This retrospective observational study was designed and reported in accordance with the Strengthening the Reporting of Observational Studies in Epidemiology (STROBE) guidelines (36). The study included women diagnosed with PCOS who had established specialized records at the Gynecology Outpatient Department of the Third Affiliated Hospital of Guangzhou Medical University between January 2022 and December 2024. Ethical approval was obtained from the Ethics Committee of the Third Affiliated Hospital of Guangzhou Medical University, and all study procedures adhered to the principles outlined in the Declaration of Helsinki. Because this study utilized retrospective data extracted from medical records and an electronic database, anonymizing all data, the Ethics Committee approved a waiver of informed consent.

We screened electronic databases of all patients who established specialized records at the Gynecology Outpatient Department between January 2022 and December 2024 to identify cases diagnosed with PCOS. A trained investigator independently reviewed the identified cases to confirm whether they met the diagnostic criteria for PCOS according to the 2003 Rotterdam consensus (37). Inclusion criteria were (1) aged 18–40 years; (2) meeting at least two of the three Rotterdam criteria: (a) oligo-ovulation or anovulation, (b) clinical and/or biochemical hyperandrogenism, or (c) polycystic ovarian morphology detected by ultrasonography. Exclusion criteria were (1) pregnant or lactating women; (2) perimenopausal or early menopausal women; (3) history of confirmed psychiatric or neurological disorders; (4) major life events within the past 6 months; (5) incomplete completion of psychological or sleep questionnaires; and (6) missing data exceeding 20% or incomplete clinical records. Following the application of these strict criteria, 1,068 women with PCOS were included in the final analysis.

We used SPSS Statistics (Version 25.0) to perform descriptive statistics, analyzing demographic information and scores across various scales. Subsequently, we employed R (Version 4.4.1) to construct a comorbidity network for anxiety, depression, and sleep problems. Before building the network, we assessed item redundancy using the correlation matrix method (i.e., correlation coefficients r > 0.7 were considered potentially redundant) (46).

This study included 1,068 patients with PCOS, aged 18–40 years (mean 27.77 ± 4.44). Within this population, 640 (59.9%) were married, 591 (55.3%) had fertility desires, and 334 (31.3%) were overweight/obese (BMI≥24). Notably, 503 participants (47.1%) displayed anxiety symptoms (GAD-7 score ≥ 5), 440 (41.2%) exhibited depressive symptoms (PHQ-9 score ≥ 5), and 288 (27.0%) reported sleep problems (PSQI global score > 7). Additional information is shown in Table 1. The mean, standard deviation (SD), expected influence (EI), bridge expected influence (BEI), and predictability of all items across the scales are presented in Supplementary Table S1. Furthermore, no redundant item was detected between any items within this network (i.e., correlation coefficient showing more than 0.7).

The anxiety, depression, and sleep problems network model of PCOS patients is depicted in Figure 1. Within this network structure, among 253 possible edges, 139 non-zero edges were identified (network density=0.55), comprising 135 positive edges and 4 negative edges. The strongest edge, which links PHQ4 “low energy” in depression and PSQI7 “Daytime dysfunction” in sleep problems, is found across communities (weight=0.403). The remaining firm edges are primarily concentrated within their respective communities. Within the sleep symptom community, the strongest edges were observed between PSQI1 “Subjective sleep quality” and PSQI2 “Sleep latency” (weight=0.396), and between PSQI3 “Sleep duration” and PSQI4 “Sleep efficiency” (weight=0.329). For anxiety symptoms, significant connections included GAD1 “Nervousness”-GAD2 “Uncontrollable worry” (weight=0.244), GAD2 “Uncontrollable worry”-GAD4 “Trouble relaxing” (weight=0.216), and GAD2 “Uncontrollable worry”-GAD3 “Excessive worry” (weight=0.202). Among depressive symptoms, the strongest connections were PHQ6 “Worthlessness”-PHQ9 “Suicidal ideation” (weight=0.238), PHQ1 “Anhedonia”-PHQ2 “Sad mood” (weight=0.218), and PHQ7 “Poor concentration”-PHQ8 “Abnormal behavior and speech” (weight=0.215). For detailed weights of all edges, please refer to Supplementary Table S2 in the Supplementary Materials.

