Latent profile analysis identifying symptom clusters of fatigue, sleep disturbance, pain, and anxiety and their associations with health-related quality of life and self-efficacy in patients undergoing peritoneal dialysis: a cross-sectional study Authors: Yiting Zeng, Dandan Zhang, Yuzhen Fang, Mingyi Xiao, Chaona Ji, Xisui Chen Published: 2026 ABSTRACT AbstractBackgroundPeritoneal dialysis patients experience a high burden of co-occurring symptoms that impair health-related quality of life and self-efficacy. This study applied latent profile analysis to identify distinct symptom profiles of fatigue, sleep disturbance, pain, and anxiety and to examine factors associated with profile membership and their associations with patient-reported outcomes.MethodsIn this cross-sectional study conducted from January to October 2025, 308 peritoneal dialysis patients from a tertiary hospital in Shantou, China, completed the Multidimensional Fatigue Inventory-20, Athens Insomnia Scale, Numerical Rating Scale, Zung Self-Rating Anxiety Scale, 36-Item Short Form Health Survey, and Chronic Disease Self-Efficacy Scale. Latent profile analysis was performed using Mplus 8.3. Factors associated with profile membership were examined using exploratory three-step analyses (R3STEP), and differences in health-related quality of life and self-efficacy across profiles were compared using the BCH approach.ResultsThree profiles were identified: Low symptom distress (n = 182, 59.1%), Moderate symptom distress (n = 57, 18.5%), and High symptom distress (n = 69, 22.4%). In exploratory R3STEP analyses accounting for classification uncertainty, higher phosphorus was associated with greater odds of membership in the moderate and high symptom distress profiles relative to the low symptom distress profile, while BMI showed a borderline association. BCH analyses further showed significant differences across profiles in physical and mental health-related quality of life and self-efficacy, with the high symptom distress profile showing the poorest outcomes.ConclusionsThree clinically meaningful symptom profiles exist among peritoneal dialysis patients. Higher symptom distress was associated with poorer health-related quality of life and lower self-efficacy, and exploratory three-step analyses suggested that phosphorus may be an exploratory correlate of membership in the moderate and high symptom distress profiles. These findings suggest that symptom cluster-based screening may help identify peritoneal dialysis patients with greater supportive care needs and may inform future multidisciplinary management strategies.Supplementary InformationThe online version contains supplementary material available at 10.1186/s12955-026-02547-3.Keywords: Latent profile analysis, Symptom clusters, Peritoneal dialysis, Health-related quality of life, Self-efficacy, Patient-reported outcomes Introduction Peritoneal dialysis (PD), as a home-based renal replacement therapy, remains the preferred treatment option for many patients with end-stage renal disease to this day. Compared with hemodialysis, peritoneal dialysis offers greater protection of residual renal function and, at the same time, greater flexibility and autonomy in daily life [1, 2]. However, patients undergoing peritoneal dialysis still have to deal with various physical and psychological symptoms, which often significantly reduce their health-related quality of life and affect their overall health status [3, 4]. Among patients undergoing peritoneal dialysis, fatigue, sleep disturbance, pain, and anxiety are among the most common and burdensome symptoms. Several large-sample cohort studies on peritoneal dialysis patients in China have shown that the incidence of fatigue ranges from 35% to 78%, sleep disturbance as high as 80%, and about half of the patients are troubled by moderate to severe pain. In comparison, the incidence of clinically significant anxiety is 30% to 50% [5–9]. It is worth noting that these symptoms rarely occur alone but often co-occur and may be interrelated through biological, neuroendocrine, and psychosocial pathways described in prior literature [10–13]. For instance, sleep disturbance can exacerbate fatigue by interfering with the body’s physiological repair process, while persistent pain and anxiety may reinforce each other, amplifying nociceptive sensations and maintaining a continuous excited state of the sympathetic nerve [12, 13]. Existing evidence consistently indicates that such interrelated symptom clusters are closely related to decreased treatment compliance, increased hospitalization risk, and reduced self-efficacy in long-term chronic disease management [14, 15]. Although human-centered analytical methods (such as latent profile analysis and network analysis) have demonstrated significant value in symptom cluster studies of hemodialysis patients and cancer patients [16–18], their application in the field of peritoneal dialysis remains very limited. The few current network analysis studies on peritoneal dialysis patients in China have found that fatigue is a key bridging node connecting psychological distress and gastrointestinal symptoms [19, 20]. However, to date, few studies have adopted latent profile analysis methods to identify different subgroups in peritoneal dialysis patients based on the coexistence patterns of fatigue, sleep disturbance, pain, and anxiety. Given the substantial burden of kidney disease symptoms and impaired health-related quality of life among patients undergoing peritoneal dialysis, identifying clinically meaningful symptom subgroups may help support more individualized symptom management strategies [14]. In the Chinese PD care context, where home-based dialysis places considerable demands on patient and family self-management, understanding how co-occurring symptoms relate to health-related quality of life and self-efficacy may be particularly relevant for routine clinical follow-up and supportive care planning. In this