Evaluating the inter-relationships among depression, anxiety and somatic symptoms in haemodialysis patients: a cross-sectional structural equation modelling study Authors: Anika Krishna, Kari O’Donnell, Jacqueline Dolata, Maria Figueroa, Martha Sajatovic, Ashwini R Sehgal, Douglas D Gunzler Published: 2026 ABSTRACT AbstractBackgroundSomatic symptoms frequently occur in haemodialysis patients, whereas depression and anxiety are also common but are often under-recognised issues. This study aims to examine the associations among depression, anxiety and somatic symptoms in haemodialysis patients. These symptoms may be linked together under a general distress construct in haemodialysis patients.MethodsWe surveyed 1085 dialysis patients between 1 March 2018 and 28 February 2023 at 17 dialysis facilities in Northeast Ohio with the 9-item Patient Health Questionnaire (PHQ-9), 7-item Generalized Anxiety Disorder and Kidney Disease Quality of Life instrument somatic questions. To evaluate associations across these multi-item questionnaires, we used latent variables and a structural equation modelling approach. We first compared single-factor, two-factor and three-factor models. We next considered bifactor and higher order factor models. We controlled for race, gender and age in all models.ResultsOur study sample included a high percentage of black patients (75%) and 150 (14.3%) survey participants screened positive for depression (PHQ-9 ≥10). All models fit relatively well, with the three-factor model fitting slightly better than the single and two-factor models. In the three-factor model, the associations for depression and anxiety (r=0.818), depression and somatic (r=0.800) and anxiety and somatic (r=0.784) were all high, potentially indicative of poor discriminant validity. We found that the bifactor model fit the best of the five models tested (comparative fit index=0.957, Tucker-Lewis fit index=0.959, root mean square error of approximation=0.043) and was the most substantively meaningful, with an overall internalising construct and specific domains explaining additional covariance across the questions. Being black (vs all other races), male (vs female) and older age all had significantly (p<0.05) lower levels of reported symptoms on most constructs.ConclusionWe found very strong associations among depression, anxiety and somatic symptoms in our sample of haemodialysis patients. All questionnaire items may be linked together under a common general distress disorder, while each still retains unique features of depression, anxiety and somatic symptoms not fully explained by internalising processes. Patient care approaches should generally consider screening for all three together, rather than only one. STRENGTHS AND LIMITATIONS OF THIS STUDY This study used a large, multisite sample of haemodialysis patients and incorporated the 9-item Patient Health Questionnaire and Kidney Disease Quality of Life questionnaires. This study used a large, multisite sample of haemodialysis patients and incorporated the 9-item Patient Health Questionnaire and Kidney Disease Quality of Life questionnaires. The use of structural equation modelling with latent variables allowed for explicit modelling of measurement error and underlying constructs of depression, anxiety and somatic. The use of structural equation modelling with latent variables allowed for explicit modelling of measurement error and underlying constructs of depression, anxiety and somatic. Ordinal questionnaire data were modelled to evaluate a general distress construct and the inter-relationships among depression, anxiety and somatic symptoms using a series of confirmatory factor analysis models. Ordinal questionnaire data were modelled to evaluate a general distress construct and the inter-relationships among depression, anxiety and somatic symptoms using a series of confirmatory factor analysis models. The absence of socioeconomic and educational variables introduces potential residual confounding. The absence of socioeconomic and educational variables introduces potential residual confounding. The use of modified symptom scales, which are not formally validated, limits interpretability and comparison to other studies. The use of modified symptom scales, which are not formally validated, limits interpretability and comparison to other studies. Introduction Depression, anxiety and somatic symptoms are common and debilitating among individuals undergoing haemodialysis. These psychological and physical burdens significantly impact patients’ quality of life and treatment outcomes.1 Major depressive disorder (MDD) is the most frequently diagnosed psychiatric condition in such individuals.2 Studies estimate that around 20–30% of dialysis patients are affected by MDD.36 This condition is linked to several adverse health outcomes, such as reduced treatment adherence, poor nutritional status, higher