In addition, the predictability of each node is represented by the outer rings surrounding the nodes in Figure 1. The average predictability across all network nodes is 0.42, indicating that 42% of node variance can be explained by their neighboring nodes. Among them, “Excessive worry” (GAD3) and “Trouble relaxing” (GAD4) show high predictability, whereas “Use of sleep medication” (PSQI6) had the lowest predictability.

The EI results for each node are shown in Figure 2. “Trouble relaxing” (GAD4) and “Sad mood” (PHQ2) exhibit the highest expected impact values, followed by “Low energy” (PHQ4), indicating that these three symptoms are the most influential and occupy a central position in the current network. The BEI results for each node are shown in Figure 3. “Daytime dysfunction” (PSQI7), “Low energy” (PHQ4), and “Sad mood” (PHQ2) had the highest bridge expected influence. These symptoms serve as crucial bridge nodes linking anxiety, depression, and sleep problems.

The findings of robustness analyses were as follows. First, nonparametric bootstrapping yielded relatively narrow 95% confidence intervals for edge weights, suggesting precise estimates. Second, the correlation stability coefficient (CS-C) for both expected influence and bridge expected influence reached 0.75, which exceeds the recommended threshold of 0.50. The centrality metrics exhibit strong stability. Finally, bootstrapped difference tests confirmed that the strongest edges (e.g., PHQ4–PSQI7, PSQI1–PSQI2) and the influence values of core (GAD4, PHQ2) and bridge (PSQI7, PHQ4) nodes were significantly greater than those of most others (Supplementary Figure S1).

Network comparison results revealed invariance in network structure or overall strength between subgroups defined by clinical features (Supplementary Figures S2-S6). Regarding global strength, no significant differences were observed between the non-overweight and overweight/obese groups (9.42 vs. 9.07, P=0.383), infertility and without infertility groups (9.25 vs. 9.28, P=0.950), hirsutism and without hirsutism groups (9.61 vs. 9.30, P=0.419), acne and without acne groups (9.15 vs. 9.10, P=0.882), or acanthosis nigricans and without acanthosis nigricans groups (9.04 vs. 9.35, P=0.531). Similarly, in network structure comparisons, no significant differences were found between weight status (M=0.17, P=0.546), infertility(M=0.15, P=0.873), hirsutism(M=0.18, P=0.480), acne(M=0.15, P=0.778), or acanthosis nigricans(M=0.20, P=0.528) groups. After Bonferroni-Holm correction, all edge weight differences within subgroups remained non-significant (P>0.05).

The symptom network in this study reveals the tightly intertwined relationship among anxiety, depression, and sleep problems. This pattern aligns with findings from other populations, further supporting the existence of a stable associative pattern among these three psychiatric symptoms (57, 58).

Distinct from most network studies in which the strongest associations typically occur within the same symptom community, the present study identified the strongest edge across communities, linking “daytime dysfunction” in sleep problems with “low energy” in depressive symptoms (PSQI7-PHQ4). Existing research indicates that sleep disorders frequently are often accompanied by daytime functional impairment and low-energy states (20, 59, 60). Physiological evidence suggests that sleep deprivation may disrupt the hypothalamic-pituitary-adrenal (HPA) axis and energy metabolism (61, 62), partly contributing to low-energy experiences. In women with PCOS, disease-related pathophysiological features and persistent psychosocial stress may further accentuate this association, making daytime dysfunction a salient correlate of low energy in this population. This cross-community association suggests that daytime functioning may play an important role in the co-occurrence of sleep problems and depressive symptoms among women with PCOS. The next strongest edges were observed within the sleep problems community, indicating tight associations different specific sleep dimensions. These findings suggest that sleep disorders in women with polycystic ovary syndrome rarely occur in isolation (63, 64). Clinical management should focus on the comorbidity and mutual reinforcement of multiple sleep issues, rather than targeting interventions at individual symptoms alone.