cross-sectional study, we aimed to: (1) apply latent profile analysis to identify subgroups of peritoneal dialysis patients based on the co-occurrence patterns of fatigue, sleep disturbance, pain, and anxiety; and (2) compare these subgroups with respect to health-related quality of life, chronic disease self-efficacy, and selected demographic, clinical, and biochemical characteristics. To our knowledge, few studies have applied latent profile analysis to examine these four co-occurring symptoms in patients undergoing peritoneal dialysis, and even fewer have linked the resulting symptom profiles to health-related quality of life and self-efficacy using methods that account for classification uncertainty. Methods Participants and setting This cross-sectional study was conducted at the Peritoneal Dialysis Center of the First Affiliated Hospital of Shantou University Medical College (a Grade-A tertiary hospital in Shantou City, China) from January to October 2025. The study continuously included adult patients aged 18 years or older who received regular peritoneal dialysis treatment for at least 3 months. Patients were excluded if they were judged by the attending clinicians and research nurses to have evident cognitive impairment or insufficient decisional capacity to understand the study procedures or complete the questionnaires. No formal cognitive screening instrument was administered. All research subjects signed a written informed consent form before enrollment. This research protocol was approved by the Medical Ethics Committee of the First Affiliated Hospital of Shantou University Medical College (Ethics Approval Number: B-2025-013), and the entire process strictly adhered to the ethical principles stipulated in the Declaration of Helsinki. Measures Symptom assessment Fatigue was evaluated using the Chinese version of the Multidimensional Fatigue Inventory-20 (MFI-20) [21]. This scale consists of 20 items, covering five dimensions: general fatigue, physical fatigue, reduced activity, decreased motivation, and mental fatigue. The subjects assigned a 5-point Likert score to each item, with the total score ranging from 20 to 100. The higher the score, the more severe the fatigue. The Chinese version of MFI-20 has good reliability and validity in the dialysis population, and the Cronbach’s α coefficient of the sample in this study is 0.937. Sleep disturbance were measured using the Chinese version of the Athens Insomnia Scale (AIS) [22]. This scale consists of 8 items, which respectively assess difficulty falling asleep, nocturnal awakening, early awakening, total sleep time, sleep quality, daytime well-being, functional status, and daytime sleepiness. Each item is scored from 0 to 3 points, with a total score ranging from 0 to 24 points. The higher the score, the more severe the insomnia. The Chinese version of AIS shows stable performance in patients with chronic diseases, and Cronbach’s α in this study is 0.907. The intensity of pain was evaluated using an 11-point Numerical Rating Scale (NRS) (0 = “no pain,” 10 = “the most severe pain imaginable”), and this single-item tool has been widely validated among patients with chronic kidney disease in China [23]. Single-item scales do not calculate internal consistency reliability. The anxiety state was evaluated by the Chinese version of the Zung Self-Rating Anxiety Scale (SAS) [24]. This scale consists of 20 items, with scores ranging from 20 to 80. The higher the score, the more severe the anxiety. A standard score of ≥ 50 indicates the presence of clinically significant anxiety. The Chinese version of SAS has excellent reliability and validity in dialysis patients. The Cronbach’s α in this study was 0.875. Health-related quality of life was measured using the Chinese version of the Medical Outcomes Study 36-Item Short Form Survey (SF-36) [25]. This scale provides scores in 8 domains as well as the total Physical Health score (PCS) and the Total Mental Health Score (MCS). The scores of each dimension are converted from 0 to 100 points, with higher scores indicating better functional status. This study adopted a standardized score based on the 1998 norm of the general population in the United States (mean 50, standard deviation 10). The Chinese version of SF-36 has been fully validated in Chinese patients with end-stage renal disease, and the overall Cronbach’s α of this sample is 0.946. The self-efficacy of chronic disease management was evaluated using the Chinese version of the Chronic Disease Self-Efficacy Scale (CDSES) [26]. This scale consists of 6 items. Subjects rate their confidence in coping with disease-related challenges on a scale of 1 to 10 and take the average of the items. The higher the score, the stronger the self-efficacy. The Chinese CDSES has demonstrated solid psychometric performance in dialysis patients, and the Cronbach’s α in this study is 0.849. Demographic and clinical variables We gathered demographic information—age, sex, education level, marital status, employment, monthly household income, residential area (urban or rural), and type of health insurance—together with key clinical and laboratory parameters. These included the primary etiology of end-stage renal disease, duration of peritoneal dialysis (dialysis vintage), method of catheter placement, body mass index, serum albumin, hemoglobin, serum creatinine, blood urea nitrogen, serum potassium, calcium, phosphate, C-reactive protein, and total weekly Kt/V. Data were extracted from electronic medical records and supplemented by a structured general information questionnaire completed by each participant. Data collection and statistical analysis subsection Data collection and quality control Data were collected from January to October 2025 among patients receiving peritoneal dialysis who were either hospitalized in the Department of Nephrology or attending regular follow-up visits at the Peritoneal Dialysis Centre of the First Affiliated Hospital of Shantou University