Network structure of depression and anxiety symptoms with psychosocial factors among Chinese women with primary infertility: a multi-center cross-sectional study

Abstract

Background

Women with primary infertility experience elevated psychological distress, yet the interplay between individual symptoms and psychosocial factors remains unclear. This study aimed to identify central and bridge symptoms within a network of depression, anxiety, and psychosocial factors in Chinese women with primary infertility.

Methods

A total of 1,062 women with primary infertility were recruited from four reproductive medicine centers in China between February and December 2025. Depression (PHQ-9), anxiety (GAD-7), perceived stress (PSS-10), fertility-specific stress (COMPI-FPSS-SF), infertility stigma (ISS), social support (MSPSS), and resilience (CD-RISC-10) were assessed. Network analysis using regularized partial correlation networks was conducted to estimate network structure, centrality, bridge centrality, and predictability.

Results

Of the participants, 9.6% met clinical thresholds for depression and 5.3% for anxiety. Difficulty relaxing exhibited the highest centrality among symptoms. Public stigma showed the strongest centrality among psychosocial factors, with a robust connection to family stigma (edge weight = 0.538). Perceived stress emerged as the primary bridge linking psychosocial factors to symptoms (bridge expected influence = 0.297), followed by fertility-specific stress (bridge expected influence = 0.143). Guilt was the key bridge symptom (bridge expected influence = 0.086). Resilience showed the strongest negative association with perceived stress (edge weight = − 0.275). Network stability was excellent (CS-coefficient = 0.75).

Conclusions

Difficulty relaxing, perceived stress, and guilt emerged as potentially important nodes that warrant further investigation as candidate intervention targets for psychological distress in women with primary infertility. Interventions targeting relaxation training, stress appraisal, and self-compassion, alongside resilience enhancement, warrant further investigation.

Clinical trial number

Not applicable.

Supplementary Information

The online version contains supplementary material available at 10.1186/s12888-026-08156-0.

Introduction

Infertility, defined as the failure to achieve a clinical pregnancy after 12 months or more of regular unprotected sexual intercourse, has emerged as a significant global public health concern [1]. According to the World Health Organization, approximately 17.5% of the adult population worldwide, affecting nearly 1 in 6 individuals of reproductive age, experiences infertility [2]. In China, the prevalence is particularly striking, with estimates indicating that 15.5% to 25% of reproductive-age couples are affected [3, 4]. Notably, primary infertility, the condition wherein a woman has never achieved pregnancy despite attempting conception, accounts for a substantial proportion of these cases and may confer distinct psychological challenges compared to secondary infertility [5, 6]. Epidemiological evidence suggests that women with primary infertility exhibit significantly higher rates of depression and lower quality of life compared to those with secondary infertility [6, 7], underscoring the need for dedicated investigation of this population.

Beyond its reproductive implications, infertility imposes a profound psychological burden on affected women. Meta-analytic evidence indicates that the global prevalence of depressive symptoms among infertile women is 42%, with anxiety affecting approximately 41% of this population [6]. Compared to fertile women, those experiencing infertility demonstrate a 1.63-fold increased risk of psychological distress and a 1.40-fold elevated risk of depression [8]. Furthermore, infertility-related stress, characterized by overwhelming mental and emotional strain arising from the desire but inability to conceive and from associated treatment procedures, affects up to 78.8% of infertile women globally [9]. This stress manifests across multiple domains, including social concerns, sexual concerns, relationship strain, and preoccupation with parenthood [10].

In the Chinese sociocultural context, infertility carries particularly salient stigma. Traditional beliefs emphasizing filial piety and the continuation of family lineage remain deeply embedded, with childlessness often perceived as a failure of women’s primary social role [11]. Research indicates that 53% to 64% of infertile women worldwide experience stigma, with Chinese women reporting moderate to high levels that manifest primarily as self-devaluation and social withdrawal [12, 13]. This stigma not only directly impairs mental health but also intensifies infertility-related stress and reduces quality of life [14]. The internalization of negative societal attitudes creates a self-reinforcing cycle wherein stigmatized women may withdraw from social interactions and experience compounded psychological distress [15].

Although considerable research has examined risk factors for psychological distress in infertile women, understanding protective factors is equally important for developing comprehensive interventions. Social support, encompassing emotional, informational, and instrumental assistance from family, friends, and significant others, has consistently demonstrated protective effects against depression and anxiety in this population [16]. Similarly, psychological resilience, defined as the capacity to adapt positively in the face of adversity, plays a crucial role in mediating the impact of infertility-related stressors on mental health outcomes [16]. Studies have shown that resilience mediates the relationship between infertility stress and psychological well-being, suggesting that enhancing resilience may represent a viable intervention target [17]. However, the specific mechanisms through which these protective factors interact with psychological symptoms at a granular level remain incompletely understood.