Notably, the strongest association within the depressive symptom network was identified between “Suicidal ideation” (PHQ9) and “Worthlessness” (PHQ7). This finding has been relatively underreported in previous network analyses, which commonly focused on the relationship between “suicidal ideation” and “Abnormal behavior and speech “ (65, 66). Although the centrality of suicidal ideation was relatively low in our network, it remains a clinically critical high-risk symptom. Research indicates that “Worthlessness” on the PHQ-9 scale is a significant predictor of suicidal ideation among primary care patients (67). Individuals with suicidal ideation commonly exhibit negative self-beliefs (e.g., low self-worth) (68), and that persistent self-deprecation and feelings of failure may intensify hopelessness, thereby fostering suicidal ideation. This psychological pathway might be further amplified in patients with complex and severe PCOS, who often experience body image dissatisfaction, repeated treatment frustration, and reproductive stress (22, 23, 69). Meanwhile, the physiological changes associated with PCOS also may exacerbate emotional dysregulation and feelings of worthlessness (70, 71). Although the specific mechanisms require further validation, this study suggests that clinicians should pay close attention to early signs of worthlessness (e.g., self-blame, feeling useless, or disappointing others) to prevent the escalation toward suicidal ideation.

Nodes with higher expected impact usually occupy core positions within symptom networks, offering potential targets for psychological interventions. “Trouble relaxing” (GAD4) had the highest centrality value in this study, suggesting that it plays a crucial role in triggering and maintaining symptom networks of anxiety, depression, and sleep problems. As a core symptom of anxiety disorders, trouble relaxing manifests as persistent somatic tension and difficulty achieving relaxation (72). While “Trouble relaxing” has been widely recognized as clinically relevant in psychopathology (73, 74), its prominent centrality in the PCOS patients may be associated with the prevalence of chronic psychological stress and physiological hypervigilance; however, this hypothesis requires further validation. Although network analysis cannot establish direct causality, this pattern can provide useful insights for clinical psychological management. Specifically, interventions targeting “Trouble relaxing”, identified as a core symptom in the network, may represent a potential entry point, with their impact on the overall psychological symptom network in women with PCOS requiring further validation in future studies.

Furthermore, “sad mood” (PHQ2) was the second core symptom and played an important bridging role within the network. This pattern finds support in a network study of general populations from El Salvador and Peru (75), suggesting that the central role of sad mood may not be unique to PCOS. The DSM-5 lists depressed mood as a core criterion for diagnosing depressive disorders (76), primarily manifested as negative emotional experiences such as sadness, fear, and feelings of failure. Consistent with this, a cross-sectional study focusing on PCOS patients found that sad mood was the most common and prominent symptom of depression (77). The chronic disease burden, psychological stress, and inadequate medical support associated with PCOS may increase the risk of persistent sad mood (78, 79), which may partly explain the higher centrality of sad mood in the present network. Moreover, as a bridging symptom, “sad mood” showed significant associations with “nervousness” (GAD1) and “feeling afraid” (GAD7) within the anxiety symptom cluster. The additional association with “feeling afraid,” which had been less frequently reported, suggests that symptom interrelations may present certain population-specific features in women with PCOS and warrants further investigation. Given evidence that sad mood represents a shared vulnerability factor for comorbid anxiety and depressive disorders (80), it may function as a key emotional linkage within the psychological symptom network of PCOS. In clinical practice, this suggests that addressing and intervening in sad mood may alleviate patients’ broader emotional distress during early psychological management.