Medical College. Trained nurses distributed and supervised completion of the paper questionnaires during routine clinical encounters, whether on ward rounds for inpatients or in the outpatient clinic. The questionnaire battery consisted of a general information form (demographic and basic clinical variables), the Chinese Multidimensional Fatigue Inventory-20 (MFI-20), the Athens Insomnia Scale (AIS), the 11-point Numerical Rating Scale (NRS) for pain, the Zung Self-Rating Anxiety Scale (SAS), the Chinese 36-Item Short Form Health Survey (SF-36), and the Chronic Disease Self-Efficacy Scale (CDSES). To ensure data quality, two trained research assistants independently reviewed all completed questionnaires on the day of collection to check for completeness, legibility, and internal consistency. Any omissions or ambiguous responses were clarified immediately with the participant whenever possible. After verification, data were entered into a secure electronic database and subsequently exported to IBM SPSS Statistics 26.0 for data cleaning, including range checks and identification of implausible values. Of the 338 participants who completed the questionnaire battery, 30 (8.9%) were excluded from the analytic sample because of missing data on at least one of the four symptom indicators used to derive the latent profiles (MFI-20, AIS, NRS, or SAS). Because these indicators were required for latent profile estimation, a complete-case approach was used for the primary analyses. To evaluate the robustness of this decision, we additionally conducted a basic sensitivity analysis in the full sample using full-information maximum likelihood (FIML). Detailed item-level or scale-specific missingness patterns within the excluded cases were not retained. To assess the potential impact of these exclusions, included and excluded participants were compared on available demographic and clinical characteristics (Supplementary Table S1). Latent profile analysis Latent profile analysis (LPA) was conducted in Mplus 8.3 [27] using four standardized symptom indicators: fatigue, sleep disturbance, pain, and anxiety. Models with one to five profiles were estimated sequentially using robust maximum likelihood (MLR). We first evaluated a parsimonious specification with equal indicator variances across profiles and zero within-profile covariances. For the retained 3-profile solution, we additionally explored more flexible variance-covariance specifications. The comparative results for these alternative variance-covariance specifications are presented in Supplementary Table S3. However, these more complex models did not converge normally and were therefore not interpreted. The optimal number of profiles was determined based on a combination of statistical indices and clinical judgment: (1) lower values of the Akaike Information Criterion (AIC), Bayesian Information Criterion (BIC), and sample-size-adjusted BIC (aBIC) [28, 29]; (2) entropy ≥ 0.80, indicating high classification accuracy; (3) statistically significant Lo-Mendell-Rubin adjusted likelihood ratio test (LMR) and bootstrapped likelihood ratio test (BLRT) when comparing the k-profile model with the (k − 1)-profile model [28–30]; (4) no profile representing less than 5% of the sample; and (5) clinically meaningful and theoretically coherent profile patterns. After careful evaluation of these criteria, the three-profile solution was selected as the best-fitting and most parsimonious model. The full fit indices are presented in Table 1. Table 1Model fit indices for latent profile models with one to five profilesModelLogLikParametersAICBICaBICEntropyLMR pBLRT pSmallest class %Interpretable?1-profile-4187.78788391.5758421.4158396.043———100Yes2-profile-3733.405137492.817541.3017500.0710.97500.0000*35.4Yes3-profile-3650.819187337.6387404.787347.6920.9610.03790.0000*18.1Yes4-profile-3596.774237239.5497325.3417252.3950.9070.09320.0000*16Yes5-profile-3546.685287149.3717253.8137165.0090.9280.00650.0000**1.6NoLogLik, loglikelihood; AIC, Akaike information criterion; BIC, Bayesian information criterion; aBIC, sample-size adjusted BIC; LMR, Lo-Mendell-Rubin adjusted likelihood ratio test; BLRT, bootstrap likelihood ratio test. BLRT results should be interpreted cautiously because the number of successful bootstrap draws was very small in some models. ** For the 5-class model, although the LMR and BLRT remained significant, the smallest class comprised only 1.6% of the sample and was therefore considered too small for a stable and clinically meaningful solution Model fit indices for latent profile models with one to five profiles LogLik, loglikelihood; AIC, Akaike information criterion; BIC, Bayesian information criterion; aBIC, sample-size adjusted BIC; LMR, Lo-Mendell-Rubin adjusted likelihood ratio test; BLRT, bootstrap likelihood ratio test. BLRT results should be interpreted cautiously because the number of successful bootstrap draws was very small in some models. ** For the 5-class model, although the LMR and BLRT remained significant, the smallest class comprised only 1.6% of the sample and was therefore considered too small for a stable and clinically meaningful solution Although no formal a priori sample size calculation was performed specifically for LPA, the present analysis included four indicators, a total sample of 308 participants, and a retained three-profile solution with adequate class proportions and high entropy, which is generally considered acceptable for estimating relatively simple latent profile models in applied research [28–30]. Validation of latent profiles To examine variables associated with latent profile membership while accounting for classification uncertainty, exploratory three-step analyses (R3STEP) were conducted in Mplus. Based on clinical relevance and parsimony, the following candidate covariates were examined individually: age, comorbidity, dialysis vintage, albumin, hemoglobin, phosphorus, log-transformed C-reactive protein, and body