Abstract
Background
Maintenance hemodialysis (MHD) patients commonly experience stigma during long-term treatment, yet the characteristics of their subgroups and the associated patterns with symptom burden remain unclear. This study aimed to identify latent subgroups of stigma among MHD patients and depict the symptom network structure of each subgroup.
Methods
This cross-sectional study enrolled 558 MHD patients from two hospitals between March and December 2024. Stigma was assessed using the Social Impact Scale (SIS), symptom burden was assessed using the Modified Dialysis Symptom Index (DSI), subgroups of stigma were identified using latent profile analysis, and complex relationships between symptoms were explored using network analysis. Network comparison tests were performed to evaluate between-group differences in network connectivity.
Results
The study found that there were three subgroups of MHD patients with stigma: low stigma group (38.5%), moderate stigma group (49.6%), and high stigma group (11.8%). The network analysis showed that muscle pain, sadness, and chest pain were the core symptom nodes in the overall sample. Stable symptom networks were observed in the low and moderate stigma groups, with core symptoms differing by subgroup (physical symptoms in the low stigma group vs. emotional symptom in the moderate stigma group), though no significant difference in global network strength was found between these two groups.
Conclusions
Stigma among MHD patients presents distinct heterogeneous subgroup characteristics, and there are core symptoms with high connection strength in the symptom network. The difficulty in recruiting patients with high stigma reflects the concealment of this population. Future studies should adopt longitudinal designs and expand the sample size of the high stigma subgroup to verify the value of core symptoms in symptom management.
Clinical trial number
Not applicable.
Supplementary Information
The online version contains supplementary material available at 10.1186/s12882-026-04798-0.
Keywords: Maintenance hemodialysis, Stigma, Latent profile analysis, Network analysis, Symptom burden
Introduction
Chronic kidney disease (CKD) is a significant global public health challenge. For patients progressing to kidney failure, maintenance hemodialysis (MHD) serves as a prevalent life-sustaining treatment [1]. The number of people receiving maintenance hemodialysis treatment worldwide exceeded 2.5 million and is expected to more than double to 5.4 million by 2030 [2]. While MHD extends survival, patients endure a considerable burden of co-occurring physical and psychological symptoms—such as pain, fatigue, itching, anxiety, and depression—which are associated with reduced quality of life [3].
Beyond the symptom burden, individuals on MHD frequently experience stigma, a distressing psychosocial experience arising from perceived social devaluation and discrimination due to their illness and treatment dependency [4]. Stigma is associated with adverse outcomes, including social isolation, reduced treatment adherence, and exacerbated psychological distress, forming a vicious cycle that further compromises patient well-being and prognosis [5, 6].
While physical discomfort and psychological distress are presumed to heighten feelings of difference and shame [7], prior research has primarily examined these factors in isolation or using variable-centered approaches that assume population homogeneity. This overlooks the potential for distinct patient subgroups who may experience unique configurations of stigma and symptoms. Furthermore, symptoms likely do not operate independently but interact within dynamic networks, where certain “core” symptoms may be strongly associated with others and potentially linked to higher levels of stigma [8]. Understanding this intricate, potentially subgroup-specific symptom-stigma network is crucial for developing targeted interventions.
To systematically investigate the aforementioned issues, this study constructed an integrated framework (Fig. 1) based on social cognitive theory [9] and the network theory of psychopathology [10]. Social cognitive theory helps understand stigma as a multidimensional psychological experience formed by an individual in interactions with the social environment, where its heterogeneity (i.e., subgroups) may stem from different cognitive evaluations and coping processes. Network theory conceptualizes co-occurring symptoms as a dynamic interactive system, where certain “core” symptoms may be strongly associated with others, providing a new perspective for identifying key targets for intervention. This framework was designed to test whether there is an association between the heterogeneous subgroups of stigma and specific structural features of the symptom network (such as core hubs, modules). To test this framework: First, latent profile analysis was used to identify stigma subgroups; Second, network analysis was employed to depict the symptom interaction patterns within the overall network and different subgroups. This strategy allows us to jointly reveal the complex landscape of stigma and symptom experience from the two levels of “group classification” and “system structure.”
Fig. 1.
Integrated theoretical framework for investigating stigma-symptom. Networks in patients on MHD. Note: Solid arrows indicate known association pathways based on existing theories or literature; dashed arrows represent the subgroup-specific network interaction mechanisms that are the focus of this study
Methods
Research design
A cross-sectional study was conducted, involving patients undergoing MHD at the Second Affiliated Hospital of Shandong First Medical University and Tai’an 88th Hospital (both in Tai’an City, Shandong Province) between March 2024 and December 2024.
Participants
The inclusion criteria were as follows: (1) age ≥ 18 years; (2) regular hemodialysis for at least 3 months; (3) stable dialysis schedule of 2–3 sessions per week; (4) voluntary participation with informed consent. Exclusion criteria were: (1) severe illness preventing cooperation with the survey (2) a history of mental or language disorders.
Instruments
General information questionnaire
The questionnaire was developed by the researcher, including age, gender, residence, education level, occupation, marital status, cohabitant, family relationship, family members who studied medicine, monthly income, medical expenses, medical insurance type, primary disease, complication, dialysis time and frequency, and 24-hour urine volume. Family relationship quality was assessed via a single self-report item (“How would you describe your current family relationship?”) with responses “good” or “poor.”
Social impact scale (SIS)
The Social Impact Scale (SIS), developed by Fife et al. [11], was used to assess stigma in cancer and AIDS patients. In 2007, Pan et al. [12] translated it into Chinese. SIS consists of 24 items divided into four categories: social rejection, financial insecurity, internalized shame, and social isolation. The total score ranged from 24 to 96. A higher SIS score indicated a greater level of perceived stigma. All 24 items were rated on a 4-point Likert-type scale, with 4 representing ‘strongly agree’ and 1 representing ‘strongly disagree’. In this study, the Cronbach’s α coefficient was 0.962.
