Development and Preliminary Psychometric Evaluation of the Intradialytic Hypotension–Related Health Beliefs Scale for Patients Undergoing Maintenance Hemodialysis Authors: Ting Luo, Guangyi Yang, Siying Li, Zhenye Mu, Hongdan Wang, Min Yang Published: 2026 ABSTRACT AbstractPurposeTo develop the Intradialytic Hypotension (IDH)–Related Health Beliefs Scale for patients undergoing maintenance hemodialysis (MHD) and preliminarily evaluate its psychometric properties.MethodsA multistage scale-development study combined literature review, patient interviews, Delphi consultation, and pretesting. Convenience sampling recruited 451 patients from four tertiary hospitals in China. Before item analysis, participants were randomly divided into development (n=226) and validation (n=225) samples. The development sample was used for item analysis and exploratory factor analysis (EFA), whereas the validation sample was used for confirmatory factor analysis (CFA) and evaluation of convergent and discriminant validity. Content validity, known-groups validity, internal consistency, test–retest reliability, and score distributions were also examined.ResultsThe final scale contained 29 items across four domains: perceived susceptibility, perceived severity, perceived benefits, and self-efficacy. EFA supported a four-factor structure explaining 65.29% of item variance. After two content-justified post hoc within-domain residual correlations were specified, CFA fit was acceptable (CFI=0.978, TLI=0.976, RMSEA=0.064, SRMR=0.061). The scale-level content validity index using the average method (S-CVI/Ave) was 0.969. Across domains, Cronbach’s α ranged from 0.794 to 0.922, McDonald’s ω from 0.805 to 0.961, and 2-week intraclass correlation coefficients from 0.839 to 0.894. Three of six prespecified known-groups hypotheses were supported. The self-efficacy domain showed a 34.37% ceiling effect.ConclusionTo our knowledge, this is the first patient-reported instrument developed specifically to assess IDH-related health beliefs in this population. The findings provide preliminary evidence for its four-domain structure, reliability, and validity. Scores should be calculated and interpreted by domain. The scale may help identify priorities for patient education and risk communication but cannot replace objective clinical assessment. Independent multicenter and longitudinal validation is required before routine use.Keywords: intradialytic hypotension, maintenance hemodialysis, health beliefs, scale development, psychometric evaluation, patient-reported instrument Introduction Maintenance hemodialysis (MHD) is a major kidney replacement therapy for patients with end-stage kidney disease. By the end of 2022, 850,809 patients in China were receiving MHD.1 Intradialytic hypotension (IDH) is a common acute complication, although reported incidence varies by diagnostic definition.2,3 IDH may cause dizziness, fatigue, nausea, and muscle cramps. Management may require reducing or discontinuing ultrafiltration, administering fluids, or terminating dialysis early, potentially compromising volume control and dialysis adequacy.4–6 Recurrent IDH is also associated with myocardial stunning, cerebral ischemia, loss of residual kidney function, vascular access thrombosis, cardiovascular events, and mortality.3,7 Thus, IDH affects dialysis safety, treatment quality, and long-term outcomes. IDH prevention and management rely on objective monitoring of intradialytic blood pressure, symptoms, ultrafiltration, and volume status. Strategies include optimizing dry weight, controlling ultrafiltration rate, individualizing dialysis prescriptions, adjusting antihypertensive therapy, and using pharmacologic interventions when needed.4,6,8 However, several measures require sustained patient participation, including restricting fluid and sodium intake, controlling interdialytic weight gain, avoiding peridialytic food intake that may worsen hemodynamic instability, and promptly reporting symptoms.9–11 Objective monitoring and professional management therefore remain essential but cannot replace patient involvement. Patients’ appraisals of health threats, preventive actions, and self-management capability may influence health behaviors and management decisions.12 The Health Belief Model (HBM) is an important theoretical framework for explaining preventive health behaviors. According to the model, health actions are influenced by perceived susceptibility and severity, perceived