mass index. Distal outcomes, including the SF-36 Physical Component Summary (PCS), Mental Component Summary (MCS), and the Chronic Disease Self-Efficacy Scale (CDSES), were compared across latent profiles using the BCH approach in Mplus, which accounts for uncertainty in profile classification. Descriptive analyses were performed in IBM SPSS Statistics 26.0. A two-tailed p value < 0.05 was considered statistically significant. Results Description of the sample Of the 350 consecutive peritoneal dialysis patients initially approached, 338 completed the questionnaires, yielding a response rate of 96.6%. After excluding 30 individuals with missing data on any of the four core symptom measures, the final sample for analysis included 308 patients. The mean age of the total sample was 53.5 ± 13.0 years (range 20–90), with 52.6% of participants being male. Most participants had an education level of junior high school or below (81.5%), a monthly household income of less than 1000 RMB (63.0%), were unemployed (67.9%), and were married (90.9%). Chronic glomerulonephritis was the leading cause of end-stage renal disease (62.0%). The mean dialysis vintage was 47.83 ± 34.85 months (range 3–161), Kt/V was 1.749 ± 0.681, hemoglobin was 98.40 ± 24.83 g/L, and serum albumin was 32.6 ± 4.95 g/L. Detailed demographic, clinical, and biochemical characteristics, stratified by latent profile, are presented in Table 2. Table 2Demographic, clinical, and biochemical characteristics of the overall sample and by latent symptom profile (N = 308)CharacteristicsTotal(N = 308)Profile 1(N = 182)Profile 2(N = 57)Profile 3(N = 69) P Age(y)53.5 ± 1352.15 ± 12.6454.49 ± 14.7956.10 ± 12.050.079Age classification (%)0.15518-44y79(25.65)51(28.02)15(26.32)13(18.84)45-59y128(41.56)81(44.51)20(35.09)27(39.13)≥ 60y101(32.79)50(27.47)22(38.60)29(42.03)Sex(%)0.470Male162(52.6)94(51.65)34(59.65)34(49.28)Female146(47.4)88(48.35)23(40.35)35(50.72)Education Level(%)0.846Less than high school251(81.49)145(79.67)47(82.46)59(85.51)High school36(11.69)23(12.64)6(10.53)7(10.14)College/Undergraduate21(6.82)14(7.69)4(7.01)3(4.35)Marital Status(%)0.562Married280(90.91)162(89.01)53(92.98)65(94.20)Unmarried25(8.12)18(9.89)3(5.26)4(5.80)Divorced3(0.97)2(1.1)1(1.75)0Occupation(%)0.662White-collar employees27(8.77)14(7.69)6(10.53)7(10.14)Blue-collar and self-employed72(23.4)45(24.73)15(26.32)12(17.39)Not currently employed209(67.9)123(67.58)36(63.16)50(72.46)Economic status(%)0.674Below 1000 RMB/month194(63.0)119(65.38)32(56.14)43(62.32)1000–3000 RMB/month58(18.8)33(18.13)12(21.05)13(18.84)3000–5000 RMB/month34(11.0)17(9.34)10(17.54)7(10.14)Above 5000 RMB/month22(7.1)13(7.14)3(5.26)6(8.70)Living arrangement(%)0.970Living alone15(4.87)9(4.95)3(5.26)3(4.35)Living with others293(95.13)173(95.05)54(94.74)66(95.65)Health Insurance Type(%)0.510New Rural Cooperative Medical Insurance74(24.03)47(25.82)14(24.56)13(18.84)Urban Resident Health Insurance234(75.97)135(74.18)43(75.44)56(81.16)Primary Disease(%)0.935Chronic glomerular nephritis191(62.01)112(61.54)37(64.91)42(60.87)Diabetic nephropathy37(12.01)23(12.64)5(8.77)9(13.04)Hypertensive nephropathy39(12.66)25(13.74)7(12.28)7(10.14)Other41(13.31)22(12.09)8(14.04)11(15.94)Dialysis Catheter Type(%)0.514Open abdominal surgery176(57.14)107(58.79)30(52.63)39(56.52)Laparoscopic surgery117(37.99)65(35.71)23(40.35)29(42.03)Unknown15(4.87)10(5.49)4(7.02)1(1.45)Dialysis vintage(months)47.83 ± 34.8541.47 ± 34.5045.37 ± 36.066.65 ± 27.85<0.001Dialysis frequency (bags/day)3.91 ± 0.4823.92 ± 0.5093.91 ± 0.5103.87 ± 0.3800.735BMI (kg/m²)0.438Underweight (< 18.5)12(3.90)7(3.85)1(1.75)4(5.80)Normal weight (18.5–23.9)158(51.30)96(52.75)27(47.37)35(50.72)Overweight (24-29.9)122(39.61)72(39.56)25(43.86)25(36.23)Obese (≥ 30)16(5.19)7(3.85)4(7.02)5(7.25)Laboratory parametersAlb(g/L)32.6 ± 4.9532.4 ± 4.9533.0 ± 4.7532.6 ± 5.170.703Hb (g/L)98.40 ± 24.83106.41 ± 23.6399.03 ± 22.9975.93 ± 13.05<0.001Scr (µmol/L)934.87 ± 339.14968.10 ± 334.58910.28 ± 325.23867.55 ± 355.140.092BUN (mmol/L)19.23 ± 7.3420.18 ± 7.6218.89 ± 6.6617.02 ± 6.660.008Serum Potassium (mmol/L)4.0 ± 2.414.03 ± 2.744.01 ± 1.653.82 ± 0.840.848Serum Calcium (mmol/L)2.15 ± 0.262.13 ± 0.222.16 ± 0.262.16 ± 0.360.635Serum Phosphorus (mmol/L)1.67 ± 0.541.68 ± 0.541.78 ± 0.551.57 ± 0.50.089CRP(mg/L)4.70 (1.39–15.90)6.20 (1.42–16.90)3.68 (1.55–9.17)4.70 (1.10–16.90)0.5908Kt/V1.749 ± 0.6811.806 ± 0.7351.69 ± 0.561.649 ± 0.6190.206SF-36 Physical Component Summary (PCS)*58.83 ± 24.3973.29 ± 13.4843.84 ± 19.6033.07 ± 21.54<0.001SF-36 Mental Component Summary (MCS)*63.12 ± 16.7471.03 ± 10.5356.05 ± 15.8648.09 ± 18.02<0.001CDSES25.6 ± 13.8732.93 ± 10.6218.19 ± 10.7112.36 ± 10.27<0.001Categorical variables are shown as n (%). Normally distributed continuous variables are presented as mean ± SD, whereas CRP is presented as median (interquartile range). Group differences were tested using one-way ANOVA, Kruskal–Wallis tests, or χ²/Fisher’s exact tests as appropriate. Abbreviations: KT/V, urea clearance; BMI, body mass index; BUN, blood urea nitrogen; Scr, serum creatinine; Hb, hemoglobin; Alb, serum albumin; CRP, C-reactive protein; HRQoL, health-related quality of life assessed using the Chinese version of the SF-36 Health Survey; CDSES, Chronic Disease Self-Efficacy Scale. PCS, Physical Component Summary; MCS, Mental Component Summary. Both PCS and MCS are norm-based scores using 1998 U.S. general population orthogonal factor coefficients (mean = 50, SD = 10). Higher scores indicate better health-related quality of life Demographic, clinical, and biochemical characteristics of the overall sample and by latent symptom profile (N = 308) Categorical variables are shown as n (%). Normally distributed continuous variables are presented as mean ± SD, whereas CRP is presented as median (interquartile range). Group differences were tested using one-way ANOVA, Kruskal–Wallis tests, or χ²/Fisher’s exact tests as appropriate. Abbreviations: KT/V, urea clearance; BMI, body mass index; BUN, blood