(29) = 50.54, p < .01; RMSEA = 0.037; CFI = 0.991; TLI = 0.985). Both structures are well supported in previous EPDS literature.^48,49

Discussion

The present study sought to analyze a network model of PND based on the EPDS scores of 132 women in Bihar, India. Analyses included network structure estimation and calculation of centrality indices. An accuracy analysis was also undertaken to assess the robustness of the network estimates. Since psychological research—including the present study—typically involves limited sample sizes, the accuracy and subsequent interpretation of the estimated network and its parameters remain a concern, making this an essential step in network estimation.⁴¹

The following section discusses the strongest edges and strength centrality estimated in the present network. Although the stability of strength centrality fell below the recommended threshold given by Epskamp et al.⁴¹ the authors note that their recommended cutoff values should not be taken as definitive guidelines and are somewhat arbitrary. Nevertheless, the discussion of central symptoms that follows should be interpreted with appropriate caution and regarded as exploratory. Similarly, since the accuracy analysis does not support the edge weights being significantly different from one another, interpretation of the order of the edges in the network should also be done with care.

The strongest edge in the network was between item one (I have been able to laugh and see the funny side of things) and item two (I have looked forward to things with enjoyment). Other network analyses of maternal depression during pregnancy have similarly reported this edge as among the strongest in their network.²⁰ This link between humor and pleasure experience may hold promise for understanding symptom–symptom interactions in PND. For instance, targeting humor as a point of intervention may be a fruitful line of inquiry toward improving mood or pleasure experience. Though the present findings do not allow for causal inferences, they reveal a key association that researchers and clinicians may leverage to better understand and target the humor-pleasure link. Notably, while reviews examining humor-based interventions and symptoms and outcomes related to depression have highlighted some promising results, the overall evidence remains mixed, and such studies need to be viewed in light of variation in design and operationalization.^50,51 Thus, rigorous research is needed to establish the clinical utility of these findings. Furthermore, this association between pleasure and humor processing is not new; research has found that impaired humor appreciation is associated with an inability to anticipate pleasure among individuals with social anhedonia.⁵² These findings raise interesting questions about the mechanics of the humor-pleasure relationship, and interventions targeting maternal depression may thus benefit from a deeper understanding of this association.

Additionally, results indicated that item nine (crying), followed by item two (looked forward/enjoyment) and item eight (sad/miserable), were comparatively higher-ranked nodes in the strength centrality estimates. In previous studies with pregnant and/or postpartum women, item eight (sad/miserable) has been found to have the highest strength centrality among pregnant mothers.²⁰ Thus, while the present findings regarding the relevance of mood-related symptoms align with prior research, caution is warranted before extending or generalizing these results to the broader network literature on PND.

The observed centrality of “crying,” “sad mood,” and “looking forward to things with enjoyment”—though an interesting exploratory finding—should be viewed as a preliminary trend rather than robust findings, given the low CS-coefficient. Notably, depressed mood and anhedonia are core features of depression, and at least one of them is required for a diagnosis.^52,53 Interestingly, these three nodes were also connected to each other in the present network. Given that prior research has found that depressed mood and anhedonia not only predict each other over time but that the experience of one may make the other more likely at a later point in time,⁵⁴ this interconnectedness warrants further investigation. It may offer promising avenues for clinical interventions.

Other findings from the present study were associations of “crying” and “looking forward with enjoyment” with “I have been anxious or worried for no good reason” (item four). Offering an evidence-backed call for further exploration of these symptom interactions, previous research has hinted at a pattern of mutual reinforcement of depression and anxiety symptoms throughout development.⁵⁵ It has even been proposed that anxiety may lead to depression through anhedonia, as high anxiety levels can fuel avoidance behaviors that deplete enjoyment and contribute to depressive symptoms.⁵⁶ Future research may benefit from focusing on the intersection of depressive, anhedonia, and anxiety-related processes, and how they influence and possibly maintain one another.

Thus, in the present network, mood and anhedonia symptoms appeared elevated in centrality—an interesting preliminary finding, given that these symptoms are core features of major depressive disorder.⁵³

A CFA was also performed to assess the underlying factor structure of the EPDS. Consistent with prior literature,^48,49 the CFA results in our sample also supported a multidimensional EPDS structure. This is relevant for network-analytic work because, while partial correlation networks estimate unique associations between two items after conditioning on all other observed variables, they “retain shared variance due to outside sources that cannot fully be partialed out by controlling for other variables in the network” (p.4).⁴² As such, this possibility should be taken into account when interpreting the associations estimated in psychological networks.