Traditional analytical approaches to understanding psychological distress in infertility have predominantly relied on latent variable models, which conceptualize depression and anxiety as unitary constructs measured through aggregated symptom scores [18]. While informative, such approaches obscure the heterogeneous nature of these conditions and fail to capture the complex, dynamic interactions among individual symptoms, which may have differential clinical implications [18, 19]. The aggregation of symptoms into total scores inevitably results in information loss regarding which specific symptoms are most influential in driving and maintaining psychological distress.

Network analysis offers an alternative framework grounded in the network theory of psychopathology [20], which posits that mental disorders arise from direct causal interactions among symptoms rather than reflecting latent disease entities. In this approach, symptoms are represented as nodes and their statistical relationships as edges [21]. Central symptoms, those with the strongest connections to other symptoms in the network, may play pivotal roles in the onset and maintenance of psychological distress, making them priority targets for intervention [19]. Bridge symptoms, which connect different symptom clusters or psychological domains, are thought to drive comorbidity and may explain why depression, anxiety, and stress frequently co-occur in clinical populations [22]. This symptom-level approach thus provides actionable insights that aggregate-score analyses cannot offer.

Recent applications of network analysis to infertility mental health have begun to illuminate the architecture of psychological distress in this population, yet important gaps remain. Cao et al. examined depression and anxiety networks in 740 Chinese infertile women, identifying restlessness as the most central symptom and guilt as a culturally significant node connected to suicidal ideation [18]. Wu and colleagues [23] extended this work longitudinally, demonstrating that central and bridge symptoms shift dynamically across IVF treatment stages. Liu et al. [24] investigated stress and stigma networks, finding that relationship concern and public stigma emerged as central symptoms with strong associations to quality of life. Despite these contributions, three critical limitations characterize the existing literature. First, prior studies have examined either depression and anxiety symptom networks or psychosocial factor networks in isolation, without simultaneously integrating risk factors and protective factors within a unified framework. Second, existing investigations have generally combined primary and secondary infertility populations, despite evidence that primary infertility confers distinct psychological challenges [6, 7]. Third, no study has examined how protective factors such as social support and resilience interact with psychological symptoms at the individual symptom level, leaving the buffering mechanisms through which these factors may attenuate distress poorly understood.

It should also be noted that the broader mental health research landscape has witnessed rapid methodological diversification. Computational approaches such as machine learning and deep learning have been applied to automated detection and classification of depression using behavioral, linguistic, or physiological signals [25, 26]. Recent innovations include adversarial domain adaptation techniques that address cross-corpus generalization challenges in EEG-based depression recognition [27], and multimodal fusion architectures that integrate heterogeneous data sources to improve classification accuracy [25]. While these methods offer promising avenues for screening and monitoring, they serve a fundamentally different purpose from network analysis. Machine learning approaches primarily address the question of whether an individual is depressed, whereas network analysis addresses the question of how symptoms and psychosocial factors are structurally interconnected, thereby generating hypotheses about where to intervene. Network analysis thus occupies a unique niche within the computational toolkit for mental health research, and the relationship between these complementary paradigms is examined in the Discussion.

To address the identified gaps, the present study employed network analysis to examine the interrelationships among depression symptoms, anxiety symptoms, perceived stress, infertility-related stress, stigma, social support, and resilience in a large, multi-center sample of Chinese women with primary infertility. The present study extends the existing literature in three ways: (1) it simultaneously integrates both risk factors (perceived stress, fertility-specific stress, and four dimensions of infertility stigma) and protective factors (three dimensions of social support and resilience) alongside individual depression and anxiety symptoms within a single network, enabling examination of cross-domain dynamics; (2) it focuses exclusively on women with primary infertility, a subpopulation with potentially distinct psychological profiles that has not been examined separately in prior network studies; and (3) it employs a multi-center design across four reproductive medicine centers to enhance sample representativeness. Specifically, we aimed to: (a) estimate the network structure of psychological symptoms and psychosocial factors, identifying central nodes; (b) detect bridge nodes that connect psychological symptom clusters with psychosocial factor clusters; and (c) explore how risk and protective factors simultaneously relate to depression and anxiety symptoms within an integrated network framework.

Methods

Study design and participants

This multi-center cross-sectional study was conducted at the Reproductive Medicine Centers of four tertiary hospitals in China from February to December 2025. The inclusion criteria were: (1) women aged 20–45 years; (2) diagnosed with primary infertility, defined as the inability to conceive after 12 months of regular unprotected intercourse without any prior pregnancy [5]; (3) ability to read and understand Chinese; and (4) willingness to provide informed consent. The exclusion criteria were: (1) secondary infertility (history of previous pregnancy); (2) current pregnancy; (3) diagnosed psychiatric disorders currently requiring medication; and (4) severe cognitive impairment affecting questionnaire completion.