urea nitrogen; Scr, serum creatinine; Hb, hemoglobin; Alb, serum albumin; CRP, C-reactive protein; HRQoL, health-related quality of life assessed using the Chinese version of the SF-36 Health Survey; CDSES, Chronic Disease Self-Efficacy Scale. PCS, Physical Component Summary; MCS, Mental Component Summary. Both PCS and MCS are norm-based scores using 1998 U.S. general population orthogonal factor coefficients (mean = 50, SD = 10). Higher scores indicate better health-related quality of life Latent profile analysis Latent profile models with one to five profiles were estimated (Table 1). For the retained 3-profile solution, comparison of alternative variance-covariance specifications showed that the model with equal variances and zero within-profile covariances converged normally and yielded an interpretable solution, whereas the more flexible specifications with free variances and/or free covariances did not terminate normally and were therefore not interpreted (Supplementary Table S3). The AIC, BIC, and aBIC consistently decreased as additional profiles were added. Entropy remained high (> 0.90) up to the five-profile solution, except for the four-profile model (0.907). The LMR test was significant for the two-, three-, and five-profile solutions, while the four-profile model did not significantly improve over the three-profile model (p = 0.0932). The BLRT remained significant (p < 0.001) up to the four-profile model. Although the statistical indices continued to improve with more profiles, the four- and five-profile solutions resulted in fewer theoretically meaningful or stable classes, including a very small class of n = 5 in the five-profile model. Consequently, the three-profile solution was chosen for its optimal balance of statistical fit, parsimony, entropy (0.961), and clinical interpretability, and it was retained for further analyses. Sensitivity analysis As a basic sensitivity analysis, we re-estimated the latent profile models in the full sample using full-information maximum likelihood (FIML). The 3-profile solution was again supported, and the class distribution was similar to that of the primary complete-case analysis (192/68/78 vs. 182/57/69), suggesting that the identified symptom profile structure was robust to the original handling of missing data. Detailed fit indices for the sensitivity analysis are provided in Supplementary Table S2. Symptom profiles of the three latent profile analyses Figure 1 illustrates the three latent profiles based on standardized z-scores for the four symptom indicators. Profile 1 (Low symptom distress; n = 182, 59.1%) displayed low levels across all four symptoms. Profile 2 (Moderate symptom distress; n = 57, 18.5%) showed moderate elevations in all domains. Profile 3 (High symptom distress; n = 69, 22.4%) was characterized by significantly elevated scores across all four symptoms. The average latent profile posterior probabilities were 0.998, 0.961, and 0.957 for Profiles 1, 2, and 3, respectively, indicating excellent classification quality. Latent profiles of fatigue, sleep disturbance, pain, and anxiety among 308 peritoneal dialysis patients (N = 308) Low Symptom Distress (Profile 1): included 182 patients, accounting for 59.1% of the total sample. The scores of all four symptoms were at a relatively low level, and the overall symptom burden was the lightest. Therefore, it was named the low symptom distress profile. Moderate Symptom Distress (Profile 2): included 57 patients, accounting for 18.5% of the total sample. All four symptoms showed a moderate increase. The patients complained of obvious fatigue, intermittent insomnia, mild to moderate pain, and an increase in anxiety. Although it has had a certain impact on daily functions, it has not yet reached a severe level, so it was named the moderate symptom distress group. High Symptom Distress (Profile 3): included 69 patients, accounting for 22.4% of the total sample. The scores of all four symptoms were significantly higher than those of the other two groups. Patients were burdened with heavier physical and psychological symptoms, and their daily lives and overall health were significantly impaired. Therefore, they were named the highly troubled symptom group. Three latent symptom profiles were identified through latent profile analysis in 308 peritoneal dialysis patients, based on standardized z-scores (mean = 0, SD = 1) for the four symptom indicators: Differences in clinical variables, HRQoL, and self-efficacy between profiles Univariate analysis across the latent profile Univariate comparisons of demographic and clinical characteristics among the three latent profiles are shown in Table 2. There were no significant differences across the profiles in terms of age categories, sex, educational level, marital status, occupation, economic status, living arrangement, health insurance type, primary renal disease, dialysis catheter type, BMI categories, or dialysis frequency (all p > 0.05). Descriptive comparisons of demographic, clinical, and biochemical characteristics across the three latent profiles are presented in Table 2. Most sociodemographic characteristics did not differ significantly across profiles. Some clinical and laboratory indicators, including dialysis vintage, blood urea nitrogen, and hemoglobin, showed between-profile differences in the descriptive analyses. However, subsequent exploratory R3STEP analyses were used to evaluate associations with profile membership while accounting for classification uncertainty. Furthermore, the three latent profiles exhibited marked differences in health-related quality of life and chronic disease self-efficacy. In descriptive analyses, mean PCS scores were 73.29 ± 13.48 in the low symptom distress profile, 43.84 ± 19.60 in the moderate symptom distress profile, and 33.07 ± 21.54 in the high symptom distress profile. The corresponding MCS scores were 71.03 ± 10.53, 56.05 ± 15.86, and 48.09 ± 18.02, respectively (Table 2). Formal distal