hospitalisation rates and increased risk of early mortality.3710 Additionally, the presence of depression in patients with chronic kidney disease (CKD) is known to significantly lower their quality of life.11 Anxiety is another common psychological symptom in dialysis patients, with prevalence estimates ranging from 20% to 52%, depending on the assessment tools and population studied.1214 It is associated with poorer adherence to treatment, reduced quality of life and increased morbidity and mortality in this population.1 14 Despite their high prevalence and clinical impact, both anxiety and depression are often under-recognised and undertreated in dialysis settings.1 15 In addition, anxiety and depression have been found to be highly correlated in dialysis patients, often co-occurring and reinforcing each other. Research consistently shows moderate to strong correlations, with correlation coefficients (r) typically ranging from 0.60 to 0.80, indicating a substantial overlap in symptoms and psychological burden.1 14 16 Somatic symptoms are more generally recognised in dialysis patients and include fatigue, muscle cramps, itching, pain and gastrointestinal discomfort.1719 These symptoms are often persistent and impact patients’ quality of life.18 19 Importantly, these physical complaints may arise from both medical and psychological origins. For example, a study of haemodialysis patients from Perales-Montilla et al found that emotional factors, particularly anxiety and depression, significantly predicted somatic symptoms such as cardiovascular and gastrointestinal issues.20 The authors state that these findings suggest psychological-emotional variables can help explain somatic symptoms that do not have a clear aetiology. A similar study from Weisbord et al21 also documented the high prevalence and severity of both physical and emotional symptoms in dialysis patients, highlighting the need for integrated approaches to symptom management. Despite the clinical implications of these findings, the links between somatic and psychological symptoms remain insufficiently understood in this population. The primary goal of this study is to gain a better understanding of the relationship between anxiety, depression and somatic symptoms in patients undergoing haemodialysis. While anxiety and depression have been studied frequently together as complementary mood disorders,22 less is known about how somatic symptoms fit into this framework. In primary care populations, Simms et al23 found that a bifactor model best explained the co-occurrence of the three domains. Their work mentions that in primary care settings, anxiety and depression were more prevalent compared with somatic symptoms. In contrast, somatic symptoms are especially prominent in patients undergoing haemodialysis.24 Additionally, patients receiving haemodialysis report notably high levels of depression, anxiety and somatic symptoms.1 This raises the question of whether a similar bifactor structure observed in primary care applies to a haemodialysis population. We thus, similarly to Simms et al,23 compare latent variable models in our study population using structural equation modelling (SEM)2528 to determine the underlying dimensionality and structure of symptom co-occurrence among anxiety, depression and somatic symptoms. Our hypothesis is the same: there is an overall internalising construct that links together anxiety, depression and somatic symptoms; however, residual covariance in the form of anxiety, depression and somatic symptoms still persists. In addition, by modelling these relationships concurrently, the research seeks to provide a more comprehensive understanding of the psychological and physical symptom burden in this population. Methods Setting and participants The study was conducted at 17 in-centre haemodialysis facilities that are part of the Centers for Dialysis Care (CDC) system in Northeast Ohio. Patients were identified from these facilities using our inclusion/exclusion criteria described below from the CDC Clarity database. All adult, English-speaking, cognitively intact patients who had been on maintenance haemodialysis for at least 3 months were invited to participate in the study. The median facility size is 92; 47% of patients are female, 48% are African-American and 3% are Hispanic. These facility and patient characteristics are roughly comparable to national figures, except that there are more African-American and fewer Hispanic dialysis patients in the Cleveland area compared with the United States as a whole. Children aged <18 were excluded from the study because the performance of the 9-item Patient Health Questionnaire (PHQ-9) and the Diagnostic and Statistical Manual of Mental Disorders, Fifth Edition criteria for depression are slightly different among children compared with adults.29 30 We excluded new patients because the first 3 months of dialysis treatment