Notably, a significant sociocultural finding from the present study warrants emphasis: with a mean age of 17.9 years, a considerable proportion of the sample had adolescent pregnancies. A recent review by Lesinskienė et al. highlights the substantial mental health risks associated with adolescent pregnancy, including high rates of depression, anxiety, and suicidal ideation, as well as multiple psychosocial risk factors that contribute to a complex mental health landscape in this population.⁵⁷ Empirical evidence further suggests that factors associated with PND differ between adolescents and adults; for instance, adolescents may be particularly affected by low knowledge of postpartum complications.⁵⁸ These sociocultural considerations are particularly relevant when interpreting the network structure of the present sample, as network models are not restricted to a single level of explanation, such as only biological or psychological factors,¹⁰ and mental disorders may, to an extent, be shaped by external mechanisms.²⁷

This study is one of the few studies employing a network analysis approach to mental disorders in India. Furthermore, a community-based sample from rural Bihar, comprising women flagged by the EPDS—a standardized screening tool—was selected, which more closely mirrors the actual clinical picture of PND as opposed to a psychologically-healthy sample. While this is a major strength of the present study, it also has limitations.

The low stability of strength centrality indicates that the present findings should be interpreted with appropriate caution. Replication in larger or more diverse samples is recommended to confirm the relative importance of symptoms. Because the present study specifically used a symptom severity threshold (EPDS ≥ 10) as an inclusion criterion, the findings may be subject to Berkson’s bias—a key methodological concern in clinical network research.⁵⁹ Consequently, the results should be interpreted with caution, as the sample was selected based on a severity threshold calculated from the symptoms, which also comprise the nodes in the network. In other words, the network rests on a selection criterion defined by a function of the same variables included in the analysis, raising concerns about interpretability.⁵⁹

It should be noted that the EPDS is a screening tool, not a diagnostic one. Additionally, random measurement error may impact network structure,⁶⁰ and unmeasured confounding variables may similarly contribute to spurious edges.⁶¹ Growing literature cautions against interpreting centrality indices without accounting for latent confounding, as this may lead to misrepresentations of symptom relationships.⁶² Although the EPDS was originally proposed as a unidimensional measure,³² subsequent works have frequently reported a multidimensional structure, including depression, anxiety, and anhedonia factors,^63–65 suggesting the presence of latent variables. As Hallquist and colleagues note, a highly central node may denote a symptom strongly related to the underlying latent construct rather than signify its importance as a symptom in the network.⁶² In addition, the duration of depression symptoms was not separately assessed, even though all women flagged by the EPDS were given the appropriate intervention, including pharmacological and non-pharmacological treatment. Given the small sample size, the sparsity of the observed network may have resulted from the small sample size relative to the number of nodes.⁴² Cross-sectional data also restrict the ability to make causal inferences. Given these constraints, replication studies with larger samples are imperative.

Conclusion

This study makes an important contribution to the growing network analysis literature, particularly in the realm of PND. Having highlighted important symptoms in the present PND network, this study paves the way for generating evidence-based clinical interventions. However, centrality by itself is not directly indicative of clinical relevance and must be interpreted in the context of other relevant information about the sample and the network.⁶⁶ Building on the present study, future research may assess the relative efficacy of interventions targeting central versus peripheral symptoms and more traditional intervention options.⁸ Further insights may be gained by gauging the utility, clinical relevance, and amenability to change of these central symptoms to offer up targets for PND interventions to suit a rural Indian context. The need for additional work in this area is imperative, given the heavy toll exacted by PND on patients and families alike. Recent studies from India have called for increased PND awareness among expecting mothers and their family members,⁶⁷ and the findings from this study, upon appropriate replication, may offer initial directions for grassroots-level psychoeducational endeavors as well.

Supplemental Material

Supplemental material for this article is available online.

Footnotes

Acknowledgements

Nil.

Consent to Participate

Participants provided written informed consent in the local language, that is, Hindi, before being recruited into the study.

Data Availability Statement

The datasets generated during and/or analyzed during the current study are available from the corresponding author on reasonable request.

Declaration of Conflicting Interests

The authors declared no potential conflicts of interest with respect to the research, authorship, and/or publication of this article.

Declaration Regarding the Use of Generative AI

No part of this article was written or generated by a generative AI tool. The authors take full responsibility for the accuracy, integrity, and originality of the published article.

Ethical Approval

Institutional ethics committee approval was obtained before the start of the study. The study protocol was approved by the Institutional Review Board and Ethics Committee of Schizophrenia Research Foundation, Chennai, Tamil Nadu, India: (SRF-CR/04/May-2019).