benefits and barriers, and confidence in one’s ability to perform the behavior.13,14 The HBM therefore provides a theoretical basis for understanding the initiation and maintenance of preventive behaviors. Within the HBM, this study defined IDH-related health beliefs as multidimensional cognitions concerning IDH and its prevention and management. The resulting scale focused on four domains: perceived susceptibility, perceived severity, perceived benefits, and self-efficacy. The first two reflect perceived threat, whereas perceived benefits and self-efficacy reflect the value assigned to preventive behaviors and confidence in performing them. Patients who underestimate IDH risk or consequences may give less priority to controlling interdialytic weight gain or reporting early symptoms. Even when risk is recognized, low perceived benefits or low self-efficacy may impede sustained fluid management and symptom reporting. Objective clinical data are fundamental for identifying IDH and evaluating physiologic risk but do not capture patients’ health beliefs about prevention and management. Among the patient-reported instruments reviewed, existing measures primarily addressed broader illness perceptions or hemodialysis self-management. The Brief Illness Perception Questionnaire (Brief IPQ) evaluates cognitive and emotional representations of illness.15 The Hemodialysis Self-Management Instrument (HDSMI) assesses partnership, self-care, problem-solving, and emotional management.16 Neither instrument is IDH-specific, and their measurement targets differ from the HBM-based IDH-related health beliefs assessed in the present study. An IDH-specific patient-reported instrument is therefore needed to complement objective clinical data and existing general instruments by assessing IDH-related patient cognitions. To our knowledge, based on the literature searches conducted for this study, no HBM-based patient-reported instrument specifically assessing IDH-related health beliefs among patients undergoing MHD was identified. This study therefore aimed to develop the IDH-Related Health Beliefs Scale, a patient-reported and IDH-specific instrument grounded in the HBM. Its psychometric properties were then preliminarily evaluated. The scale was designed to characterize domain-specific beliefs and identify priorities for patient education and risk communication. After further feasibility testing, domain-level assessment could potentially be incorporated before patient education or nursing communication to help identify belief areas warranting further discussion alongside objective clinical information. Materials and Methods Study Design A multistage design was used to develop the scale and conduct a preliminary psychometric evaluation within the HBM framework. The study comprised item generation, item refinement, and psychometric evaluation (Figure 1). Before item analysis, the analytic sample was randomly divided into development and validation subsamples. The development sample was used for item analysis and EFA, whereas the validation sample was used for CFA and CFA-based evaluation of convergent and discriminant validity.17,18 Development and psychometric evaluation process of the IDH-Related Health Beliefs Scale. Abbreviations: EFA, exploratory factor analysis; HBM, Health Belief Model; IDH, intradialytic hypotension; WLSMV, weighted least squares mean- and variance-adjusted estimation. Stage I: Item Generation Literature Review PubMed, Web of Science, Embase, CINAHL, China National Knowledge Infrastructure, and Wanfang Data were searched from inception to April 2025, with an updated search before the first Delphi round in September 2025. Search concepts included hemodialysis, MHD, IDH, risk perception, health beliefs, the HBM, and patient-reported instruments; corresponding Chinese terms were used in Chinese databases. Relevant guidelines, theoretical and empirical studies, existing instruments, and reference lists informed candidate domains and items. Sources and preliminary theoretical assignments are provided in Supplementary Table S1. Semi-Structured Interviews A semi-structured interview guide was developed from the literature review, the HBM, and research-team discussions, with the HBM also informing the initial theoretical categorization of interview content (Supplementary Appendix 1). Maximum-variation sampling recruited 15 patients undergoing MHD from one tertiary hospital in China, considering age, sex, educational