are often a time of complications. Finally, cognitively impaired patients were excluded from the study because they would be unable to reliably report on depressive symptoms and dialysis complications.31 We did not remove patients on antidepressants, as we needed to observe the full range of depression levels for our algorithm. In addition, standard practice is to administer PHQ-9 to all patients, regardless of whether they are on antidepressants or not. The total number of patients screened by chart was n=2357. Of these, n=1272 were excluded (27 non-English speaking, 43 deceased, 756 refused and 446 could not be contacted by our team). We collected data on 1085 study participants between 1 March 2018 and 28 February 2023. Pre-COVID-19 pandemic (1 March 2020), the quality of life and depression screening surveys under study (see the Measures assessed section) were applied and taken in person during the dialysis sessions. The surveys were verbally administered by a study coordinator. Post-COVID-19 pandemic, letters were sent out to all eligible patients at the 17 in-centre haemodialysis facilities introducing the study. Surveys were then administered via a phone call during or after dialysis sessions. All participants provided written informed consent. Patient and public involvement Patients or the public were not involved in the design, conduct, reporting or dissemination plans of our research. Measures assessed The PHQ-9 is a brief, self-report depression severity measure.32 Scores on this self-reported instrument are often used to guide treatment decisions.32 A PHQ-9 ≥10 has been previously established as a screening cut-off for depressive disorder. The Kidney Disease Quality of Life (KDQOL) survey is a kidney disease-specific measure of health-related quality of life.33 The long form of the KDQOL (134 items) was developed in 1994 by the Kidney Disease Quality of Life Working Group with support from Amgen. In our study, we used 14 specific items chosen by our team of experts from the long form of the KDQOL to comprehensively describe the somatic symptoms of dialysis. The specific questions are detailed in online supplemental table 1. The 7-item Generalized Anxiety Disorder-7 (GAD-7)34 is a widely used self-report scale designed to assess the severity of generalised anxiety symptoms over the past 2 weeks. The GAD-7 has demonstrated strong reliability and validity in various populations, making it a valuable tool for screening and monitoring anxiety symptoms in clinical settings.34 To ensure more consistency with the KDQOL, our teams administered the GAD-7 items as part of a 56-item questionnaire with KDQOL and used five categories instead of the usual three for the GAD-7 for individual items. Therefore, total scoring on our modified GAD-7 will be higher than established cut-offs and not directly comparable. Demographic variables incorporated in our analysis plan included age, race (white, black, other) and gender. Statistical analyses We reported descriptive statistics as mean (SD) for continuous measures and number of subjects (percentage) for each category for discrete measures to summarise demographic information and scores on our measures of depression screening, anxiety and the KDQOL somatic items. SEM is a flexible multivariate technique that allows relationships among latent variables and observed variables to be examined. A latent variable is a variable that is assumed to be not directly observed and instead is inferred using observed variables. For example, as a depression screening tool, the nine items of the PHQ-9 can be used to infer a latent variable of depression, since these items are assumed to describe different symptoms of depression and in aggregate depression. A similar latent variable of anxiety and somatic can be inferred in our study using the GAD-7 and KDQOL somatic items, respectively. Note that we italicise depression, anxiety and somatic here (and throughout this paper) to denote we are referring to a latent variable rather than the condition. A latent variable model, also termed a confirmatory factor analysis (CFA) model, relates a latent variable to the observed variables postulated to infer the latent variable. In such a model, random measurement error represents the difference between what is assumed to be the true score, the latent variable and the observed variables. One does not have to assume a single latent variable, as sometimes multiple, correlated latent variables are inferred for a given set of observed variables. In our sample, we conducted CFA using ordered categorical outcomes to model latent constructs of depression, anxiety and somatic symptoms. The observed indicators (items from the PHQ-9, GAD-7 and KDQOL) were right skewed and ordinal, with a limited number of discrete categories. Therefore, we treated these variables as ordered categorical rather than continuous. In this modelling