A total of 1,106 eligible participants were initially recruited using convenience sampling. After data quality control, including removal of incomplete responses (n = 44), 1,062 participants were retained for the final analysis. The sample size substantially exceeded the recommended minimum of 10 observations per estimated parameter for network analysis [21], yielding a sample-to-node ratio of 40.8:1. This study was approved by the Ethics Committees of four participating hospitals. Written informed consent was obtained from all participants.

Measures

Patient Health Questionnaire-9 (PHQ-9)

The PHQ-9 is a nine-item self-report instrument based on DSM-IV criteria that assesses depressive symptoms over the past two weeks [28]. Items evaluate anhedonia (PHQ1), depressed mood (PHQ2), sleep disturbance (PHQ3), fatigue (PHQ4), appetite changes (PHQ5), guilt/worthlessness (PHQ6), concentration difficulties (PHQ7), psychomotor changes (PHQ8), and suicidal ideation (PHQ9). Each item is rated on a 4-point Likert scale from 0 (not at all) to 3 (nearly every day), with total scores ranging from 0 to 27. A cut-off score of ≥ 10 indicates clinically significant depressive symptoms. The Chinese version has demonstrated excellent psychometric properties [29]. In this study, Cronbach’s α was 0.873.

Generalized Anxiety Disorder-7 (GAD-7)

The GAD-7 is a seven-item instrument measuring generalized anxiety symptoms over the past two weeks [30]. Items assess nervousness (GAD1), uncontrollable worry (GAD2), excessive worry (GAD3), difficulty relaxing (GAD4), restlessness (GAD5), irritability (GAD6), and fearfulness (GAD7). Items are rated on a 4-point scale from 0 (not at all) to 3 (nearly every day), with total scores ranging from 0 to 21. A cut-off score of ≥ 10 indicates clinically significant anxiety symptoms. The Chinese version has been validated in clinical populations [31]. In this study, Cronbach’s α was 0.918.

Perceived Stress Scale-10 (PSS-10)

The PSS-10 measures the degree to which individuals perceive situations in their life as stressful over the past month [32]. Items are rated on a 5-point scale from 0 (never) to 4 (very often), with items 4, 5, 7, and 8 reverse-scored. Total scores range from 0 to 40, with higher scores indicating greater perceived stress. The Chinese version has demonstrated good reliability and validity [33]. In this study, Cronbach’s α was 0.816.

Copenhagen Multi-center Psychosocial Infertility-Fertility Problem Stress Scale-Short Form (COMPI-FPSS-SF)

The COMPI-FPSS-SF is a nine-item scale developed by Sobral et al. specifically designed to assess fertility-related stress [34]. Items evaluate stress related to infertility across personal, marital, and social domains. Two items are rated on a 5-point Likert scale (1 = strongly disagree to 5 = strongly agree), while the remaining seven items use a 4-point scale (1 = not at all to 4 = a great deal). Total scores range from 9 to 38, with higher scores indicating greater fertility-related stress. The Chinese version has been validated in infertile populations [35]. In this study, Cronbach’s α was 0.868.

Infertility Stigma Scale (ISS)

The ISS is a 27-item instrument developed and validated specifically for Chinese infertile women measuring perceived stigma across four dimensions [36]: self-devaluation (7 items), social withdrawal (5 items), public stigma (9 items), and family stigma (6 items). Items are rated on a 5-point Likert scale from 1 (strongly disagree) to 5 (strongly agree), with higher scores indicating greater perceived stigma. The scale has demonstrated excellent psychometric properties, with Cronbach’s α of 0.94, split-half reliability of 0.90, and test-retest reliability of 0.91. In the present study, Cronbach’s α for the total scale was 0.978.

Multidimensional Scale of Perceived Social Support (MSPSS)

The MSPSS is a 12-item scale measuring perceived social support from three sources [37]: family (items 3, 4, 8, 11), friends (items 6, 7, 9, 12), and significant others (items 1, 2, 5, 10). Items are rated on a 7-point scale from 1 (very strongly disagree) to 7 (very strongly agree), with higher scores indicating greater perceived support. The Chinese version has demonstrated good psychometric properties [38]. In this study, the overall Cronbach’s α was 0.957.