level, dialysis duration, and previous IDH experience. With participants’ permission, all interviews were audio-recorded and transcribed verbatim. Directed content analysis was used.19 Two researchers independently coded transcripts and consolidated similar codes; disagreements were resolved by team discussion. No new codes emerged after interview 13, and two additional interviews confirmed code saturation. Literature evidence, interview findings, and team discussion informed the initial item pool. Codes, theoretical assignments, and candidate-item links are presented in Supplementary Table S2. Stage II: Item Refinement Delphi Expert Consultation Two Delphi rounds were conducted from September to October 2025. Forty-two experts were invited to round 1, and the 38 respondents were invited to round 2. Eligibility required at least 10 years of experience in psychiatric nursing, nephrology, clinical nursing, hemodialysis nursing, or scale development; a bachelor’s degree or higher; and an intermediate or senior professional title. Participation was voluntary. Experts rated item relevance, importance, clarity, and theoretical assignment and suggested revisions. Outcomes included response rate, expert authority coefficient (Cr), Kendall’s coefficient of concordance (W), mean importance rating, and coefficient of variation (CV). Items with mean importance <4.00 or CV >0.25 underwent focused review. After both rounds, 10 Delphi panelists rated the relevance of the 32-item preliminary scale on a 4-point scale. Item-level content validity index (I-CVI) and S-CVI/Ave were calculated; I-CVI ≥0.78 and S-CVI/Ave ≥0.80 were considered acceptable.20–22 The complete item-refinement process is shown in Supplementary Table S3. Pretesting Convenience sampling recruited 30 patients undergoing MHD to pretest the 32-item preliminary scale.23 Standardized procedures matched those planned for the formal survey. Completion time, unclear items, and participant suggestions were recorded to assess readability, comprehensibility, and feasibility. Formal cognitive interviews were not conducted. Stage III: Psychometric Evaluation Participants and Settings A cross-sectional survey was conducted from October 2025 to February 2026. Convenience sampling recruited patients receiving MHD from hemodialysis centers at four tertiary hospitals in Shandong and Sichuan Provinces, China. The inclusion criteria were as follows: (1) meeting the diagnostic criteria for end-stage kidney disease and receiving MHD as the main treatment modality; (2) age ≥18 years; (3) dialysis duration ≥3 months and dialysis frequency ≥2 sessions per week; and (4) providing informed consent and voluntarily participating in the study. The exclusion criteria were as follows: (1) severe complications, such as infection or acute heart failure; (2) symptoms related to mental disorders or a confirmed psychiatric disorder; (3) inability to communicate in writing or verbally; and (4) severe visual, hearing, or expressive impairment that prevented questionnaire completion despite standardized assistance. Sample Size According to recommendations for factor analysis, the sample size should generally be 5–10 times the number of items. The preliminary scale contained 32 items; therefore, at least 160 participants were required for each factor-analysis sample. Because EFA and CFA were conducted using independent samples, the minimum total sample size required for this study was 320.24 A total of 451 valid questionnaires were ultimately included. Before item analysis, the sample was randomly divided approximately 1:1 into two non-overlapping subsamples using the SPSS random-number function. Sample A (n=226) was used for item analysis and EFA, whereas Sample B (n=225) was used for CFA and CFA-based evaluation of convergent and discriminant validity. Scale Format and Scoring The 32-item preliminary scale used a 5-point response format. Perceived susceptibility, severity, and benefits were rated from 1 (strongly disagree) to 5 (strongly agree), whereas self-efficacy was rated from 1 (not at all confident) to 5 (very confident). All items were positively scored. The final scale yields separate scores for perceived susceptibility, perceived severity, perceived benefits, and self-efficacy; no total score is calculated. Higher scores indicate higher levels of the corresponding domain. The complete scale and scoring instructions are provided in Supplementary Table S4. Data Collection and Quality