Modified dialysis symptom index (DSI)
We used the Chinese version revised by Hao [13] from the original DSI [14]. The original DSI assessed symptom bother only; the revised version added two additional dimensions (frequency and severity) for more comprehensive symptom burden assessment. The instrument contained 30 items (25 physiological, 5 psychological). Participants first indicated the presence (yes/no) of each symptom during the past seven days. If present, frequency and severity were rated on a 4-point Likert scale (1 = occasionally/mild to 4 = always/extremely serious), and distress was rated on a 5-point Likert scale (0 = not at all bothersome to 4 = bothers very much). The total score ranged from 0 to 390, with higher scores indicating greater symptom burden. The Chinese version demonstrated satisfactory validity and reliability in patients on hemodialysis [15, 16]. In this study, Cronbach’s α coefficient was 0.956. The full item pool and scoring procedures were provided in Appendix 1, 2, 3, ensuring reproducibility.
Data collection procedures
This study collected data using paper questionnaires, and all the questionnaires were obtained from publicly available online sources. When MHD patients came to the hemodialysis center for treatment, the researchers used a uniform statement to explain the purpose of the study to patients and conducted the survey after their informed consent. The complete questionnaire (including demographic items, SIS, and DSI) contained a total of 67 questions and took approximately 15–20 min to complete. A total of 575 questionnaires were collected in this study, and 17 (2.96%) were excluded because more than 20% of items on either SIS or DSI were incomplete. This pre-specified exclusion criterion was applied to ensure data quality for primary analyses. Missing-pattern analysis showed no significant differences in age, gender or dialysis vintage between excluded and analyzed participants (p > 0.10). The final analytical sample comprised 558 complete questionnaires (effective response rate 97.04%).
Statistical analyses
Statistical analysis was performed using SPSS 26.0, R software (version 4.2.2), and Mplus version 8.0. First, descriptive statistics were processed using SPSS 26.0 software to evaluate the scores of demographic variables and stigma. Categorical variables were analyzed using frequency and percentage, while continuous variables were analyzed using mean and standard deviation. In Latent Profile Analysis (LPA), models were compared using the Akaike Information Criterion (AIC), Bayesian Information Criterion (BIC), Adjusted Bayesian Information Criterion (aBIC), Lo-Mendell-Rubin test (LMR), Bootstrap likelihood ratio test (BLRT), and entropy. The optimal number of latent profiles was selected by the relatively low values of AIC, BIC, and aBIC, high entropy values, and significant p-values for LMR and BLRT [17]. In addition, the clinical interpretability of each latent profile was also taken into account [18]. After identifying the optimal latent profile solution, demographic and clinical characteristics were compared across profiles. Continuous variables were analyzed using one-way ANOVA, while categorical variables were compared using the chi-square test of independence or Fisher’s exact test, as dictated by expected cell frequencies. Given 13 variables were tested, the Benjamini-Hochberg (BH) correction was applied to the primary intergroup comparison p-values to control the false discovery rate (FDR) at 5%. Subsequently, the significant factors (including symptom burden) were incorporated into the multinomial logistic regression model to evaluate the association between demographic and clinical characteristics and stigma subgroup membership. For symptom burden (assessed by DSI total score), the standardized score (z-score) was used in the model to avoid scale bias. The z-score was calculated as (individual raw score - sample mean) / sample standard deviation. For all statistical tests, p < 0.05 was considered statistically significant.
Additionally, we estimated the network using the EBICglasso function in the qgraph package of R software (version 4.2.2). The qgraph package generated a Gaussian graphical model (GGM) by using the least absolute shrinkage and selection operator (LASSO) function. LASSO generated sparser networks by removing relatively weakly connected links [19]. A node in a network visualization represented a symptom or a measurement item. In the networks, blue edges represented positive connections and red edges represented negative connections, and the thickness of the edges indicated the strength of the connections. The more strongly connected nodes clustered together, and the more influential nodes were located more centrally in the network. Edge weights were interpreted as the strength of the unique association between two symptoms.
Subsequently, three commonly used indices of node centrality, including strength, closeness, and betweenness, were calculated (standardized values) to quantify the importance of each item to the network structure. Strength was defined as the sum of all the edge weights linked to a node. Symptoms with higher strength were more likely to occur simultaneously with other symptoms and were more centrally connected within the network. Since previous studies had provided more reliable estimates of strength centrality, we focused on this indicator to identify the most relevant symptoms with stigma. Finally, the bootnet package was employed to assess network stability. The accuracy of network connections was evaluated by estimating the 95% confidence intervals (CIs) of edge weights. Smaller 95% CIs indicated higher accuracy of edge weights [20]. Network stability was evaluated using the correlation stability coefficient (CS-C). The CS-C was recommended to be above 0.5 and not below 0.25. To statistically compare symptom networks across stigma subgroups, a permutation-based test of global strength was performed using the NetworkComparisonTest package (version 2.2.1) in R [21]. This procedure (10,000 permutations) evaluated pairwise differences in overall network connectivity between stigma groups, with p-values adjusted using Bonferroni correction for multiple comparisons.
Results
Participant characteristics
A total of 558 patients were included in the study. There were more male patients (59.0%). Most of the patients were married (87.5%), had good family relations (72.8%), had no medical members in their families (90.5%), spent between 1,000 and 2,000 yuan on dialysis per month (56.6%), and had chronic diseases (66.7%). Other demographic and clinical characteristics were shown in Table 3.
Table 3.