Connor-Davidson Resilience Scale-10 (CD-RISC-10)

The CD-RISC-10 is a brief, unidimensional measure of psychological resilience [39]. Items are rated on a 5-point scale from 0 (not true at all) to 4 (true nearly all the time), with total scores ranging from 0 to 40. Higher scores indicate greater resilience. The Chinese version has been validated in infertile couples with good reliability and validity [40]. In this study, Cronbach’s α was 0.965.

Statistical analysis

Network estimation

We employed Gaussian Graphical Models (GGM) to estimate the network structure. The graphical LASSO (Least Absolute Shrinkage and Selection Operator) algorithm with Extended Bayesian Information Criterion (EBIC) model selection was used to estimate sparse network structures while minimizing spurious edges [21]. The tuning hyperparameter γ was set to 0.25, which provides a balance between sensitivity and specificity in edge detection. Due to the ordinal nature of symptom-level items, Spearman correlations were used as input for network estimation.

The network comprised 26 nodes organized into two communities: (1) a symptom community containing nine PHQ-9 items and seven GAD-7 items (nodes 1–16); and (2) a psychosocial factors community containing perceived stress (PSS-10 total), fertility-specific stress (COMPI-FPSS-SF total), four infertility stigma dimensions (self-devaluation, social withdrawal, public stigma, family stigma), three social support dimensions (family, friends, significant others), and resilience (CD-RISC-10 total) (nodes 17–26).

The use of a mixed-level approach, combining item-level symptom nodes with scale-level psychosocial factor nodes, warrants methodological consideration. This strategy was adopted because our primary research question concerned the interplay between specific symptoms and broader psychosocial constructs. Including all individual items from all psychosocial measures would have resulted in a network exceeding 80 nodes, which would compromise both interpretability and statistical power. This approach follows established precedent in network analysis literature [41] and has been employed in a comparable study of infertility-related psychological distress [18]. Nevertheless, scale-level composites typically exhibit higher reliability and different distributional properties compared to individual items, which may influence edge weight estimation and bridge centrality metrics. All network analyses were conducted using the bootnet package and qgraph package in R (version 4.5.0).

Centrality analysis

Expected influence (EI) was computed to identify the most influential nodes within the network. EI represents the sum of signed edge weights connected to a node, accounting for both positive and negative connections [42]. Unlike strength centrality, EI is particularly informative when negative edges exist in the network, as it captures the direction of influence. All centrality indices were standardized (z-scores) for visualization and comparison.

Bridge centrality analysis

Bridge expected influence (BEI) was computed to identify symptoms that connect different communities within the network [22]. Two communities were defined a priori based on theoretical considerations: (1) symptom community (PHQ-9 and GAD-7 items); and (2) psychosocial factors community (stress, stigma, support, and resilience variables). Nodes with high positive BEI values represent pathways through which activation may spread from one community to another, making them potential intervention targets. Bridge centrality was calculated using the networktools package in R.

Predictability analysis

Node predictability (R²) was estimated using mixed graphical models to quantify the proportion of variance in each node explained by its neighboring nodes [43]. Predictability provides information about the controllability of each node, nodes with higher predictability are more strongly determined by other network variables and may be more amenable to intervention through their neighbors. Predictability was computed using the mgm package in R.

Network stability and accuracy

Network robustness was evaluated using two bootstrap procedures with 1,000 iterations each [21]. First, nonparametric bootstrapping was conducted to estimate 95% confidence intervals for edge weights, providing information about edge estimation accuracy. Second, case-dropping bootstrap was performed to assess the stability of centrality indices by repeatedly estimating networks with subsets of cases. The correlation stability coefficient (CS-coefficient) was calculated, representing the maximum proportion of cases that can be dropped while maintaining a correlation of at least 0.7 with the original centrality ordering. According to established guidelines, CS-coefficients above 0.50 indicate excellent stability, values between 0.25 and 0.50 indicate acceptable stability, and values below 0.25 suggest unreliable centrality estimates.

Additionally, split-half cross-validation was conducted to assess network replicability. The full sample was randomly divided into two equal subsets (n₁ = 531, n₂ = 531), and network estimation was performed independently on each subset using identical estimation procedures. The similarity between the two resulting networks was quantified using the Spearman correlation coefficient between the vectorized edge weight matrices. This procedure was repeated 100 times with different random partitions, and the mean and standard deviation of the resulting correlation coefficients were reported. Furthermore, bootstrapped difference tests were performed using the nonparametric bootstrap samples to evaluate whether the observed differences in edge weights and expected influence between specific nodes were statistically significant. For each pair of nodes or edges, the difference in the statistic of interest was computed across 1,000 bootstrap iterations, and 95% bootstrapped confidence intervals were obtained. A difference was considered statistically significant if the corresponding 95% confidence interval did not include zero [21].