Funding

The authors disclosed receipt of the following financial support for the research, authorship, and/or publication of this article: Grand Challenges Exploration (GCE-USA), Round 21, Bill and Melinda Gates Foundation.

Citation Diversity Statement

Recent work across academic disciplines has identified a bias in citation practices such that papers authored by underrepresented groups are cited at lower rates relative to their presence in the literature. We acknowledge this bias and made efforts to cite work from a diverse range of researchers. We acknowledge the limitations of our approach and commit to continued attention to equitable citation practices.

References

Shrivastava

, Shrivastava

and Ramasamy

Antenatal and postnatal depression: A public health perspective. J Neurosci Rural Pract, 2015; 6: 116–119.

Woody

, Ferrari

, Siskind

, . A systematic review and meta-regression of the prevalence and incidence of perinatal depression. J Affect Disord, 2017; 219: 86–92.

Raghavan

, Khan

, Seshu

, . Prevalence and risk factors of perinatal depression among women in rural Bihar: A community-based cross-sectional study. Asian J Psychiatry, 2021; 56: 102552.

Grigoriadis

, VonderPorten

, Mamisashvili

, . The impact of maternal depression during pregnancy on perinatal outcomes: A systematic review and meta-analysis. J Clin Psychiatry, 2013; 74: e321–e341.

Da Costa

, Dritsa

, Rippen

, . Health-related quality of life in postpartum depressed women. Arch Womens Ment Health, 2006; 9: 95–102.

Dennis

and McQueen

Does maternal postpartum depressive symptomatology influence infant feeding outcomes?

Acta Paediatr, 2007; 96: 590–594.

Priel

, Zeev-Wolf

, Djalovski

, . Maternal depression impairs child emotion understanding and executive functions: The role of dysregulated maternal care across the first decade of life. Emotion, 2020; 20: 1042–1058.

Santos

, Fried

, Asafu-Adjei

, . Network structure of perinatal depressive symptoms in Latinas: Relationship to stress and reproductive biomarkers. Res Nurs Health, 2017; 40: 218–228.

Beard

, Millner

, Forgeard

MJC

, . Network analysis of depression and anxiety symptom relationships in a psychiatric sample. Psychol Med, 2016; 46: 3359–3369.

10.

Borsboom

. A network theory of mental disorders. World Psychiatry, 2017; 16: 5–13.

11.

Borsboom

and Cramer

AOJ

. Network analysis: An integrative approach to the structure of psychopathology. Annu Rev Clin Psychol, 2013; 9: 91–121.

12.

Shin

and Park

Perceived stress shapes symptom and social network dynamics: A network analysis of depression, anxiety, and relationship-specific support and strain. BMC Psychiatry, 2025; 25: 715.

13.

McNally

, Robinaugh

, Deckersbach

, . Estimating the symptom structure of bipolar disorder via network analysis: Energy dysregulation as a central symptom. J Psychopathol Clin Sci, 2022; 131: 86–97.

14.

Opsahl

, Agneessens

and Skvoretz

Node centrality in weighted networks: Generalizing degree and shortest paths. Soc Netw, 2010; 32: 245–251.

15.

Fried

, Epskamp

, Nesse

, . What are ‘good’ depression symptoms? Comparing the centrality of DSM and non-DSM symptoms of depression in a network analysis. J Affect Disord, 2016; 189: 314–320.

16.

Malgaroli

, Calderon

and Bonanno

. Networks of major depressive disorder: A systematic review. Clin Psychol Rev, 2021; 85: 102000.

17.

Santos

, Tan

and Salomon

Heterogeneity in perinatal depression: How far have we come? A systematic review. Arch Womens Ment Health, 2017; 20: 11–23.

18.

Liu

, Huang

, Gao

, . Network analysis of depression and anxiety symptoms and their associations with cognitive fusion among pregnant women. BMC Psychiatry, 2025; 25: 537.

19.

Ebina

, Ito

, Yamaguchi

, . Stability and variability of edinburgh postnatal depression scale networks among postpartum women: Insights from the Japan environment and children’s study. J Affect Disord, 2025; 391: 119992.

20.

Zhang

, Sun

, Li

, . Maternal and paternal depression during pregnancy in China: Prevalence, correlates, and network analysis. Neuropsychiatr Dis Treat, 2021; 17: 2269–2280.

21.