framework, each categorical response is assumed to reflect an underlying, normally distributed latent response variate, which is continuous but unobserved.25 35 We denote this underlying normally distributed latent response variate with an asterisk, as is common nomenclature in the SEM literature.27 Responses are determined by where an individual’s value on this latent response falls in relation to a set of item-specific threshold.27 For instance, the PHQ-9 item assessing anhedonia includes four observed categories: ‘not at all’, ‘several days’, ‘more than half the days’ and ‘nearly every day’. These correspond to three thresholds along the latent continuum of anhedonia*. If an individual’s latent score is below the first threshold, they will respond ‘not at all’. As the latent score crosses successive thresholds, the response shifts into higher categories. In this way, the observed ordinal response is a coarse indicator of the continuous latent severity of the symptom. This thresholding process is repeated for each item in the model. All models included regressions of each latent construct on covariates for age, gender (male, female) and race (black, white) to evaluate differences in symptom severity across demographic groups. By treating the indicators as ordinal and modelling the measurement process explicitly, this approach preserves the categorical nature of the data while capturing the underlying symptom dimensions with greater accuracy. To test our primary hypothesis—that depression, anxiety and somatic symptoms may not be meaningfully distinct but rather indicators of a single underlying internalising/general distress factor—we compared five latent variable models. First, we estimated a single-factor model (figure 1) in which all items loaded onto one general internalising factor, representing the shared variance across depression, anxiety and somatic symptoms. Second, we assessed a two-factor model (figure 2), where depression and anxiety items loaded onto a combined factor, and somatic symptom items loaded onto a separate somatic factor. This specification reflects prior literature in which depression and anxiety are seen as strongly associated mood disorders.22 In this model, the depression/anxiety factor is allowed to correlate with the somatic factor, as indicated by the bidirectional path labelled ρ in the diagram. This correlation (ρ) represents the degree to which general emotional distress and somatic complaints co-occur, capturing their shared variance while still modelling them as distinct constructs. Third, we estimated a three-factor model (figure 3), where items were allowed to load only onto their respective domains (depression, anxiety, somatic), and the three latent factors were permitted to correlate. In this model, the correlations among the three latent variables are represented as r1, r2 and r3. These correlations allow us to estimate the degree of association between each pair of constructs while still modelling them as distinct factors. Fourth, we tested a higher order model (figure 4), where depression, anxiety and somatic symptoms were modelled as separate first-order factors that loaded onto a second-order internalising factor. This structure assumes that general distress influences symptoms indirectly through the specific domains. Unlike a bifactor model, the higher order model attributes shared variance solely to the second-order factor, with no direct loadings from general distress to individual items.27 Fifth, we estimated a bifactor model (figure 5), in which all items simultaneously loaded onto both a general factor and a symptom-specific factor (depression, anxiety or somatic). In the bifactor model, none of the latent factors are allowed to correlate with each other, which allows for the unique contribution of each factor to be interpreted independently. This specification enables unambiguous interpretation of associations with the general factor, as they are not confounded by correlations with the specific symptom domains. Importantly, while the general factor captures shared variance among all symptoms, it does not account for all the covariance; the specific factors explain additional variance unique to each domain, improving construct clarity. Single-factor model. Two-factor model. Three-factor model. Higher order model. Bifactor model. In order to test the overall plausibility of our hypothesised models and make comparisons across them, we evaluated the model fit using the χ2, comparative fit index (CFI), Tucker-Lewis fit index (TLI), root mean square error of approximation (RMSEA) and standardised root mean square residual (SRMR) statistical test values.3639 A non-significant χ2 value, CFI and TLI ≥0.95, RMSEA value of ≤0.05 and SRMR value ≤0.05 represent an excellent fitting model, while a smaller χ2 test statistic, CFI and TLI ≥0.90, RMSEA value of ≤0.08 and SRMR value ≤0.08 represent an