Results

Sample characteristics

A total of 1,062 women with primary infertility were included in the final analysis. Participants were predominantly in their early thirties (M = 33.57 years, SD = 4.60), with over half having attained college or university education (54.9%). The majority resided in urban areas (68.9%) and were employed (81.5%). Most women were in their first marriage (86.3%), with an average marriage duration of 5.62 years (SD = 4.25). Regarding infertility history, the most common duration was 1 to 3 years (43.7%), and approximately one-third had undergone assisted reproductive treatments (IUI: 23.3%; IVF: 31.1%). Detailed demographic and clinical characteristics are presented in Table 1.

Table 1.

Sociodemographic and clinical characteristics (N = 1,062)

Characteristic	n (%) or M ± SD
Age (years)	33.57 ± 4.60
Education level
Junior high school or below	289 (27.2)
High school/Technical secondary	190 (17.9)
College/University or above	583 (54.9)
Employment status
Employed	865 (81.5)
Unemployed/Housewife	197 (18.5)
Residence
Urban	732 (68.9)
Rural	330 (31.1)
Monthly household income (CNY)
< 5,000	328 (30.9)
5,000–10,000	466 (43.9)
> 10,000	268 (25.2)
Medical insurance type
Employee medical insurance	677 (63.8)
Resident medical insurance	353 (33.2)
Uninsured	32 (3.0)
Marital status
First marriage	911 (86.3)
Remarriage	145 (13.7)
Duration of marriage (years)	5.62 ± 4.25
Duration of infertility
≤ 1 year	153 (14.4)
1–3 years	464 (43.7)
3–5 years	234 (22.0)
> 5 years	211 (19.9)
Cause of infertility
Female factor	395 (37.2)
Male factor	131 (12.3)
Combined factors	200 (18.8)
Unexplained	336 (31.6)
Previous ART treatment
IUI	246 (23.2)
IVF	330 (31.1)

CNY, Chinese Yuan; ART, Assisted reproductive technology; IUI, Intrauterine insemination; IVF, In vitro fertilization

Table 2 summarizes the descriptive statistics and reliability coefficients for all study measures. Depression and anxiety symptoms were generally low in this sample, with mean PHQ-9 and GAD-7 scores of 3.66 (SD = 4.30) and 2.66 (SD = 3.61), respectively. Nevertheless, 9.6% of participants met the clinical threshold for depression (PHQ-9 ≥ 10) and 5.3% for anxiety (GAD-7 ≥ 10), indicating a notable minority experiencing clinically significant symptoms. Among the four dimensions of infertility stigma, public stigma was most pronounced (M = 15.74, SD = 7.21), followed by self-devaluation (M = 12.93, SD = 5.84). Participants reported relatively high levels of social support across all sources and moderate-to-high resilience (M = 35.79, SD = 8.81). All instruments demonstrated good to excellent internal consistency.

Table 2.

Descriptive statistics and reliability coefficients for study measures (N = 1,062)

Measure	Items	M	SD	Cronbach’s α
Symptom Measures
Depression (PHQ-9)	9	3.66	4.30	0.873
Anxiety (GAD-7)	7	2.66	3.61	0.918
Psychosocial Factors
Perceived Stress (PSS-10)	10	27.14	4.83	0.816
Fertility-specific Stress (COMPI)	9	22.65	6.08	0.868
Infertility Stigma (ISS)	27	50.53	20.78	0.978
Self-devaluation	7	12.93	5.84
Social withdrawal	5	11.06	4.66
Public stigma	9	15.74	7.21
Family stigma	6	10.79	4.99
Social Support (MSPSS)	12	66.90	12.61	0.957
Family support	4	22.59	4.44
Friend support	4	21.74	4.71
Significant other support	4	22.57	4.35
Resilience (CD-RISC-10)	10	35.79	8.81	0.965

PHQ-9, Patient Health Questionnaire-9; GAD-7, Generalized Anxiety Disorder-7; PSS-10, Perceived Stress Scale-10; COMPI, Copenhagen Multi-center Psychosocial Infertility Fertility Problem Stress Scale; ISS, Infertility Stigma Scale; MSPSS, Multidimensional Scale of Perceived Social Support; CD-RISC-10, Connor-Davidson Resilience Scale-10

Network structure

Figure 1 illustrates the estimated network comprising 26 nodes. Of the 325 possible edges, 151 (46.5%) were nonzero after regularization. Cross-community connections were relatively sparse, with only 40 edges (26.5%) linking symptoms to psychosocial factors (Supplementary Table S1).

Fig. 1.