outcome comparisons accounting for classification uncertainty are presented in Table 3 using the BCH approach. Table 3BCH comparisons of health-related quality of life and self-efficacy across latent profilesOutcomeLow profile Mean (SE)Moderate profile Mean (SE)High profile Mean (SE)Overall pSignificant pairwise comparisonsPCS73.309 (0.997)44.307 (2.743)32.722 (2.661)< 0.001Low > Moderate, Low > High, Moderate > HighMCS71.040 (0.779)56.428 (2.219)47.833 (2.226)< 0.001Low > Moderate, Low > High, Moderate > HighCDSES32.942 (0.785)18.453 (1.498)12.172 (1.269)< 0.001Low > Moderate, Low > High, Moderate > HighDistal outcomes were compared across latent profiles using the BCH approach in Mplus, which accounts for classification uncertainty in latent profile membership. Values are BCH-estimated means and may differ slightly from the raw descriptive means shown in Table 2. The low, moderate, and high profiles correspond to the low, moderate, and high symptom distress profiles, respectively BCH comparisons of health-related quality of life and self-efficacy across latent profiles Distal outcomes were compared across latent profiles using the BCH approach in Mplus, which accounts for classification uncertainty in latent profile membership. Values are BCH-estimated means and may differ slightly from the raw descriptive means shown in Table 2. The low, moderate, and high profiles correspond to the low, moderate, and high symptom distress profiles, respectively Exploratory R3STEP analyses of factors associated with profile membership Exploratory R3STEP analyses were conducted to examine factors associated with latent profile membership while accounting for classification uncertainty. Using the low symptom distress profile as the reference category, higher phosphorus was associated with greater odds of membership in the moderate symptom distress profile (OR = 1.795, 95% CI 1.003–3.214, p = 0.049) and the high symptom distress profile (OR = 2.696, 95% CI 1.590–4.572, p = 0.020). BMI showed a borderline association with profile membership, particularly for the moderate symptom distress profile (OR = 1.085, 95% CI 1.000–1.176, p = 0.049), while the association for the high symptom distress profile was marginal (OR = 1.080, 95% CI 0.999–1.168, p = 0.052). Because the R3STEP analyses are model-based and account for classification uncertainty, these results should be interpreted as exploratory associations and may not align exactly with raw descriptive between-profile comparisons. In the exploratory R3STEP analyses, Phosphorus showed an exploratory model-based association with latent profile membership in the R3STEP analyses. However, raw descriptive phosphorus values in Table 2 did not show a statistically significant monotonic gradient across profiles, so this finding should be interpreted cautiously and as hypothesis-generating (Table 4). Table 4Exploratory R3STEP analyses of factors associated with latent profile membershipVariableModerate vs. Low OR (95% CI) p High vs. Low OR (95% CI) p Age0.986 (0.964–1.007)0.191.018 (0.995–1.040)0.125Comorbidity (Yes vs. No)0.929 (0.086–10.033)0.951.148 (0.109–12.044)0.915Dialysis vintage1.007 (0.999–1.016)0.0981.000 (0.992–1.009)0.956Albumin1.026 (0.964–1.092)0.4231.011 (0.952–1.072)0.729Hemoglobin1.007 (0.993–1.020)0.3241.010 (0.999–1.022)0.078Phosphorus1.795 (1.003–3.214)0.0492.696 (1.590–4.572)0.02log(CRP + 1)0.871 (0.697–1.089)0.1931.011 (0.814–1.255)0.924BMI1.085 (1.000–1.176)0.0491.080 (0.999–1.168)0.052Results are from exploratory three-step analyses (R3STEP) in Mplus, which account for classification uncertainty in latent profile membership. The low symptom distress profile was used as the reference category. Odds ratios greater than 1 indicate higher odds of belonging to the comparison profile relative to the low symptom distress profile Exploratory R3STEP analyses of factors associated with latent profile membership Results are from exploratory three-step analyses (R3STEP) in Mplus, which account for classification uncertainty in latent profile membership. The low symptom distress profile was used as the reference category. Odds ratios greater than 1 indicate higher odds of belonging to the comparison profile relative to the low symptom distress profile BCH analyses showed significant differences in distal outcomes across the three latent profiles. For physical health-related quality of life, PCS scores were highest in the low symptom distress profile (mean 73.309, SE 0.997), followed by the moderate symptom distress profile (mean 44.307, SE 2.743), and lowest in the high symptom distress profile (mean 32.722, SE 2.661). The overall difference was significant (χ² = 290.209, p < 0.001), and all pairwise comparisons were significant (low vs. moderate: χ² = 98.535, p < 0.001; low vs. high: χ² = 204.010, p < 0.001; moderate vs. high: χ² = 8.465, p = 0.004). For mental health-related quality of life, MCS scores were also highest in the low symptom distress profile (mean 71.040, SE 0.779), followed by the moderate symptom distress profile (mean 56.428, SE 2.219), and lowest in the high symptom distress profile (mean 47.833, SE 2.226). The overall difference was significant (χ² = 131.469, p < 0.001), and all pairwise comparisons were significant (low vs. moderate: χ² = 38.525, p < 0.001; low vs. high: χ² = 96.870, p < 0.001; moderate vs. high: χ² = 6.883, p = 0.009). Similarly, self-efficacy differed significantly across the three profiles. CDSES scores were highest in the low symptom distress profile (mean 32.942, SE 0.785), intermediate in the moderate symptom distress profile (mean 18.453, SE 1.498), and lowest in the high symptom distress profile (mean 12.172, SE 1.269), with a significant overall difference (χ² = 230.900, p < 0.001) and significant pairwise differences between all three profiles (low vs. moderate: χ² = 73.104, p < 0.001; low vs. high: χ² = 193.628, p < 0.001; moderate vs. high: χ² = 9.453, p = 0.002). Discussion This study identified three