acceptable fitting model.27 RMSEA, CFI, TLI and SRMR values should all reach acceptable (preferably excellent) levels before designating a model as good fitting.27 The RMSEA statistic is especially useful in that it provides a CI for our point estimate. We used a mean and variance-adjusted weighted least squares estimator (option in lavaan) in our models. Note that there was a small amount of missing data (3.4%), which was assumed to be missing completely at random and thus handled via listwise deletion. Given that only 3.4% of the data were missing, we chose listwise deletion on the basis that the extent of missingness was minimal and unlikely to materially affect the results. We also considered that more complex imputation approaches under such low levels of missingness could introduce unnecessary modelling assumptions and potential bias. We define α=0.05 for our level of significance in all statistical tests. All statistical tests are two-tailed. SEM was carried out using lavaan in R program. R program in the RStudio environment was also used for data management, graphical displays and other descriptive analyses. Results Descriptive analyses The study sample was 48% female, 75% black and had an average age of 66 years (SD=14), as shown in table 1. Online supplemental table 1 additionally shows all the scores on each of the categories for each questionnaire (PHQ-9, GAD-7, KDQOL somatic questions) under study. Similar to national reports, men outnumber women, and the study sample is on average an older age.40 41 The percentage of black patients available in this study sample (of our overall sample: 39% black male, 36% black female) was much higher than in other large studies, for example, Myers et al40 and Kucirka et al.42 More generally, about one in three people receiving dialysis are black.43 Table 1Descriptive summaryVariables/questionsn=1048*Gender Female500 (48%) Male548 (52%)Race Black784 (75%) White256 (24%) Other8 (<1%)Ethnicity Non-Hispanic1031 (98%) Hispanic or Latino16 (<2%) Declined1 (<1%)Age66 (14)PHQ-9 total4.5 (4.6)GAD-7† total (max score=35)10.6 (5.1)KDQOL† somatic questions total (max score=72)24 (7)*n (%); mean (SD).†Modified.GAD-7, 7-item Generalized Anxiety Disorder; KDQOL, Kidney Disease Quality of Life; PHQ-9, 9-item Patient Health Questionnaire. Descriptive summary n (%); mean (SD). Modified. GAD-7, 7-item Generalized Anxiety Disorder; KDQOL, Kidney Disease Quality of Life; PHQ-9, 9-item Patient Health Questionnaire. At their entry, 14% (n=150) of patients had PHQ-9 ≥10. The distribution of PHQ-9 scores represents a wide range of depression screening severity levels. However, the prevalence of MDD among dialysis patients has been estimated at approximately 20–30%.36 The percentage of patients screening positive for depression via the PHQ-9 in haemodialysis patients has been found to be similarly high. For example, Agrawal et al44 found the prevalence to be 37% in a study sample of 100 with stage 5 CKD at a tertiary care centre, and Anderson et al found the prevalence to be 26% in a study sample of 485.45 We evaluated the correlation across total aggregate scores of the PHQ-9, GAD-7 and the KDQOL somatic items. We found moderate to high Pearson’s correlations: 0.68 (95% CI 0.64 to 0.71) between PHQ-9 and GAD-7 total scores, 0.64 (95% CI 0.60 to 0.67) between PHQ-9 and KDQOL somatic total scores and 0.66 (95% CI 0.63 to 0.70) between GAD-7 and KDQOL somatic total scores. While the PHQ-9 is typically scored and interpreted using established clinical thresholds, the GAD-7 was modified in our dataset to include additional response categories, so the total score cannot be interpreted using standard cut-points. Similarly, the KDQOL somatic symptom total score does not have established clinical thresholds and is not intended to function as a standalone severity index. To assess the inter-relationships among individual symptom items and evaluate the degree of shared variance across depression, anxiety and somatic symptoms, we computed a polychoric correlation matrix for the 30 ordinal items from the PHQ-9, GAD-7 and KDQOL somatic symptom scales. Within-construct correlations were strong, particularly among PHQ-9 items (r=0.45–0.71) and GAD-7 items (r=0.58–0.75), reflecting high internal consistency. Somatic items demonstrated a wider range of correlations (r=0.15–0.65). Statistical analysis We evaluated five structural models to determine the latent structure underlying depression, anxiety and somatic symptoms (table 2). Table 2Comparison of model fit indicesModelΧ2dfP valueCFITLIRMSEARMSEA 90% CISRMROne-factor2359492<0.0010.9080.9190.060(0.055 to 0.06)0.078Two-factor1592427<0.0010.9370.9440.051(0.044 to 0.05)0.069Three-factor1740483<0.0010.9380.9440.050(0.043 to 0.048)0.066Higher order1307422<0.0010.9520.9570.045(0.037 to 0.043)0.062Bifactor1317453<0.0010.9570.9590.043(0.032 to 0.037)0.054CFI, comparative fit index; RMSEA, root