Demographic and clinical characteristics of MHD patients by stigma subgroup
| Variables | Categories | Mean ± SD/N(%) | Low stigma group n = 215 |
Moderate stigma group n = 277 |
High stigma group n = 66 |
Effect value | p | BH-corrected p-value (FDR = 5%) |
|---|---|---|---|---|---|---|---|---|
| age | age | 55.45 ± 12.65 | 54.27 ± 12.445 | 56.31 ± 13.296 | 55.71 ± 10.163 | 1.606 | 0.202 | 0.358 |
| gender | male | 329(59.0) | 126(58.6) | 170(61.4) | 33(50.0) | 2.867 | 0.238 | 0.386 |
| female | 229(41.0) | 89(41.4) | 107(38.6) | 33(50.0) | ||||
| residence | village | 142(25.4) | 72(33.5) | 64(23.1) | 6(9.1) | 48.552 | < 0.001 | < 0.001 |
| township | 182(32.6) | 79(36.7) | 92(33.2) | 11(16.7) | ||||
| county | 38(6.8) | 13(6.0) | 14(5.1) | 11(16.7) | ||||
| city | 196(35.1) | 51(23.7) | 107(38.6) | 38(57.6) | ||||
| education level | primary school or below | 96(17.2) | 52(24.2) | 37(13.4) | 7(10.6) | 23.773 | < 0.001 | < 0.001 |
| junior high school | 222(39.8) | 92(42.8) | 101(36.5) | 29(43.9) | ||||
| high school and technical secondary school | 149(26.7) | 47(21.9) | 88(31.8) | 14(21.2) | ||||
| junior college or above | 91(16.4) | 24(11.2) | 51(18.4) | 16(24.2) | ||||
| marital status | unmarried | 40(7.2) | 21(9.8) | 19(6.9) | 0 | 12.584 | 0.038 | 0.109 |
| married | 488(87.5) | 185(86.0) | 243(87.7) | 60(90.9) | ||||
| divorced | 16(2.9) | 6(2.8) | 8(2.9) | 2(3.0) | ||||
| widowed | 14(2.5) | 3(1.4) | 7(2.5) | 4(6.1) | ||||
| cohabitant | solitude | 54(9.7) | 24(11.2) | 26(9.4) | 4(6.1) | 17.543 | 0.007 | 0.037 |
| with spouse | 287(51.4) | 112(52.1) | 129(46.6) | 46(69.7) | ||||
| with children/parents | 53(9.5) | 26(12.0) | 24(8.6) | 3(4.5) | ||||
| with spouse and children | 164(29.4) | 53(24.7) | 98(35.4) | 13(19.7) | ||||
| family relationship | good | 406(72.8) | 148(68.8) | 199(71.8) | 59(89.4) | 11.001 | 0.004 | 0.026 |
| poor | 152(27.2) | 67(31.2) | 78(28.2) | 7(10.6) | ||||
| family members studies medicine | yes | 53(9.5) | 14(6.5) | 32(11.6) | 7(10.6) | 3.685 | 0.158 | 0.305 |
| no | 505(90.5) | 201(93.5) | 245(88.4) | 59(89.4) | ||||
| monthly income (yuan) | < 1000 | 259(46.4) | 117(54.4) | 126(45.5) | 16(24.2) | 28.807 | < 0.001 | < 0.001 |
| 1000–3000 | 153(27.4) | 58(27.0) | 69(24.9) | 26(39.4) | ||||
| 3000–5000 | 88(15.8) | 31(14.4) | 45(16.2) | 12(18.2) | ||||
| > 5000 | 58(10.4) | 9(4.2) | 37(13.4) | 12(18.2) | ||||
| medical expenses (yuan) | 0-1000 | 147(26.3) | 51(23.7) | 77(27.8) | 19(28.8) | 3.526 | 0.474 | 0.613 |
| 1000–2000 | 316(56.6) | 129(60.0) | 154(55.6) | 32(48.5) | ||||
| 2000–3500 | 95(17.0) | 35(16.3) | 46(16.6) | 15(22.7) | ||||
| chronic disease | no | 186(33.3) | 69(32.1) | 102 (36.8) | 15(22.7) | 5.008 | 0.082 | 0.189 |
| yes | 372(66.7) | 146(67.9) | 175(63.2) | 51(77.3) | ||||
| the year when dialysis began | 2000–2015 | 126 | 45(20.9) | 62(22.4) | 19(28.8) | 4.026 | 0.402 | 0.569 |
| 2016–2020 | 174(31.2) | 74(34.4) | 79(31.8) | 21(31.8) | ||||
| 2021–2024 | 258(46.2) | 96(44.7) | 136(49.1) | 26(39.4) | ||||
| 24-hour urine volume | < 100 ml | 355(63.6) | 135(62.8) | 177(63.9) | 43(65.2) | 0.740 | 0.946 | 0.946 |
| 100-400 ml | 101(18.1) | 42(19.5) | 47(17.0) | 12(18.2) | ||||
| > 400 ml | 102(18.3) | 38(17.7) | 53(19.1) | 11(16.7) |
Note: Multiple comparisons were adjusted using the Benjamini-Hochberg (BH) method (number of tested variables m = 13) to control the false discovery rate (FDR) at 5%
Latent profile analysis
To explore the latent profiles of stigma in MHD patients, a latent profile model was established using each dimension of the SIS as an explicit variable. Starting from the benchmark model of a profile, the number of latent profiles was increased successively for model fitting estimation. Then, we obtained latent profile models with 1 to 5 profiles, shown in Table 1. In this study, the model with Profile 3 (C1, C2, and C3) was selected as the best latent profile model (entropy = 0.910). The average posterior probabilities for the Profile 3 were shown in Table 2. The average posterior probability of assignment for MHD patients was 94.6%-96.8%, indicating that the result of the Profile 3 was credible. Profile 1 (38.5%; n = 215) exhibited the lowest levels of stigma, classified as the “low stigma group”. Profile 2 (49.6%; n = 277) demonstrated moderate stigma scores, designated as the “moderate stigma group”. Profile 3 (11.8%; n = 66) showed the highest stigma levels, categorized as the “high stigma group” (Supplementary Fig. 1).