Network structure of depression-anxiety symptoms and psychosocial factors in women with primary infertility (N = 1,062). Blue edges represent positive associations; red edges represent negative associations. Edge thickness indicates the strength of the association. The network comprises 26 nodes: 16 symptom nodes (PHQ1–PHQ9, GAD1–GAD7) and 10 psychosocial factor nodes

The strongest associations emerged within the psychosocial factors community. Public stigma and family stigma exhibited the highest edge weight (0.538), followed by family support and significant other support (0.464), and friend support and significant other support (0.373). Self-devaluation and social withdrawal also showed a strong positive association (0.349). Within the symptom community, notable associations included anhedonia and depressed mood (0.279), sleep disturbance and fatigue (0.267), and difficulty relaxing and irritability (0.235).

The strongest negative edge was observed between perceived stress and resilience (− 0.275). Other negative associations included self-devaluation and resilience (− 0.070), and family stigma and family support (− 0.047). The complete list of strongest edges is provided in Supplementary Table S2.

Central symptoms

Centrality indices are presented in Fig. 2. GAD4 (difficulty relaxing) exhibited the highest strength (z = 1.539) and expected influence (1.195), followed by PHQ2 (depressed mood; strength = 1.347) and GAD3 (excessive worry; strength = 1.127). Among psychosocial factors, public stigma showed the highest strength (z = 1.106) and expected influence (0.708), followed by self-devaluation (strength = 0.879). Detailed centrality indices for all nodes are provided in Supplementary Table S3.

Fig. 2.

Centrality indices (Strength and Expected Influence) for all network nodes. Values are standardized (z-scores). Nodes are ordered by Expected Influence from highest to lowest. Higher values indicate greater centrality in the network

Bridge symptoms

Bridge expected influence (BEI) analysis results are displayed in Fig. 3. Perceived stress showed the highest positive BEI (0.297), followed by fertility-specific stress (BEI = 0.143), social withdrawal (BEI = 0.104), and PHQ6 guilt/worthlessness (BEI = 0.086). Among symptom nodes, PHQ6 (guilt/worthlessness) exhibited the highest positive BEI, followed by GAD1 (nervousness; BEI = 0.082) and GAD2 (uncontrollable worry; BEI = 0.074). Several nodes showed negative BEI values, including resilience (BEI = − 0.050), PHQ9 suicidal ideation (BEI = − 0.049), and friend support (BEI = − 0.048). Complete bridge centrality values are available in Supplementary Table S3.

Fig. 3.

Bridge expected influence centrality indices for all network nodes

Node predictability

Across all nodes, the mean predictability was 0.552 (SD = 0.153). The highest predictability was observed for public stigma (R² = 0.793), significant other support (R² = 0.767), and family support (R² = 0.764). The lowest predictability was found for resilience (R² = 0.264) and suicidal ideation (R² = 0.292). Predictability values for all nodes are visualized as ring-shaped pie charts in Fig. 1 and detailed in Supplementary Table S3.

Network stability and accuracy

The correlation stability (CS) coefficient was 0.75 for both strength and expected influence, exceeding the recommended threshold of 0.50 (Supplementary Figure S2). Bootstrapped 95% confidence intervals for edge weights were relatively narrow, indicating accurate edge estimation (Supplementary Figure S1). Split-half cross-validation demonstrated good network replicability, with a mean Spearman correlation of 0.717 (SD = 0.025, range: 0.662 to 0.770) between edge weight matrices across 100 random partitions (Supplementary Figure S3). Bootstrapped difference tests for expected influence revealed that GAD4 (difficulty relaxing) did not differ significantly from the next three highest-ranked nodes (GAD3, PHQ4, and PHQ2; all 95% confidence intervals included zero), but was significantly higher than ISS_Public (95% CI of difference: [0.018, 0.273]) and GAD6 (95% CI: [0.031, 0.318]) (Supplementary Figure S4). For edge weights, the difference between the strongest edge (public stigma and family stigma, weight = 0.538) and the second-strongest edge (family support and significant other support, weight = 0.464) was not statistically significant (95% CI of difference: [− 0.006, 0.148]).

Discussion

This study is among the first to employ network analysis to examine depression and anxiety symptoms alongside multiple psychosocial factors in a multi-center sample exclusively comprising Chinese women. Our sample exclusively comprised individuals with primary infertility. Our findings revealed that difficulty relaxing was the most central symptom, perceived stress was the primary bridge connecting psychosocial factors to symptoms, and guilt was a key bridge symptom. These results provide insights into the architecture of psychological distress in this population and generate hypotheses regarding candidate intervention targets that require prospective validation.