mean square error of approximation; SRMR, standardised root mean square residual; TLI, Tucker-Lewis fit index. Comparison of model fit indices CFI, comparative fit index; RMSEA, root mean square error of approximation; SRMR, standardised root mean square residual; TLI, Tucker-Lewis fit index. The one-factor model demonstrated acceptable fit to the data (CFI=0.908, TLI=0.919, RMSEA=0.060), suggesting a unidimensional structure was plausible. The two-factor model showed improved fit (CFI=0.937, TLI=0.944, RMSEA=0.051), indicating a potential distinction between emotional and physical symptoms, with a high correlation between factors (r=0.741). The three-factor model, which specified separate but correlated depression, anxiety and somatic constructs, fit similarly (CFI=0.938, TLI=0.944, RMSEA=0.050). From a theoretical perspective, the three-factor model was more clinically meaningful than the two-factor model, as it did not group anxiety and depression as a single construct; however, the correlations among the three latent factors were strong—r1=0.818 (depression–anxiety), r2=0.784 (anxiety–somatic) and r3=0.800 (depression–somatic)—indicating substantial overlap and potentially suggesting poor discriminant validity.27 Thus, we tested a higher order model, which demonstrated strong fit (CFI=0.952, TLI=0.957, RMSEA=0.045), but had some limitations in regard to interpretability because covariates could only be regressed onto the second-order factor (not on both the general and specific factors) to meet the condition of model identification (ie, to ensure a unique solution).27 Finally, we tested a bifactor model, which provided the best fit to the data among the five models (CFI=0.957, TLI=0.959, RMSEA=0.043) and offered a conceptually coherent representation of symptom structure from a clinical perspective. In the bifactor model, all items except one loaded highly (≥0.40) on the general factor with standardised loadings as high as 0.799. The item with the lowest loading was Q29 (bowel problems, λ=0.368). Additionally, eight items loaded strongly (>0.40) on their specific factors, of which two were on anxiety, two on depression and four on somatic. Overall, the pattern of loadings (table 3) supports the utility of modelling both general and specific factors to explain item covariation, justifying the bifactor model as the optimal model for this study. Table 3Standardised loadings and residual variances from bifactor modelItemInternalisingDepressionAnxietySomatisationResidual varianceG1: anxious0.741 0.282 0.364G2: control worrying0.744 0.534 0.146G3: different worrying0.697 0.598 0.141G4: relaxing0.769 0.253 0.337G5: restless0.734 0.250 0.392G6: irritable0.723 0.266 0.399G7: afraid0.735 0.302 0.360P1: anhedonia0.6920.281 0.443P2: depressed0.7990.382 0.216P3: sleep0.6230.198 0.574P4: fatigue0.6560.224 0.520P5: appetite0.6180.229 0.566P6: guilt0.6900.436 0.334P7: concentration0.6760.276 0.468P8: psychomotor0.6300.236 0.547P9: self-harm0.5710.498 0.427Q9: sleeping0.714 −0.0670.488Q10: energy0.612 0.0250.624Q18: chest0.402 0.5150.564Q22: breath0.413 0.3400.708Q23: fainting0.411 0.2910.741Q24: dizziness0.562 0.2970.589Q27: indigestion0.481 0.5220.486Q29: bowel0.368 0.2190.813Q32: joints0.552 0.3290.579Q36: headaches0.429 0.4000.649Q38: stomach0.413 0.6260.426Q39: back0.486 0.3430.639Q40: heart0.425 0.4190.636Q55: sexual0.552 −0.0570.694P1–P9=PHQ-9; G1–G7=GAD-7; Q9–Q55=KDQOL.GAD-7, 7-item Generalized Anxiety Disorder; KDQOL, Kidney Disease Quality of Life; PHQ-9, 9-item Patient Health Questionnaire. Standardised loadings and residual variances from bifactor model P1–P9=PHQ-9; G1–G7=GAD-7; Q9–Q55=KDQOL. GAD-7, 7-item Generalized Anxiety Disorder; KDQOL, Kidney Disease Quality of Life; PHQ-9, 9-item Patient Health Questionnaire. We also examined how demographic variables were associated with the general and specific factors in the bifactor model (table 4). Black participants had lower scores on the internalising factor (β (standardised regression estimate)=−0.178); female participants had higher scores on depression (β=0.138), anxiety (β=0.228) and somatic (β=0.244) factors; and higher age was associated with lower general distress (β=−0.137), anxiety (β=−0.231) and somatisation (β=−0.103). The remainder of the coefficients were marginal or not significant. Table 4Standardised regression coefficients for covariates from bifactor modelFactorBlack (P value)Female (P value)Age (P value)Internalising−0.178 (<0.001)0.059 (0.111)−0.137 (<0.001)Depression0.083 (0.213)0.138 (0.048)−0.049 (0.489)Anxiety0.114 (0.056)0.228 (<0.001)−0.231 (<0.001)Somatisation0.012 (0.799)0.244 (<0.001)−0.103 (0.028) Standardised regression coefficients for covariates from bifactor model Discussion This study investigated the underlying relationships between depression, anxiety and somatic symptoms among individuals undergoing haemodialysis. The bifactor model was chosen as