Table 1.
Fitting index table of different latent profile models
| model | AIC | BIC | aBIC | LMR | BLRT | Entropy | Category probability |
|---|---|---|---|---|---|---|---|
| 1 | 3865.712 | 3900.307 | 3874.911 | - | - | - | - |
| 2 | 2989.897 | 3046.114 | 3004.845 | < 0.001 | < 0.001 | 0.854 | 0.66/0.34 |
| 3 | 2365.025 | 2442.864 | 2385.723 | < 0.001 | < 0.001 | 0.910 | 0.38/0.50/0.12 |
| 4 | 2181.737 | 2281.197 | 2208.185 | 0.002 | < 0.001 | 0.868 | 0.25/0.41/0.26/0.08 |
| 5 | 2087.109 | 2208.191 | 2119.305 | 0.155 | 0.001 | 0.897 | 0.04/0.32/0.32/0.08/0.24 |
Table 2.
Average posterior probabilities of latent profile membership
| category | Number of subjects | Probability probability | ||
|---|---|---|---|---|
| C1 | C2 | C3 | ||
| 1 | 215(38.5%) | 0.952 | 0.048 | < 0.001 |
| 2 | 277(49.6%) | 0.027 | 0.968 | 0.005 |
| 3 | 66(11.8%) | < 0.001 | 0.054 | 0.946 |
Characteristics associated with stigma subgroups
A comparative analysis of the demographic and clinical characteristics of different latent profiles of stigma in patients with MHD was performed. The results showed significant univariate differences in residence, education, cohabitant type, family relationship, and income (p < 0.05). These associations remained significant after BH correction for multiple comparisons, with the exception of marital status, which became non-significant (p = 0.109). Consequently, marital status was excluded from subsequent multinomial logistic regression analysis. The full distribution of demographic variables by stigma subgroup was presented in Table 3. Post-hoc pairwise comparisons were performed for the five significant categorical variables with Bonferroni correction, as shown in Supplementary Table 1.
Multinomial logistic regression analysis was conducted to identify factors associated with stigma subgroup membership, using the low stigma group as the reference (Table 4). The overall model was statistically significant (χ² = 100.88, p < 0.001). The results showed that higher symptom burden was associated with lower odds of being in the moderate stigma group (OR = 0.818, 95% CI: 0.669-1.000, p = 0.049), but not the high stigma group (p = 0.725). Compared to urban residents (reference), those living in villages (OR = 0.529, p = 0.028) or townships (OR = 0.557, p = 0.023) had significantly lower odds of being in the moderate stigma group; this protective effect was even stronger for the high stigma group (village: OR = 0.193, p = 0.002; township: OR = 0.243, p = 0.001). Income also showed significant associations: patients with the lowest monthly income (< 1000 yuan) had significantly lower odds of being in the high stigma group compared to those with the highest income (> 5000 yuan) (OR = 0.281, 95% CI: 0.084–0.940, p = 0.039). For the moderate stigma group, a middle-income level (3000–5000 yuan) was associated with lower odds compared to the highest income group (OR = 0.404, p = 0.044).
Table 4.
Multivariable logistic regression analysis of latent profiles of stigma in patients on MHD
| Variables | Moderate stigma group | High stigma group | ||||||||
|---|---|---|---|---|---|---|---|---|---|---|
| B | p-value | Exp(B) | Exp (B) 95% CI | B | p-value | Exp(B) | Exp (B) 95% CI | |||
| Lower Bound | Upper Bound | Lower Bound | Upper Bound | |||||||
| intercept | 2.086 | 0.000 | -0.237 | 0.748 | ||||||
| symptom | -0.201 | 0.049 | 0.818 | 0.669 | 1.000 | 0.053 | 0.725 | 1.054 | 0.786 | 1.414 |
| village | -0.636 | 0.028 | 0.529 | 0.300 | 0.934 | -1.645 | 0.002 | 0.193 | 0.068 | 0.545 |
| township | -0.585 | 0.023 | 0.557 | 0.336 | 0.923 | -1.417 | 0.001 | 0.243 | 0.107 | 0.550 |
| county | -0.784 | 0.074 | 0.457 | 0.193 | 1.079 | -0.053 | 0.913 | 0.948 | 0.363 | 2.478 |
| city | 0b | 0b | ||||||||
| primary school or below | -0.776 | 0.051 | 0.460 | 0.211 | 1.003 | -0.520 | 0.429 | 0.595 | 0.164 | 2.154 |
| junior high school | -0.370 | 0.280 | 0.691 | 0.353 | 1.351 | -0.206 | 0.684 | 0.814 | 0.302 | 2.195 |
| high school and technical secondary school | 0.147 | 0.667 | 1.158 | 0.594 | 2.257 | -0.608 | 0.232 | 0.544 | 0.201 | 1.477 |
| junior college or above | 0b | 0b | ||||||||
| family relationship-good | -0.062 | 0.779 | 0.940 | 0.608 | 1.452 | 0.691 | 0.140 | 1.996 | 0.797 | 4.996 |
| family relationship-poor | 0b | 0b | ||||||||
| income < 1000 | -0.781 | 0.076 | 0.458 | 0.193 | 1.085 | -1.270 | 0.039 | 0.281 | 0.084 | 0.940 |
| income 1000–3000 | -0.847 | 0.055 | 0.429 | 0.180 | 1.019 | -0.553 | 0.341 | 0.575 | 0.184 | 1.794 |
| income 3000–5000 | -0.907 | 0.044 | 0.404 | 0.167 | 0.977 | -0.944 | 0.105 | 0.389 | 0.124 | 1.218 |
| income > 5000 | 0b | 0b | ||||||||
| solitude | -0.711 | 0.044 | 0.491 | 0.246 | 0.981 | -0.494 | 0.457 | 0.610 | 0.166 | 2.240 |
| with spouse | -0.441 | 0.049 | 0.643 | 0.415 | 0.998 | 0.717 | 0.066 | 2.048 | 0.955 | 4.395 |
| with children/parents | -0.613 | 0.081 | 0.541 | 0.272 | 1.078 | -0.380 | 0.598 | 0.684 | 0.167 | 2.805 |
| with spouse and children | 0b | 0b | ||||||||
Note: Refer to the low stigma group
Symptom network characteristics: overall and across stigma subgroups
The network comprised 30 nodes with 197 non-zero edges (mean weight = 0.032) (Fig. 2). The detailed weight matrix for all connections was provided in Supplementary Table 2. Among these connections, the strongest positive edges (weight ≥ 0.50) were observed between: S2 and S3 (weight = 0.66), S21 and S20 (weight = 0.60), S25 and S24 (weight = 0.57), S30 and S29 (weight = 0.88), indicating strong associations between these symptoms. Strength centrality, considered the most reliable indicator, identified the following core symptoms: Muscle pain (S18): strength = 1.807, Sadness (S27): strength = 1.463, Chest pain (S16): strength = 1.321, Headache (S17): strength = 1.142. (Fig. 3, detailed values in Supplementary Table 3). This means that these symptoms had the highest sum of connection strengths to all other symptoms in the network, suggesting they may act as key hubs or influencers within the symptom system. The correlation stability coefficient (CS-C) for strength centrality was 0.672, which is substantially above the recommended threshold of 0.50, indicating excellent stability of the network structure (Fig. 4). And calculation results also showed that the bootstrapped CIs were small, which showed good accuracy of the network (Fig. 5).