Control Demographic and clinical data were collected using a general information questionnaire. For participants reporting an IDH-related event in the previous 3 months, severity was operationally classified from recalled symptoms, clinical management, and dialysis disruption. Mild events involved symptoms without targeted management or premature termination. Moderate events required immediate management, such as normal saline or glucose. Severe events caused early interruption or termination of dialysis. This self-reported classification was not independently verified against monitoring data or medical records. Among participants reporting an IDH-related event, event frequency during the previous 3 months was categorized as once, 2–4 times, or ≥5 times. No standardized blood pressure threshold was used to define this self-reported event variable. Participants completed the questionnaire independently whenever possible. For those able to understand and answer but unable to mark responses because of reading, visual, or positioning limitations, trained investigators read items neutrally and recorded responses verbatim without explanation or guidance. Administration mode was recorded. Two researchers reviewed questionnaire completeness and response patterns and independently entered and cross-checked the data. Questionnaires were excluded for missing scale items, completion by family members, or response patterns judged invalid by consensus based on the paper questionnaires and field records, including implausibly short completion times combined with highly concentrated responses. All analyzed questionnaires had complete scale data; no imputation was performed. Thirty formal-survey participants were selected by convenience sampling for a 2-week retest. Before retesting, investigators confirmed that dialysis prescriptions were unchanged, no new IDH-related education had been provided, and no hospitalization, acute complication, or marked clinical change had occurred. Statistical Analysis Analyses were conducted in SPSS 27.0 and R 4.4.1. Continuous variables were summarized as mean ± standard deviation or median (interquartile range), as appropriate; categorical variables as n (%). Between-group comparisons used independent-samples t tests, Mann–Whitney U-tests, χ2-tests, or Fisher’s exact tests. Demographic and clinical characteristics, recruitment center, and administration mode were compared between Samples A and B. Administration-mode analyses compared participant characteristics and domain scores between self- and assisted-completion groups. Generalized linear models with a normal distribution, identity link, and robust covariance estimator adjusted for age, educational level, and recruitment center; Holm correction was applied across the four tests of administration-mode effects on domain scores. With only four centers, intracluster correlation coefficients and multilevel models were not estimated. Center heterogeneity was assessed by between-center comparisons, and center was included as a fixed categorical factor in adjusted models. Item analysis in Sample A included critical-ratio statistics for the upper and lower 27% groups, corrected item–domain correlations, and Cronbach’s α if item deleted. An absolute critical-ratio value ≥3.00 with P<0.05 indicated adequate discrimination, and a corrected item–domain correlation ≥0.40 supported retention. Items failing a criterion or increasing α after deletion were reviewed for content, theoretical assignment, and domain coverage.18 Sampling adequacy and factorability were assessed using the Kaiser–Meyer–Olkin (KMO) measure and Bartlett’s test of sphericity. EFA in Sample A used a polychoric correlation matrix, principal axis factoring, and direct oblimin rotation. Factor retention considered 1,000-iteration parallel analysis, the scree plot, theoretical consistency, and interpretability. Items were evaluated using a primary loading criterion of ≥0.40 and a ≥0.20 difference between the two largest absolute loadings, together with theoretical consistency, interpretability, item content, and domain coverage. Parallel analysis and EFA were repeated after each deletion, and a Pearson-correlation sensitivity analysis was conducted.24 CFA in Sample B treated the five-point item responses as ordered categorical indicators and used weighted least squares mean- and variance-adjusted (WLSMV) estimation; maximum likelihood re-estimation treating item responses as continuous