, Zhu

, Li

, . Identification of core antenatal depressive symptoms: A network analysis. West J Nurs Res, 2025; 47: 732–741.

22.

Wasil

, Venturo-Conerly

, Shinde

, . Applying network analysis to understand depression and substance use in Indian adolescents. J Affect Disord, 2020; 265: 278–286.

23.

Wasil

, Gillespie

, Park

, . Which symptoms of depression and anxiety are most strongly associated with happiness? A network analysis of Indian and Kenyan adolescents. J Affect Disord, 2021; 295: 811–821.

24.

Sönmez

, Verdeli

, Malgaroli

, . Symptom networks of common mental disorders in public versus private healthcare settings in India. Camb Prisms Glob Ment Health, 2025; 12: e30.

25.

Takegata

, Ohashi

, Lazarus

, . Cross-national differences in psychosocial factors of perinatal depression: A systematic review of India and Japan. Healthcare, 2017; 5: 91.

26.

Shivalli

and Gururaj

Postnatal depression among rural women in South India: Do socio-demographic, obstetric and pregnancy outcome have a role to play?

PLoS One, 2015; 10: e0122079.

27.

Borsboom

, Cramer

AOJ

and Kalis

Brain disorders? Not really: Why network structures block reductionism in psychopathology research. Behav Brain Sci, 2019; 42: e2.

28.

von Elm

, Altman

, Egger

, . The strengthening the reporting of observational studies in epidemiology (STROBE) statement: Guidelines for reporting observational studies. Lancet, 2007; 370: 1453–1457.

29.

Chandramouli

and General

Census of India 2011. Provisional population totals. New Delhi: Government of India, pp.409–413.

30.

PRS Legislative Research. Bihar Budget Analysis 2025–2026. New Delhi: PRS Legislative Research; 2025 [cited 2025 Aug 17]. Available from: https://prsindia.org/reports-summaries/bihar-budget-analysis-2025-26.

31.

Ministry of Health and Family Welfare. National Health Policy 2017. New Delhi: Government of India; 2017 [cited 2025 Aug 17]. Available from: https://nhsrcindia.org/sites/default/files/2021-07/National%20Health%20Policy%202017%20%28English%29%20.pdf.

32.

Cox

, Holden

and Sagovsky

Detection of postnatal depression: Development of the 10-item Edinburgh postnatal depression scale. Br J Psychiatry, 1987; 150: 782–786.

33.

Dubey

, Gupta

, Bhasin

, . Prevalence and associated risk factors for postpartum depression in women attending a tertiary hospital, Delhi, India. Int J Soc Psychiatry, 2012; 58: 577–580.

34.

Milgrom

and Gemmill

. Screening for perinatal depression. Best Pract Res Clin Obstet Gynaecol, 2014; 28: 13–23.

35.

Joshi

, Lyngdoh

and Shidhaye

Validation of Hindi version of Edinburgh postnatal depression scale as a screening tool for antenatal depression. Asian J Psychiatry, 2020; 48: 101919.

36.

Team

RDC

. R: A language and environment for statistical computing. Vienna, Austria: R Foundation for Statistical Computing, 2009.

37.

Epskamp

, Cramer

AOJ

, Waldorp

, . qgraph: Network Visualizations of Relationships in Psychometric Data. J Stat Softw, 2012; 48(4): 1–18.

38.

Fruchterman

TMJ

and Reingold

. Graph drawing by force-directed placement. Softw Pract Exp, 1991; 21: 1129–1164.

39.

Lauritzen

. Graphical models. Oxford : New York: Clarendon Press; Oxford University Press, 1996.

40.

Tibshirani

. Regression shrinkage and selection via the LASSO. J R Stat Soc Ser B Stat Methodol, 1996; 58: 267–288.

41.

Epskamp

, Borsboom

and Fried

. Estimating psychological networks and their accuracy: A tutorial paper. Behav Res Methods, 2018; 50: 195–212.

42.

Epskamp

and Fried

. A tutorial on regularized partial correlation networks. Psychol Methods, 2018; 23: 617–634.

43.

Chen

and Chen

Extended Bayesian information criteria for model selection with large model spaces. Biometrika, 2008; 95: 759–771.

44.

Friedman

, Hastie

and Tibshirani

Sparse inverse covariance estimation with the graphical lasso. Biostatistics, 2008; 9: 432–441.

45.

McNally

. Can network analysis transform psychopathology? Behav Res Ther, 2016; 86: 95–104.

46.

Bringmann

, Elmer

, Epskamp

, . What do centrality measures measure in psychological networks?. J Abnorm Psychol, 2019; 128: 892–903.