Fig. 2.
Network of stigma in MHD patients. Note: Blue edges indicate positive correlations, red edges indicate negative correlations and thicker edge paths indicate stronger path relationships
Fig. 3.
Strength centrality of symptoms in the overall network. Note: Symptoms were ordered by strength centrality (z-scored). S1; Constipation, S2; Nausea, S3; Vomiting, S4; Diarrhea, S5; Decreased Appetite, S6; Muscle cramps, S7; Swelling in legs, S8; Shortness of breath, S9; Dizziness, S10; Restless legs, S11; Numbness in the feet, S12; Fatigue, S13; Cough, S14; Dry mouth, S15; Joint pain, S16; Chest pain, S17; Headache, S18; Muscle pain, S19; Difficulty concentrating, S20; Dry skin, S21; Itching, S22; Worry, S23; Nervousness, S24; Trouble falling asleep, S25; Trouble staying asleep, S26; Feeling irritable, S27; Feeling sad, S28; Anxiety, S29; Decreased interest in sex, S30; Difficulty becoming sexually aroused
Fig. 4.
Stability of centrality indices by case dropping subset bootstrap. Note: The x-axis represents the percentage of cases of the original sample used at each step. The y-axis represents the average of correlations between the centrality indices from the original network and the centrality indices from the networks that were reestimated after excluding increasing percentages of cases
Fig. 5.
Bootstrap analysis results of the edge weights. Note: The red line represents the magnitude of edge weights in the sample of this study; the black line indicates the magnitude of the average edge weight value obtained through bootstrap method evaluation; the gray shading represents the 95% CIs derived from the bootstrap method
Comparison of symptom networks across stigma subgroups
Network models were estimated separately for each stigma subgroup (Fig. 6). Stability assessment revealed acceptable stability for the low stigma group (CS-C = 0.516) and moderate stigma group (CS-C = 0.671), but poor stability for the high stigma group (CS-C = 0.045), rendering its specific network structure unreliable. For the stable networks (low and moderate groups), visual inspection suggested differences in their patterns. The symptoms with the highest strength centrality were chest pain (S16) and muscle pain (S18) in the low stigma group, and sadness (S27) in the moderate stigma group (Fig. 6). A permutation test comparing global strength between these two stable groups found no statistically significant difference (p = 0.713). Regarding the high stigma group, given the unacceptable stability of its network estimate, it was not included in formal statistical comparisons. The visual pattern and centrality indices for this group (Figs. 6 and 7) were presented only to illustrate the challenges of estimating networks with small samples and should not be interpreted substantively.
Fig. 6.
Network diagram of three categories of stigma in MHD patients. Note: (a) low stigma group, (b) moderate stigma group, (c) high stigma group. Blue edges indicate positive correlations, red edges indicate negative correlations and thicker edge paths indicate stronger path relationships
Fig. 7.
Centrality index of three categories of stigma in MHD patients. Note: Symptoms were ordered by strength centrality (z-scored). S1; Constipation, S2; Nausea, S3; Vomiting, S4; Diarrhea, S5; Decreased Appetite, S6; Muscle cramps, S7; Swelling in legs, S8; Shortness of breath, S9; Dizziness, S10; Restless legs, S11; Numbness in the feet, S12; Fatigue, S13; Cough, S14; Dry mouth, S15; Joint pain, S16; Chest pain, S17; Headache, S18; Muscle pain, S19; Difficulty concentrating, S20; Dry skin, S21; Itching, S22; Worry, S23; Nervousness, S24; Trouble falling asleep, S25; Trouble staying asleep, S26; Feeling irritable, S27; Feeling sad, S28; Anxiety, S29; Decreased interest in sex, S30; Difficulty becoming sexually aroused
Discussion
This study identified three distinct stigma profiles among MHD patients. The majority (61.5%) experienced moderate to high stigma, which suggests its clinical relevance. Contrary to existing literature [22, 23], our multivariate model showed a marginal and counterintuitive inverse association between symptom burden and stigma subgroup membership. While statistically significant (p = 0.049), this finding should be interpreted with extreme caution due to the cross-sectional design, which precludes causal inference. First, small effect sizes and marginal p-values suggest the possibility of a Type I error. Alternatively, long-term hemodialysis patients may experience a “response shift”—those with severe and persistent symptoms may recalibrate their internal standards and come to view the “patient” role as normal, to prevent or hedge against stigma [24]. However, this assumption could not be verified in the present study. Additionally, there may be selection bias in this study: patients with high symptom burden and high stigma may demonstrate poorer treatment adherence or higher mortality rates, potentially resulting in “survivorship bias” within the study population [25]. This seemingly paradoxical finding should not be interpreted as evidence that high symptom burden protects patients from stigma.