The identification of difficulty relaxing as the most central symptom is noteworthy. Women with primary infertility face prolonged uncertainty regarding treatment outcomes, and the repeated cycles of hope and disappointment may engender chronic psychophysiological tension. This finding is partially consistent with that of Cao et al. [18], who identified restlessness as the most central symptom in Chinese infertile patients. The discrepancy may reflect sample differences, as their study included both primary and secondary infertility. Wu et al. [23] demonstrated that central symptoms shift dynamically across IVF treatment stages, suggesting that network properties in this population are context-dependent. This symptom may reflect sustained psychophysiological arousal associated with the chronic stress of infertility, which in turn activates and maintains other anxiety and depressive symptoms through the network. This interpretation is further supported by qualitative evidence suggesting that women undergoing fertility treatment experience persistent tension and hypervigilance related to treatment outcomes [44], and by evidence that mindfulness-based interventions targeting relaxation have demonstrated efficacy in reducing psychological distress in this population [45].

Among psychosocial factors, public stigma exhibited the highest centrality, followed by self-devaluation. Notably, the association between public stigma and family stigma was the strongest edge in the entire network. This strong association highlights the interconnected nature of stigma dimensions in Chinese cultural contexts. In collectivist societies, childlessness is often perceived as a violation of social expectations regarding family continuity, and traditional emphasis on bearing children to continue family lineage creates an environment wherein women may experience stigmatization from both public and familial spheres simultaneously [44–47]. The high predictability of public stigma further suggests that this construct is heavily determined by other variables in the network, particularly family stigma and self-devaluation. Network analyses in other stigmatized populations have similarly demonstrated that stigma dimensions serve as bridge nodes linking stigma and depressive symptoms; for instance, Yuan et al. found that specific internalized stigma items functioned as key bridges connecting stigma and depression among people living with HIV in China [48].

Bridge expected influence analysis identified perceived stress as the predominant pathway connecting psychosocial factors to psychological symptoms, which aligns with transactional stress-coping models emphasizing the role of cognitive appraisal in mediating stressor-outcome relationships [45]. Notably, fertility-specific stress emerged as the second most important bridge factor, indicating that infertility-related concerns directly connect to psychological symptom clusters. This is consistent with the findings of Liu et al. [24], who identified relationship concern as a central node in the stress-stigma network of infertile women, and extends their work by demonstrating that fertility-specific stress also serves as a bridge to depression and anxiety symptoms, supporting the clinical utility of targeted interventions addressing fertility-related concerns. Among symptom nodes, guilt exhibited the highest bridge expected influence, suggesting that guilt may facilitate activation spread from psychosocial stressors to depressive symptoms. The clinical relevance of guilt is further supported by evidence that this symptom shows unique connections to suicidal ideation among Chinese infertile women [18]. This finding is also consistent with network analyses in other Chinese populations; for example, Zhao et al. found that guilt emerged as one of the top bridge symptoms connecting depressive symptoms and quality of life among Wuhan residents [49].

Resilience exhibited the strongest negative edge with perceived stress in the network, along with a negative bridge expected influence. This pattern suggests a potential buffering role against cross-domain activation spread, which is consistent with previous findings that resilience mediates the relationship between infertility stress and psychological well-being [16, 17]. The relatively low predictability of resilience indicates that this construct is largely independent of other network variables and may represent a stable individual characteristic. This suggests that interventions may need to target resilience directly, rather than relying on improvements through connected nodes in the network. Resilience-building interventions may therefore provide benefits by both reducing symptom severity and attenuating the impact of stress on psychological outcomes.

While these cross-sectional findings require prospective validation, they generate several testable hypotheses for future intervention development. The centrality of difficulty relaxing is consistent with the potential value of incorporating relaxation training into psychosocial care for women with primary infertility [50]. The identification of perceived stress as a key bridge factor suggests that interventions targeting cognitive appraisal of stressors may warrant investigation. Cognitive restructuring techniques targeting maladaptive stress cognitions could potentially attenuate the associations through which external stressors relate to symptom networks [51]. Given the prominence of guilt as a bridge symptom, self-compassion training may be particularly beneficial for this population, as it directly addresses self-criticism and shame that often accompany reproductive difficulties [52].

Finally, it is worth situating the present findings within the broader context of computational approaches to mental health. Network analysis and emerging computational paradigms address fundamentally different research questions. Machine learning and deep learning approaches, including adversarial domain adaptation techniques for cross-corpus generalization [27] and multimodal fusion architectures that integrate physiological signals such as electroencephalography and electrodermal activity [25, 26], primarily address the task of automated detection and classification of mental health conditions. These approaches take behavioral, linguistic, or physiological data as input and produce diagnostic classifications as output. By contrast, network analysis takes symptom and psychosocial factor measurements as input and produces a map of conditional dependencies as output, identifying structurally central and bridging nodes that represent candidate intervention targets. These two paradigms are not competing but synergistic, each addressing a distinct clinical need.