served as a sensitivity analysis.25 Fit was evaluated using χ2/df, comparative fit index (CFI), Tucker–Lewis index (TLI), root mean square error of approximation (RMSEA) with 90% confidence interval (CI), and standardized root mean square residual (SRMR). Standardized loadings, latent correlations, residual variances, and standardized residuals were also examined. Post hoc within-domain residual correlations were considered only when supported by modification indices, item content, and theory.26 Composite reliability (CR) and average variance extracted (AVE) evaluated convergent validity, and the Fornell–Larcker criterion evaluated discriminant validity.27,28 Six known-groups hypotheses were prespecified from the HBM. Higher perceived susceptibility and severity were expected among participants reporting IDH-related events or more frequent IDH, and higher perceived benefits and self-efficacy among those receiving IDH-related education. Mann–Whitney U or Kruskal–Wallis H-tests were used, with effect sizes reported as r or ε2 and Holm correction across six tests. Floor or ceiling effects were defined as more than 15% of participants attaining the theoretical minimum or maximum domain score.29 Internal consistency was assessed in the full sample using Cronbach’s α and McDonald’s ω. Two-week test–retest reliability was assessed in the retest subsample using intraclass correlation coefficients (ICCs) with 95% CIs from a two-way mixed-effects, absolute-agreement, single-measure model.17,29–31 All statistical tests were two-sided, with P<0.05 considered statistically significant unless otherwise specified. Ethical Considerations This study was approved by the Research Ethics Committee of the Affiliated Hospital of Binzhou Medical University (approval no. [2025] KYLL-266). Written informed consent was obtained from participants in the semi-structured interviews. Verbal informed consent was obtained from participants in the psychometric validation survey, as approved by the ethics committee. Participant privacy and data confidentiality were strictly protected throughout the study. The study was conducted in accordance with the Declaration of Helsinki.32 Results Participant Characteristics Of 480 questionnaires collected, 29 were excluded: 12 for missing scale items, five for family-member completion, and 12 for implausibly short completion times with highly concentrated responses. The final sample comprised 451 questionnaires (valid-response rate, 94.0%) from tertiary hospitals in Binzhou (n=160), Yantai (n=124), Nanchong (n=61), and Suining (n=106). In the previous 3 months, 257 participants (57.0%) reported an IDH-related event, and 272 (60.3%) had received IDH-related health education. The analytic sample was randomly divided into Sample A (n=226) and Sample B (n=225). The samples were similar in most demographic and clinical characteristics, center distribution, and administration mode, although the distribution of dialysis duration differed significantly. The participant-to-item ratios were 7.3:1 for the 31 items entering EFA in Sample A and 7.8:1 for the final 29-item CFA in Sample B. The questionnaire was self-completed by 190 participants (42.1%) and completed with investigator assistance by 261 (57.9%). Although the groups differed in several characteristics, administration mode was not associated with any domain score after adjustment for age, educational level, and recruitment center, and no association remained significant after Holm correction. Domain scores differed across recruitment centers; recruitment center was therefore retained as a fixed categorical factor in the adjusted models. Table 1 presents key characteristics, whereas Supplementary Tables S5–S8 provide full-sample characteristics and detailed administration-mode and recruitment-center analyses. Table 1Participant Characteristics and Comparison of Samples A and B (n=451)CharacteristicSample A: EFA (n=226)Sample B: CFA (n=225)Test StatisticP valueAge, years, mean (SD)56.39 (12.62)55.83 (13.78)t(449)=0.4560.649Sex, n (%)χ2(1)=0.1210.727 Male136 (60.2)139 (61.8) Female90 (39.8)86 (38.2)Educational level, n (%)χ2(3)=1.4710.689 Junior high school or below169 (74.8)162 (72.0) Senior high school or technical secondary school42 (18.6)42 (18.7) Junior college10 (4.4)12 (5.3) Bachelor’s degree or above5 (2.2)9 (4.0)Dialysis duration, years, n (%)χ2(2)=6.5470.038 <143 (19.0)66 (29.3) 1–5121 (53.5)106 (47.1) >562 (27.4)53 (23.6)Dialysis frequency, n (%)χ2(2)=2.9010.234 2 times/week29 (12.8)21 (9.3) 3 times/week194 (85.8)197 (87.6) 5 times per 2 weeks3 (1.3)7 (3.1)Self-reported IDH experience in the previous 3 months, n (%)χ2(1)=0.5180.472 Yes125 (55.3)132 (58.7) No101 (44.7)93 (41.3)Received IDH-related health education, n (%)χ2(1)=0.0030.954 Yes136 (60.2)136 (60.4) No90 (39.8)89 (39.6)Recruitment center, n (%) Binzhou85 (37.6)75 (33.3)χ2(3)=1.1030.776 Nanchong28 (12.4)33 (14.7) Suining52 (23.0)54 (24.0) Yantai61 (27.0)63 (28.0)Questionnaire administration mode, n (%) Assisted completion135 (59.7)126 (56.0)χ2(1)=0.6450.422 Self-completion91 (40.3)99 (44.0)Notes: Data are presented as mean (SD) or n (%).Abbreviations: CFA, confirmatory factor analysis; EFA, exploratory factor analysis; IDH, intradialytic hypotension; SD, standard deviation. Participant Characteristics and Comparison of Samples A and B (n=451) Notes: Data are presented as mean (SD) or n (%). Abbreviations: CFA, confirmatory factor analysis; EFA, exploratory factor analysis; IDH, intradialytic hypotension; SD, standard deviation. Item Generation The literature review identified candidate domains and informed preliminary items. New codes emerged through interview 13; interviews 14 and 15 yielded no additional codes or categories, supporting data saturation. Literature evidence, interview coding, and research-team discussion generated a 55-item pool across perceived susceptibility, severity, benefits, barriers, and self-efficacy. Supplementary Tables S1–S2 present item sources and interview codes. Item Refinement Delphi Expert Consultation Round 1 yielded 38 valid responses, corresponding to a response rate of 90.5%; all 38 experts completed round 2, with a response rate of 100%. Expert authority coefficients were 0.86 and 0.87, and Kendall’s W values were 0.234 and 0.387, respectively (both P<0.001). After round 1, 23 items were deleted, five were added, and three pairs of items were merged, yielding 34 items. Round 2 deleted three items and added one, producing a 32-item, four-domain preliminary scale. None of 11 candidate perceived-barrier items met prespecified retention criteria, and experts noted overlap with self-efficacy in content or wording. Perceived barriers were therefore not retained as an independent domain after considering quantitative ratings, qualitative feedback, and team discussion. Supplementary Table S3 details the refinement trajectory. Pretesting All 30 participants completed the 32-item preliminary scale within 5–10 minutes. No item was repeatedly identified as unclear or difficult to understand, and no specific wording revisions were suggested; therefore, no changes followed pretesting. Item Analysis In Sample A (n=226), critical-ratio statistics ranged from 9.134 to 22.661 (all P<0.001). D4 had a corrected item–domain correlation of 0.336, below 0.40, and deleting it increased self-efficacy α from 0.763 to 0.770; D4 was removed. Deleting B5 slightly increased α; however, the item met all other retention criteria and was retained on substantive grounds. The remaining 31 items entered EFA (Supplementary Table S9). Exploratory Factor Analysis After D4 removal, 31 items entered EFA in Sample A. The Kaiser–Meyer–Olkin value was 0.803, and Bartlett’s test was significant [χ2(465)=7117.195, P<0.001]. Initial parallel analysis suggested retention of five factors. In the corresponding EFA solution, D5 and D6 loaded on a fifth factor. D6 did not show its strongest loading on the intended self-efficacy factor and lacked a theoretically interpretable alternative assignment; it was therefore removed. Parallel analysis of the remaining 30 items supported four factors, but D5 still failed to load on its intended factor and was removed for the same reason. Parallel analysis of the final 29 items supported four factors, and every item loaded highest on its intended factor. Primary pattern coefficients ranged from 0.579 to 0.902 and communalities from 0.397 to 0.790. The four factors accounted for 65.29% of the variance, with no Heywood cases. Pearson-correlation sensitivity analysis reproduced the four-factor structure, with absolute primary loadings of 0.518–0.839. Table 2 summarizes the primary pattern coefficients and communalities for the final EFA solution, whereas Figure 2 presents the parallel analysis and scree plot. Supplementary Tables S10–S12 provide the item-reduction history, complete EFA results including the full pattern matrix and factor correlations, and sensitivity