Compared to urban residents, individuals living in villages or townships had lower odds of belonging to both the moderate and high stigma groups. Rural and township settings often feature closer-knit, more homogeneous social networks. These environments may foster mutual understanding and collective support, potentially buffering perceived stigma [26]. In contrast, urban residents usually have higher social participation, career development expectations, and self-actualization needs, and may be more sensitive to the damage to their social roles. Consequently, any perceived threat of disease to their career and social functioning may correlate more strongly with feelings of stigma [27].
Income level also showed significant associations with stigma subgroup membership. The lowest income group had lower odds of being in the high stigma group compared with the highest income group, while the middle income group had lower odds of being in the moderate stigma group than the highest income group. This finding indicates heterogeneous associations between economic status and stigma perception. One possible interpretation is that the lowest income group may have more limited access to information channels that convey social evaluations, potentially relating to a lower likelihood of perceiving external stigma [28]. Conversely, the highest income group may have greater social participation opportunities and pay more attention to their social image; these factors might correlate with increased sensitivity to the pressure of social stigma in the context of diseases [29].
Muscle pain (S18), sadness (S27), and chest pain (S16) were identified as core symptom nodes with the highest strength centrality. Pain is a subjective experience. Studies have shown that there is a significant positive correlation between stigma and pain intensity and depression [30]. These core symptoms may play a “hub” role in the entire symptom network. Their presence or severity is strongly linked to a series of other symptoms. Therefore, in clinical management, prioritizing the assessment and intervention of these core symptoms (such as by optimizing pain management and providing psychological support to alleviate sadness) could target key connections in the symptom network, potentially alleviating the overall symptom burden [31].
Our network structure revealed an important finding: the strongest connections concentrated within clinically coherent symptom pairs, forming distinct functional modules. Specifically, very strong edges were found between nausea-vomiting (gastrointestinal), itching-dry skin (dermatological), sleep onset-maintenance problems (insomnia), and decreased sexual interest-arousal (sexual dysfunction). These tightly coupled pairs likely represent shared underlying pathophysiological or psychological mechanisms. However, the nodes within these modules generally showed low strength centrality. This pattern suggests a dissociation between local cohesion within a module and global influence over the entire network. While symptoms within a module (e.g., S24 and S25) directly reinforce each other strongly, they interact less directly with the broader network of core physical and emotional symptoms (e.g., pain, sadness). These modules appear to function more as specialized subsystems. One potential clinical implication is a dual intervention strategy: (1) managing highly coupled symptom pairs as integrated units, and (2) targeting central hub symptoms (like core pain and emotional symptoms identified here) for interventions aimed at broadly reducing overall symptom burden due to their greater potential network-wide leverage.
This study found that while the symptom network structures of the low and moderate stigma subgroups were stable, there was no statistically significant difference in the global network strength between these two groups. Critically, the network estimation for the high stigma subgroup was highly unstable (CS-C = 0.045) due to insufficient sample size. This result primarily highlights the decisive impact of sample size on the robustness of network analysis—when the sample size-to-variable ratio is too low, regularized estimates are highly susceptible to sampling fluctuations. This suggests that future research must ensure an adequate sample size for each subgroup or consider employing Bayesian methods that incorporate prior information to enhance estimation reliability with small samples. Although the network for the high stigma subgroup could not be reliably estimated, the difficulty in recruiting participants for this group (n = 66) is itself an important clinical finding. This likely reflects the concealment and social isolation characteristics of this population due to high stigma, underscoring the urgent need for attention and research focused on this subgroup [32].
Within the low and moderate stigma subgroups, where results are stably interpretable, we observed a shift in core symptoms: the network of the low stigma group centered on chest pain (S16) and muscle pain (S18), suggesting a physically-driven symptom pattern; whereas the core symptom for the moderate stigma group was sadness (S27), supporting the hypothesis of an enhanced central role of emotion within the symptom network [33]. Notably, despite the difference in core nodes, there was no significant difference in global network strength between the two groups. This may indicate that the influence of stigma is more focused on reshaping the architecture of symptom connections rather than universally strengthening all connections. This finding awaits validation in future studies with larger samples and requires further confirmation through effect size estimation.
Limitations and future directions
This study also has some limitations. First, the cross-sectional design precludes causal inference and cannot account for all potential unmeasured confounders. Although key demographic variables like age did not differ between subgroups, residual confounding by other factors remains possible. Second, the data were derived from patients’ self-reports, which may have recall bias. Third, the network analysis using the EBICglasso method assumes data are continuous and approximately normally distributed. While this is a common approach for symptom severity scores, future studies could employ methods specifically designed for ordinal or mixed data types to confirm our findings. Additionally, the most critical limitation of this study lies in the insufficient sample size of the high stigma subgroup (n = 66), leading to extremely unstable estimates of its symptom network (CS-C = 0.045). Therefore, any inferences regarding the symptom interaction patterns of patients with high stigma should not be based on the results of this study.