clinically interpretable symptom profiles among patients undergoing peritoneal dialysis, with a clear gradient in symptom burden and patient-reported outcomes across profiles. Fatigue, sleep disturbance, pain, and anxiety appeared to co-occur in patterned ways rather than in isolation, supporting the potential value of multidimensional symptom assessment in PD care. In this sample, the co-occurrence of fatigue, sleep disturbance, pain, and anxiety was broadly consistent with prior literature suggesting that symptom burden in dialysis populations is multidimensional rather than symptom-specific [31–33]. However, the present cross-sectional data do not establish causal, directional, or mechanistic relationships among these symptoms [34, 35]. The findings should therefore be interpreted as describing clinically meaningful symptom patterns rather than demonstrating a shared underlying biological pathway. The co-occurrence of fatigue and insomnia in the more symptomatic profiles is consistent with prior literature suggesting a bidirectional relationship between sleep disruption and daytime fatigue [36]. Likewise, the co-occurrence of pain and anxiety is consistent with previous studies suggesting psychophysiological interactions between affective distress and pain perception [37, 38]. Similar combinations of fatigue, sleep difficulties, pain, and emotional distress have also been reported in patients with cancer, heart failure, and chronic obstructive pulmonary disease [39–41]. Taken together, these findings extend prior symptom cluster research to the peritoneal dialysis population and suggest that these symptoms may represent a clinically meaningful symptom pattern. One of the most notable findings of this study was the clear gradient across the three profiles in health-related quality of life and self-efficacy. BCH analyses showed that PCS, MCS, and CDSES were highest in the low symptom distress profile, intermediate in the moderate symptom distress profile, and lowest in the high symptom distress profile, with significant differences across all pairwise comparisons. The substantial reductions in PCS and MCS in the high symptom distress profile indicate marked impairment in both physical and mental aspects of health-related quality of life. Lower self-efficacy may also have important implications for the autonomy and self-management required in home peritoneal dialysis [42]. Prior studies have linked lower self-efficacy in peritoneal dialysis to poorer treatment adherence, technical difficulties, and a higher risk of adverse PD-related outcomes [43]. Taken together, these findings suggest that greater symptom burden tends to co-occur with poorer functioning across physical, psychological, and self-management domains, rather than indicating a single causal mechanism. In the exploratory R3STEP analyses, higher phosphorus was associated with greater odds of membership in the moderate and high symptom distress profiles, compared with the low symptom distress profile, whereas BMI showed only borderline evidence of association. Because these R3STEP estimates are model-based and account for classification uncertainty, they should be interpreted as exploratory associations rather than direct descriptive contrasts between raw profile means. In contrast, age, dialysis vintage, albumin, hemoglobin, and log-transformed CRP were not robustly associated with profile membership after accounting for classification uncertainty. These findings suggest that metabolic disturbances may be relevant to symptom burden, but the biological mechanisms underlying these symptom profiles are likely multifactorial and should be interpreted cautiously. Further longitudinal studies with more comprehensive PD-specific clinical measures and biomarker data are needed. Future interventions may benefit from adopting a multidimensional strategy that addresses symptom burden, sleep, pain, emotional distress, and self-management together, but this requires confirmation in prospective and interventional studies. In routine follow-up, brief multidimensional symptom screening tools may help identify patients with higher symptom burden who could benefit from more comprehensive assessment and multidisciplinary support. Whether such strategies can reduce symptom burden and improve longer-term outcomes should be examined in future studies. Strengths and limitations To our knowledge, this is among the first studies to use a person-centered latent profile approach to characterize co-occurring fatigue, sleep disturbance, pain, and anxiety in patients undergoing peritoneal dialysis, and to compare health-related quality of life and self-efficacy across symptom profiles. A further strength is the use of BCH and R3STEP procedures, which account for classification uncertainty more appropriately than post hoc classify-analyze approaches. The retained three-profile solution also showed good class separation and clinical interpretability. Several limitations should be noted. First, the cross-sectional design precludes causal inference and does not allow examination of temporal changes or transitions between symptom profiles. Second, this was a single-center study conducted in a tertiary hospital in China, and the relatively low educational attainment and household income of the sample may limit generalizability to other PD populations. Third, all symptom and patient-reported outcome measures were self-reported and may be subject to recall and response bias; pain was assessed using a single-item NRS and did not capture pain location, chronicity, etiology, analgesic use, or pain-related interference. Fourth, 30 participants were excluded because of missing data on one or more core symptom indicators, and detailed item-level missingness patterns were not retained; comparison of included and excluded participants should therefore be regarded as only a preliminary assessment of potential selection bias. However, a basic sensitivity analysis using full-information