Building on these findings, several points of integration can be identified. First, the central and bridge nodes identified through network analysis could serve as theoretically informed feature selection criteria for machine learning models. For instance, the finding that perceived stress functions as the primary bridge connecting psychosocial factors to symptoms suggests that stress appraisal measures may carry greater predictive value than other psychosocial variables when used as features in automated screening models. Second, the identification of difficulty relaxing as the most central symptom suggests that physiological correlates of sustained tension, such as heart rate variability or electromyographic indices, may represent informative input features for multimodal detection systems targeting this population. Third, longitudinal machine learning approaches could validate cross-sectional network findings by testing whether baseline centrality values predict subsequent symptom trajectories. Fourth, the low clinical symptom prevalence in the present sample highlights the potential value of computational screening tools for identifying at-risk individuals at the subclinical stage. Future research integrating network-analytic and machine learning approaches may advance both the understanding and prediction of psychological distress in this population.

Several limitations should be acknowledged. First, the cross-sectional design precludes causal inference regarding the directionality of associations. Central and bridge nodes identified in cross-sectional networks do not necessarily represent optimal intervention targets, as their centrality may reflect consequences rather than causes of psychological distress. Longitudinal and experimental designs are necessary to determine whether targeting the identified nodes produces cascading improvements in network-wide outcomes. Second, although our multi-center design enhanced generalizability within China, all participants were recruited from tertiary hospitals in southern China, which may limit generalizability to other regions or cultural contexts. Third, our sample exhibited relatively low mean levels of depression and anxiety, with only 9.6% and 5.3% meeting clinical thresholds. The restricted variance in symptom scores may attenuate partial correlations, and the identified network structure may not generalize to clinically severe populations. Future studies should examine whether network topology differs across symptom severity levels. Fourth, participants were at varying stages of fertility treatment, and network properties may shift across treatment phases [23]. Additionally, the mixing of item-level symptom nodes with scale-level psychosocial factor nodes may introduce asymmetry in edge weight and bridge centrality estimation; future studies could conduct sensitivity analyses comparing different network specifications.

Conclusion

In conclusion, this network analysis identified central and bridge symptoms that represent candidate targets for future intervention research. The findings highlight the importance of addressing difficulty relaxing as the most central symptom, targeting perceived stress and guilt as key bridging pathways, and considering the interconnected nature of stigma dimensions in Chinese cultural contexts. Enhancing resilience may provide additional protective benefits. Future longitudinal research is needed to clarify the temporal dynamics among network nodes and to evaluate whether interventions focusing on the identified central and bridge symptoms lead to cascading improvements in psychological outcomes.

Supplementary Information

Below is the link to the electronic supplementary material.

Abbreviations

ART

Assisted Reproductive Technology

BEI

Bridge Expected Influence

CD-RISC-10

Connor-Davidson Resilience Scale-10

COMPI-FPSS-SF

Copenhagen Multi-center Psychosocial Infertility-Fertility Problem Stress Scale-Short Form

CS

Correlation Stability

EBIC

Extended Bayesian Information Criterion

EI

Expected Influence

GAD-7

Generalized Anxiety Disorder-7

GGM

Gaussian Graphical Models

ISS

Infertility Stigma Scale

IUI

Intrauterine Insemination

IVF

In Vitro Fertilization

LASSO

Least Absolute Shrinkage and Selection Operator

MSPSS

Multidimensional Scale of Perceived Social Support

PHQ-9

Patient Health Questionnaire-9

PSS-10

Perceived Stress Scale-10

Author contributions

YF, JL, and SZ contributed to the study conception and design. YF performed the statistical analysis. YF, FZ, XL, ZX, and JH were responsible for data collection and investigation. YF and XL created the figures and tables. YF and ZF drafted the original manuscript. SZ, JL, and FZ provided resources and supervision. ZF, XL, and SZ validated the data. All authors reviewed the manuscript and approved the final version.

Funding

This research was supported by Shantou Healthcare Science and Technology Program (project ID 240422116497013).

Data availability

The dataset supporting the conclusions of this article is available in the Zenodo repository, 10.5281/zenodo.18453822.

Declarations

Ethics approval and consent to participate

This study was approved by the Ethics Committees of The First Affiliated Hospital of Shantou University Medical College (B-2024-229), Shantou Central Hospital (2025-022), Longgang District Maternity and Child Healthcare Hospital of Shenzhen City (LGFYKYXMLLM-2025-9), and The Third Affiliated Hospital of Guangzhou Medical University (2025-098). This study was conducted in accordance with the Declaration of Helsinki. Written informed consent was obtained from all participants.

Consent for publication

Not applicable.

Competing interests

The authors declare no competing interests.