Future research directions should include: (1) Conducting longitudinal cohort studies to track the dynamic evolution of stigma and symptom networks and identify whether there are causally driving symptoms in the network; (2) Adopting mixed-methods designs, combining qualitative interviews to deeply understand the differences in stigma experiences among patients in different subgroups, especially the concealment mechanisms of patients with high stigma; (3) Designing intervention studies based on the identification results of core symptoms to evaluate the potential improvement effect of targeted management of hub symptoms such as muscle pain and sadness on overall symptom burden and stigma experience; (4) On the basis of expanding the sample size in multicenter studies, ensuring that the high stigma subgroup obtains an adequate sample size, or adopting Bayesian methods to integrate prior information to improve the reliability of network estimation with small samples.
Conclusions
Using latent profile analysis and network analysis, this study revealed the three-category characteristics of stigma and the symptom network structure among maintenance hemodialysis patients. The main findings include: (1) Nearly two-thirds of patients belonged to the moderate-to-high stigma subgroups, indicating that stigma is highly prevalent in this population; (2) Muscle pain, sadness, and chest pain showed the highest connection strength in the symptom network, and these core symptoms may have strong associations with multiple other symptoms; (3) The network structures of the low and moderate stigma groups were similar in terms of global connectivity, but there were differences in the distribution of core symptoms, suggesting that the level of stigma may affect the relative importance of specific nodes in the symptom network; (4) Due to the limited sample size (n = 66), the reliability of network estimation for the high stigma subgroup was insufficient, which itself reflects the accessibility challenges of this population in research.
Supplementary Information
Below is the link to the electronic supplementary material.
Acknowledgements
Not applicable.
Author contributions
LL and NJ conceived the research project, and revised the final manuscript. NJ and HMH supervised the research work. ZFZ, JL, BL, and JYL collected and managed the data. NZ and WCG performed the data analysis, interpreted the data, and prepared the draft manuscript. NS, NZ, and WCG organized and finalized the manuscript.
Funding
This study was supported by the Shandong Provincial Natural Science Foundation (grant numbers ZR2025MS1169).
Data availability
The datasets used and analyzed during the current study are available from the corresponding author on reasonable request.
Declarations
Ethics approval and consent to participate
This study adheres to the relevant provisions of the Declaration of Helsinki and the “Measures for the Ethical Review of Biomedical Research Involving Human Subjects (Trial)”. It has been reviewed and approved by the Ethics Committee of Shandong First Medical University (Ethics Approval Number: R202401170010). All participants signed the informed consent form before the start of the study. Researchers explained the study content in detail to ensure that all participants understood. The collected data were strictly confidential and used only for this study.
Consent for publication
All authors reviewed the draft and approved the final version for submission.
Competing interests
The authors declare no competing interests.
Footnotes
Publisher’s note
Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.
Na Song, Ni Zhang and Weicui Gao contributed equally to this work.
Contributor Information
Lei Li, Email: lli@sdfmu.edu.cn.
Ning Jiang, Email: maomaoriv@163.com.
References
- 1.Flythe JE, Watnick S. Dialysis for chronic kidney failure: A review. JAMA. 2024;332(18):1559–73. 10.1001/jama.2024.16338. [DOI] [PubMed] [Google Scholar]
- 2.Zhang Y, Huang Y, Qiu Y, et al. Trajectories and influencing factors of psychological resilience among Chinese patients with maintenance Hemodialysis. Heliyon. 2024;10(17):e37200. 10.1016/j.heliyon.2024.e37200. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 3.Oexle N, Müller M, Kawohl W, et al. Self-stigma as a barrier to recovery: a longitudinal study. Eur Arch Psychiatry Clin Neurosci. 2018;268(2):209–12. 10.1007/s00406-017-0773-2. [DOI] [PubMed] [Google Scholar]
- 4.Li Y, Yang Y, Liang S, et al. Understanding perceived stigma and depression symptoms in maintenance Hemodialysis patients: a network perspective. Front Psychol. 2025;16:1552518. 10.3389/fpsyg.2025.1552518. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 5.Lee EJ, Chang AK, Chung YC. Socioecological factors affecting fluid restriction adherence among Korean patients receiving hemodialysis: A qualitative study. J Transcult Nurs. 2021;32(3):239–47. 10.1177/1043659620919162. [DOI] [PubMed] [Google Scholar]
- 6.Kane JC, Elafros MA, Murray SM, et al. A scoping review of health-related stigma outcomes for high-burden diseases in low- and middle-income countries. BMC Med. 2019;17(1):17. 10.1186/s12916-019-1250-8. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 7.Elias MA, Van Damme W, Ku GMV, et al. Lived experiences of people with chronic kidney disease on maintenance dialysis: a systematic review and thematic synthesis of qualitative studies. BMC Nephrol. 2025;26(1):22. 10.1186/s12882-025-03952-4. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 8.Zhuo Q, Xu N, Wang H, et al. Analysis of the status and associated factors of stigma in patients undergoing maintenance Hemodialysis. Clin Nephrol. 2025;103(3):161–71. 10.5414/cn111405. [DOI] [PubMed] [Google Scholar]
- 9.Bandura A. Health promotion from the perspective of social cognitive theory. Psychol Health. 1998;13(4):623–49. 10.1080/08870449808407422. [Google Scholar]
- 10.Borsboom D. Psychometric perspectives on diagnostic systems. J Clin Psychol. 2008;64(9):1089–108. 10.1002/jclp.20503. [DOI] [PubMed] [Google Scholar]