our optimal model. The model indicates that although these domains are highly inter-related under a general construct, each retains unique features not fully explained by internalising processes. The bifactor model also enhances clinical interpretability by allowing practitioners to determine whether elevated scores reflect broad internalising distress, specific symptom patterns or both. These findings are consistent with previous research in primary care populations, where bifactor models have also outperformed similar competing models for depression, anxiety and somatic symptoms: one-factor, two-factor, three-factor and higher order factor models (eg,23). However, our study extends this work by demonstrating that the bifactor structure holds in a haemodialysis population, who report particularly high levels of depression, anxiety and somatic burden,1 19 and where somatic symptoms are especially common (compared with primary care settings). We also examined how demographic variables were associated with variation in both the general and specific factors. Black participants had significantly lower scores on the general internalising factor, which aligns with evidence of underdiagnosis and under-reporting of symptoms in black communities.46 47 Female participants showed significantly higher scores on the depression, anxiety and somatic symptom factors, but not on the general internalising factor, suggesting that women may experience and report these symptoms in more domain-specific ways rather than as a reflection of generalised distress. Older age was associated with lower general distress, anxiety and somatic complaints. Some studies suggest that older adults experience improved emotional regulation and desensitisation to chronic symptoms.48 49 While our study benefited from a large sample, limitations must be acknowledged. First, the sample was drawn exclusively from a single dialysis system in Northeast Ohio, which limits generalisability. Data collection methods also differed before (n=912) and during (n=173) the COVID-19 pandemic, which may have introduced variation in how patients responded. Importantly, demographic variables did not include socioeconomic status or educational attainment, which are established determinants of mental health. As a result, observed associations related to race and gender may be partially confounded by unmeasured socioeconomic differences rather than reflecting independent effects of these variables. Additionally, because socioeconomic and educational characteristics in our sample may differ from those in other dialysis populations, the external validity of our findings may be constrained. Future studies should incorporate measures of socioeconomic context to provide a more nuanced understanding of symptom burden. Another limitation is that our modified version of the GAD-7 prevents direct comparability with findings from other studies. The decision to align the GAD-7 with the KDQOL items (1–5 scale) to reduce respondent burden and improve consistency across instruments was made to meet the aims of the original research project for which these data were collected. We recognise that this resulted in a non-validated modification of the original GAD-7 and precludes formal assessment of reliability and validity of the adapted scale. Additionally, somatic symptoms were derived from KDQOL items rather than a formally validated scale, which may have influenced model fit and factor loadings.27 Future research should incorporate an established somatic symptom measure, such as the Somatic Symptom Scale-8, to improve validity and generalisability. Other types of depression and anxiety rating scales (eg, Hamilton, Montgomery-Asberg) could also provide a broader assessment of symptoms. These findings may have important clinical implications. The strong overlap among depression, anxiety and somatic symptoms supports the use of integrated screening approaches that assess all three domains concurrently. The prominence of a general internalising factor suggests that much of this symptom burden may reflect a broader construct of distress. As a result, screening strategies that rely on a single domain may underestimate overall distress. In high-burden dialysis settings, the use of a composite screening approach may offer a more comprehensive method for identifying patients in need of further evaluation. However, the presence of domain-specific variance indicates that symptom-specific tools remain important for guiding targeted interventions once distress is identified. A stepped approach beginning with broad distress screening, followed by more detailed assessment of depression, anxiety and somatic symptoms, may be particularly useful. Clinicians should therefore consider both the shared and unique features of emotional and somatic symptoms when developing care plans for patients receiving dialysis. Supplementary material