47.

Franco-Martínez

, Alvarado

and Sorrel

. Range restriction affects factor analysis: Normality, estimation, fit, loadings, and reliability. Educ Psychol Meas, 2023; 83: 262–293.

48.

Škodová

, Bánovčinová

, Urbanová

, . Factor structure of the edinburgh postnatal depression scale in a sample of postpartum slovak women. Int J Environ Res Public Health, 2021; 18: 6298.

49.

Phillips

, Charles

, Sharpe

, . Validation of the subscales of the Edinburgh postnatal depression scale in a sample of women with unsettled infants. J Affect Disord, 2009; 118: 101–112.

50.

Ayisire

, Babalola

, Aladum

, . A comprehensive review on the effects of humor in patients with depression. Cureus, 2022; 14(9): e29263. DOI: 10.7759/cureus.29263.

51.

Sarink

FSM

and García-Montes

. Humor interventions in psychotherapy and their effect on levels of depression and anxiety in adult clients: A systematic review. Front Psychiatry, 2023; 13: 1049476.

52.

Liu

, Huang

, Shan

, . Humor processing deficits in individuals with social anhedonia. Psychiatry Res, 2019; 275: 345–350.

53.

American Psychiatric Association. Diagnostic and statistical manual of mental disorders. Fifth Edition. American Psychiatric Association, 2013.

54.

Bennik

, Nederhof

, Ormel

, . Anhedonia and depressed mood in adolescence: Course, stability, and reciprocal relation in the TRAILS study. Eur Child Adolesc Psychiatry, 2014; 23: 579–586.

55.

McElroy

, Fearon

, Belsky

, . Networks of depression and anxiety symptoms across development. J Am Acad Child Adolesc Psychiatry, 2018; 57: 964–973.

56.

Winer

, Bryant

, Bartoszek

, . Mapping the relationship between anxiety, anhedonia, and depression. J Affect Disord, 2017; 221: 289–296.

57.

Lesinskienė

, Andruškevič

, Butvilaitė

. Adolescent pregnancies and perinatal mental health—needs and complex support options: A literature review. J Clin Med, 2025; 14: 2334.

58.

Kassa

, Batchelder

and Gross

Prevalence and determinants of postpartum depression among adolescent and adult mothers in Northwest Ethiopia. Res Nurs Health, 2024; 47: 125–140.

59.

De Ron

, Fried

and Epskamp

Psychological networks in clinical populations: Investigating the consequences of Berkson’s bias. Psychol Med, 2021; 51: 168–176.

60.

Borsboom

, Deserno

, Rhemtulla

, . Network analysis of multivariate data in psychological science. Nat Rev Methods Primer, 2021; 1: 58.

61.

Hevey

. Network analysis: A brief overview and tutorial. Health Psychol Behav Med, 2018; 6: 301–328.

62.

Hallquist

, Wright

AGC

and Molenaar

PCM

. Problems with centrality measures in psychopathology symptom networks: Why network psychometrics cannot escape psychometric theory. Multivar Behav Res, 2021; 56: 199–223.

63.

Coates

, Ayers

and De Visser

. Factor structure of the Edinburgh postnatal depression scale in a population-based sample. Psychol Assess, 2017; 29: 1016–1027.

64.

Matsumura

, Hamazaki

, Tsuchida

, . Factor structure of the Edinburgh postnatal depression scale in the Japan environment and children’s study. Sci Rep, 2020; 10: 11647.

65.

Tuohy

and McVey

Subscales measuring symptoms of non-specific depression, anhedonia, and anxiety in the Edinburgh postnatal depression scale. Br J Clin Psychol, 2008; 47: 153–169.

66.

Fried

, Eidhof

, Palic

, . Replicability and generalizability of posttraumatic stress disorder (PTSD) networks: A cross-cultural multisite study of PTSD symptoms in four trauma patient samples. Clin Psychol Sci, 2018; 6: 335–351.

67.

Lodha

, Jahangir

, Karia

, . Perceptions of perinatal depression among low-income mothers and families in Mumbai, India. Asian J Psychiatry, 2022; 71: 103048.

Supplementary Material

Please find the following supplemental material available below.

For Open Access articles published under a Creative Commons License, all supplemental material carries the same license as the article it is associated with.

For non-Open Access articles published, all supplemental material carries a non-exclusive license, and permission requests for re-use of supplemental material or any part of supplemental material shall be sent directly to the copyright owner as specified in the copyright notice associated with the article.

0.00 MB

0.02 MB