- 11.Fife BL, Wright ER. The dimensionality of stigma: a comparison of its impact on the self of persons with HIV/AIDS and cancer. J Health Soc Behav. 2000;41(1):50–67. 10.2307/2676360. [PubMed] [Google Scholar]
- 12.Pan AW, Chung L, Fife BL, et al. Evaluation of the psychometrics of the social impact scale: a measure of stigmatization. Int J Rehabil Res. 2007;30(3):235–8. 10.1097/MRR.0b013e32829fb3db. [DOI] [PubMed] [Google Scholar]
- 13.Hao Y. Study of symptoms in hemodialysis patients [master’s thesis]. Beijing: Peking Union Medical College; 2016. Chinese. 10.7666/d.Y3075706
- 14.Weisbord SD, Fried LF, Arnold RM, et al. Development of a symptom assessment instrument for chronic Hemodialysis patients: the Dialysis symptom index. J Pain Symptom Manage. 2004;27(3):226–40. 10.1016/j.jpainsymman.2003.07.004. [DOI] [PubMed] [Google Scholar]
- 15.Song YY, Chen L, Yu WW, et al. Correlates of symptom burden of Hemodialysis patients. West J Nurs Res. 2021;43(5):459–67. 10.1177/0193945920957229. [DOI] [PubMed] [Google Scholar]
- 16.Zhou M, Gu X, Cheng K, et al. Exploration of symptom clusters during Hemodialysis and symptom network analysis of older maintenance Hemodialysis patients: a cross-sectional study. BMC Nephrol. 2023;24(1):115. 10.1186/s12882-023-03176-4. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 17.Sinha P, Calfee CS, Delucchi KL. Practitioner’s guide to latent class analysis: methodological considerations and common pitfalls. Crit Care Med. 2021;49(1):e63–79. 10.1097/ccm.0000000000004710. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 18.Hammersen F, Pursche T, Fischer D, et al. Psychosocial and family-centered support among breast cancer patients with dependent children. Psychooncology. 2021;30(3):361–8. 10.1002/pon.5585. [DOI] [PubMed] [Google Scholar]
- 19.Epskamp S, Borsboom D, Fried EI. Estimating psychological networks and their accuracy: A tutorial paper. Behav Res Methods. 2018;50(1):195–212. 10.3758/s13428-017-0862-1. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 20.Li Y, Jia S, Cao B, et al. Network analysis of somatic symptoms in Chinese patients with depressive disorder. Front Public Health. 2023;11:1079873. 10.3389/fpubh.2023.1079873. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 21.van Borkulo CD, van Bork R, Boschloo L, et al. Comparing network structures on three aspects: A permutation test. Psychol Methods. 2023;28(6):1273–85. 10.1037/met0000476. [DOI] [PubMed] [Google Scholar]
- 22.Williamson TJ, Garon EB, Irwin MR, et al. Sleep disturbance as a mediator of lung cancer stigma on psychological distress and physical symptom burden. Psychosom Med. 2024;86(4):334–41. 10.1097/psy.0000000000001299. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 23.Jankowiak B, Kowalewska B, Krajewska-Kułak E, et al. Stigmatization and quality of life in patients with psoriasis. Dermatol Ther (Heidelb). 2020;10(2):285–96. 10.1007/s13555-020-00363-1. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 24.Ji Q, Zhang L, Xu J, et al. The relationship between stigma and quality of life in hospitalized middle-aged and elderly patients with chronic diseases: the mediating role of depression and the moderating role of psychological resilience. Front Psychiatry. 2024;15:1346881. 10.3389/fpsyt.2024.1346881. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 25.Buchman-Wildbaum T, Váradi E, Schmelowszky Á, et al. Targeting the problem of treatment non-adherence among mentally ill patients: the impact of loss, grief and stigma. Psychiatry Res. 2020;290:113140. 10.1016/j.psychres.2020.113140. [DOI] [PubMed] [Google Scholar]
- 26.Takada S, Nyakato V, Nishi A, et al. The social network context of HIV stigma: Population-based, sociocentric network study in rural Uganda. Soc Sci Med. 2019;233:229–36. 10.1016/j.socscimed.2019.05.012. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 27.Alemu WG, Tilahun SY, Bekele E, et al. Prevalence and associated factors of perceived stigma among medically ill patients on follow-up screened positive for depression in ethiopia: facility-based cross-sectional study. BMJ Open. 2022;12(1):e056665. 10.1136/bmjopen-2021-056665. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 28.Islam A, Pakrashi D, Vlassopoulos M, et al. Stigma and misconceptions in the time of the COVID-19 pandemic: A field experiment in India. Soc Sci Med. 2021;278:113966. 10.1016/j.socscimed.2021.113966. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 29.Kato A, Fujimaki Y, Fujimori S, et al. A qualitative study on the impact of internalized stigma on type 2 diabetes self-management. Patient Educ Couns. 2016;99(7):1233–9. 10.1016/j.pec.2016.02.002. [DOI] [PubMed] [Google Scholar]
- 30.Scott W, Buchman DZ, VasiIiou VS. The multi-dimensional stigma of chronic pain: A narrative review. Curr Opin Psychol. 2025;62:101980. 10.1016/j.copsyc.2024.101980. [DOI] [PubMed] [Google Scholar]
- 31.Zhou S, Zhang Y, He H, et al. Symptom clusters and networks analysis in acute-phase stroke patients: a cross-sectional study. Sci Rep. 2025;15(1):2539. 10.1038/s41598-024-84642-3. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 32.Crooks S, Mitchell G, Wynne L, et al. Exploring the stigma experienced by people affected by parkinson’s disease: a systematic review. BMC Public Health. 2025;25(1):25. 10.1186/s12889-024-21236-8. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 33.Zhang C, Wang F, Kang Z, et al. Effect of symptom burden on demoralization in Chinese lung cancer patients: the mediating roles of family Function, Resilience, and coping behaviors. Psychooncology. 2025;34(2):e70102. 10.1002/pon.70102. [DOI] [PubMed] [Google Scholar]
Associated Data
This section collects any data citations, data availability statements, or supplementary materials included in this article.
Supplementary Materials
Data Availability Statement
The datasets used and analyzed during the current study are available from the corresponding author on reasonable request.