maximum likelihood (FIML) yielded a similar 3-profile structure, which provides some reassurance regarding the robustness of the primary latent profile solution. Fifth, several clinically relevant variables were not measured, including depression, direct measures of residual kidney function in PD (such as urine volume or renal Kt/V), peritoneal transport status, ultrafiltration failure or broader fluid-related parameters, PD modality and treatment-related exposure (e.g., CAPD vs. APD, glucose exposure, icodextrin use), prior peritonitis, hospitalization history, cardiovascular disease burden, and other common PD-related symptoms such as pruritus, restless legs symptoms, dyspnea, gastrointestinal symptoms, sexual dysfunction, and cognitive complaints. Future directions Future research should prioritize multicenter longitudinal studies to examine the evolution of symptom profiles and transitions between subgroups in patients undergoing peritoneal dialysis, and to clarify temporal relationships among fatigue, sleep disturbance, pain, and anxiety. Randomized or pragmatic intervention studies targeting modifiable domains such as sleep, emotional distress, nutrition, and symptom self-management may help determine whether multidimensional interventions can reduce symptom burden or alter symptom profile membership over time. In addition, real-world translational studies are needed to evaluate the feasibility and cost-effectiveness of brief, nurse-led multidimensional symptom screening tools in peritoneal dialysis clinics, and to support integration of symptom profile-oriented, individualized, and multidisciplinary care into routine practice. Conclusion In this study, three latent symptom profiles—low, moderate, and high symptom distress—were identified among patients undergoing peritoneal dialysis. Patients in the high symptom distress profile had substantially poorer physical and mental health-related quality of life and lower self-efficacy than those in the other two profiles. Exploratory three-step analyses suggested that higher phosphorus may be associated with membership in more symptomatic profiles. These findings suggest that symptom profile-based screening may help identify peritoneal dialysis patients with greater supportive care needs and may inform future multidisciplinary management strategies, although longitudinal studies are needed before drawing causal or predictive inferences. Supplementary Information Below is the link to the electronic supplementary material. Supplementary Material 1 Supplementary Material 2 Supplementary Material 3 Acknowledgements The authors would like to thank all patients who participated in this study and the nursing staff of the Peritoneal Dialysis Centre of the First Affiliated Hospital of Shantou University Medical College for their assistance with data collection. Abbreviations Athens Insomnia Scale Sample–size–adjusted Bayesian Information Criterion Akaike Information Criterion Analysis of Variance Bayesian Information Criterion Bootstrapped Likelihood Ratio Test Body Mass Index Blood Urea Nitrogen Chronic Disease Self–Efficacy Scale C–reactive protein Haemoglobin Health–Related Quality of Life Urea clearance index Lo–Mendell–Rubin adjusted Likelihood Ratio Test Latent Profile Analysis 36–Item Short Form Health Survey Mental Component Summary (of SF–36) Physical Component Summary (of SF–36) Multidimensional Fatigue Inventory–20 Maximum Likelihood Estimation with Robust standard errors Numerical Rating Scale Peritoneal Dialysis Zung Self–Rating Anxiety Scale Serum Creatinine Standard Deviation Author contributions XS C, CN J conceptualised and designed the study, obtained ethical approval, supervised the project, and revised the manuscript critically for important intellectual content. YT Z performed data collection, conducted statistical analyses, and drafted the initial manuscript. DD Z, YZ F, MY X contributed to participant recruitment, data collection, and data entry. All authors contributed to interpretation of the data, reviewed and revised the manuscript, and approved the final version submitted for publication. Funding This study was funded by the Shantou Science and Technology Plan Project (Medical and Health Category) (Grant No. 250707166492985; Principal Investigator: Chaona Ji). The funding body played no role in the design of the study, the collection, analysis, and interpretation of data, or in the writing of the manuscript. Data availability The datasets generated and/or analysed during the current study are not publicly available due to institutional regulations on patient privacy but are available from the corresponding author on reasonable request and with permission of the First Affiliated Hospital of Shantou University Medical College. Declarations Ethics approval and consent to participate The study was conducted in accordance with the Declaration of Helsinki and was approved by the Ethics Committee of the First Affiliated Hospital of Shantou University Medical College (approval number: B-2025-013; date of approval: 13 March 2025). Written informed consent was obtained from all individual participants included in the study prior to enrolment. Participants were informed that participation was voluntary and that they could withdraw at any time without any consequence to their medical care. Consent for publication Not applicable. 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. Contributor Information Chaona Ji, Email: candykee@126.com. Xisui Chen, Email: xisuichen@163.com. References Associated Data Supplementary Materials Supplementary Material 1 Supplementary Material 2 Supplementary Material 3 Data Availability Statement The datasets generated and/or analysed during the current study are not publicly available due to institutional regulations on patient privacy but are available from the corresponding author on reasonable request and with permission of the First Affiliated Hospital of Shantou University Medical College.