The bridge expected influence analysis helps to pinpoint key symptoms that facilitate the emergence and persistence of comorbid psychiatric conditions. In the present study, in addition to “sad mood,” “daytime dysfunction” (PSQI7) and “low energy” (PHQ4) emerged as the most significant bridge symptoms within the network. Sleep problems can generally be categorized into two dimensions: nighttime disturbances and daytime consequences, both of which lead to subjective dissatisfaction with sleep quantity or quality. Daytime dysfunction as a key manifestation of sleep problems (81), has previously been identified as a bridging symptom between sleep problems and emotional symptoms in network studies (65). Low energy typically manifests as persistent fatigue or physical exhaustion, serving as one of the core features of depression (82). Research indicates that both mental disorders and somatic complaints may jointly contribute to fatigue (83), suggesting that chronic psychological distress may translate into somatic symptoms such as reduced energy. Previous network research on individuals with depressive symptoms identified “low energy” as a key bridge symptom between insomnia and depression (84), further supporting the conclusions of this study. In summary, although depression and sleep symptoms exhibit distinct patterns within the network, the reciprocal relationship between “daytime dysfunction” and “low energy” may represent a cross-diagnostic link between them (85). Therefore, in clinical assessment, attention to these two bridge symptoms (daytime dysfunction and low energy) may aid in the earlier identification of PCOS women at elevated risk for psychological comorbidity. However, whether interventions targeting such bridge symptoms can effectively attenuate symptoms or prevent comorbidity requires future interventional studies.

Network comparisons based on clinical features (infertility, hirsutism, acne, acanthosis nigricans, and weight status) showed no significant differences between subgroups in overall network strength, structural integrity, or edge weight intensity. These findings suggest that although prior research has demonstrated associations between these clinical characteristics and levels of psychological distress (86), their influence may be insufficient to distinguish the associative patterns among psychological symptoms. A possible explanation is that symptom network analysis focuses on the interactive structure among symptoms rather than the severity of individual symptoms, which may, to some extent, reduce the capacity of traditional clinical features to differentiate between network structures. Furthermore, the incomplete correspondence between biochemical abnormalities and clinical manifestations (e.g., androgen levels versus severity of hirsutism and acne) (87), the time-dependent nature of infertility-related psychological burden (88), and individual differences in psychosocial stress perception and coping may further weaken subgroup-level structural distinctions.

From a clinical perspective, the relatively consistent network structure across clinical subgroups suggests that the interrelations among anxiety, depression, and sleep problems may represent shared psychological processes in women with PCOS. This observation supports the potential applicability of symptom-focused psychological intervention frameworks across different clinical presentations, though it does not imply complete uniformity in symptom severity or intervention needs across different clinical subgroups. Given the potential influence of factors such as sample distribution and individual differences, further validation through longitudinal designs and biochemical indicators is warranted in future research.

Of course, there are several limitations in our study. First, the cross-sectional design precludes causal inference, and longitudinal studies are needed to examine the dynamic interactions among anxiety, depression, and sleep problems in PCOS patients. Second, our study was conducted among PCOS patients at a tertiary-level hospital in a first-tier city, and the applicability of these findings across different regions, healthcare tiers, and sociocultural contexts remains to be validated. Third, psychological symptoms were assessed using self-report measures, which may introduce reporting bias. Lastly, due to the retrospective nature of the study, detailed PCOS phenotypic classification (e.g., metabolic vs. reproductive subtypes), clinical and physiological severity (e.g., infertility duration, hormonal and metabolic measures), and psychosocial factors (e.g., reproductive stress, coping capacity) were not consistently available, which may have masked psychological heterogeneity across subgroups and constrained interpretation of symptom network differences. Future studies should adopt prospective designs incorporating detailed PCOS phenotyping as well as comprehensive biological and psychosocial measures to better characterize symptom dynamics and heterogeneity in this population. Further research is also warranted to examine the potential relevance of symptom-focused strategies suggested by network analysis, as well as to explore the relationships between psychological symptoms and PCOS clinical and physiological features, which could provide insights for more targeted and effective management strategies.