analysis, respectively. Table 2EFA Results for the Final 29 Items in Sample A (n=226)ItemPerceived SusceptibilityPerceived SeverityPerceived BenefitsSelf-EfficacyCommunality(h2)A10.8630.630A20.8470.708A30.7570.661A40.7790.747A50.7640.599A60.7850.630A70.6950.727A80.7440.790B10.6600.749B20.6050.680B30.6420.565B40.6080.640B50.7290.397B60.8450.713B70.8870.730B80.7100.596B90.5790.583C10.7390.544C20.7130.615C30.8700.708C40.7280.597C50.8180.728C60.7120.676C70.7370.617C80.7250.694C90.6460.530D10.9020.771D20.7820.742D30.7470.570Notes: Only primary pattern coefficients are displayed. Primary loadings are shown in bold. The complete unsuppressed pattern matrix and item-level diagnostics are provided in Supplementary Table S11.Abbreviation: EFA, exploratory factor analysis. EFA Results for the Final 29 Items in Sample A (n=226) Notes: Only primary pattern coefficients are displayed. Primary loadings are shown in bold. The complete unsuppressed pattern matrix and item-level diagnostics are provided in Supplementary Table S11. Abbreviation: EFA, exploratory factor analysis. Parallel analysis and scree plot for the final 29-item scale in Sample A (n=226). (A) Parallel analysis. (B) Scree plot; the dashed horizontal line in panel B indicates an eigenvalue of 1. Confirmatory Factor Analysis The 29-item four-factor model was tested with WLSMV in Sample B (n=225), which had not informed item selection or EFA. The initial model showed acceptable CFI and TLI but suboptimal RMSEA [χ2=1009.658, df=371, χ2/df=2.721, CFI=0.959, TLI=0.955, RMSEA=0.088 (90% CI, 0.081–0.094), SRMR=0.076]. Two post hoc within-domain residual correlations were specified sequentially for A1–A2 and A3–A4 based on modification indices, item content, and theoretical justification. Each modification significantly improved model fit [Δχ2(1)=72.293 and 30.532; both P<0.001]. The final model showed acceptable fit [χ2=705.530, df=369, χ2/df=1.912, CFI=0.978, TLI=0.976, RMSEA=0.064 (90% CI, 0.057–0.071), SRMR=0.061]. Final standardized loadings ranged from 0.700 to 0.905 (all P<0.001), with squared multiple correlations (R2) ranging from 0.490 to 0.819. For self-efficacy, loadings were 0.750–0.895 and R2 values ranged from 0.562 to 0.800. Standardized residual correlations were 0.826 for A1–A2 and 0.886 for A3–A4; both were post hoc and content-justified. In the maximum-likelihood sensitivity analysis, the same correlations reduced RMSEA from 0.101 to 0.073, although CFI (0.898) and TLI (0.887) remained slightly below conventional thresholds. Table 3 presents loadings and R2; Supplementary Tables S13–S16 provide model comparisons, parameters, residual diagnostics, and sensitivity analyses. Table 3Standardized Factor Loadings and R2 for the Final Confirmatory Factor Analysis Model (n=225)DomainItemStandardized LoadingR2Perceived susceptibilityA10.7210.520A20.7420.551A30.8100.656A40.7870.619A50.8040.646A60.8320.691A70.8330.694A80.9050.819Perceived severityB10.8420.708B20.8680.754B30.7910.626B40.7230.522B50.7000.490B60.7850.616B70.8590.738B80.7200.519B90.8370.701Perceived benefitsC10.7270.528C20.7380.545C30.8430.711C40.8440.713C50.7880.621C60.8380.701C70.8100.657C80.8230.677C90.7520.565Self-efficacyD10.8490.720D20.8950.800D30.7500.562Notes: Estimates were obtained using the WLSMV estimator. All standardized factor loadings were statistically significant at P<0.001. R2 represents item-level explained variance.Abbreviation: WLSMV, weighted least squares mean- and variance-adjusted estimator. Standardized Factor Loadings and R2 for the Final Confirmatory Factor Analysis Model (n=225) Notes: Estimates were obtained using the WLSMV estimator. All standardized factor loadings were statistically significant at P<0.001. R2 represents item-level explained variance. Abbreviation: WLSMV, weighted least squares mean- and variance-adjusted estimator. Validity Evidence Among the 29 items retained in the final scale, I-CVI values ranged from 0.80 to 1.00, and the recalculated S-CVI/Ave was 0.969. Literature review, patient interviews, Delphi consultation, and pretesting provided preliminary evidence supporting content relevance, comprehensibility, and coverage. In Sample B, CR ranged from 0.871 to 0.940 and AVE from 0.630 to 0.694, supporting convergent validity. For each domain, the square root of AVE exceeded its correlations with the other latent variables, satisfying the Fornell–Larcker criterion and supporting discriminant validity (Table 4). Table 4Convergent and Discriminant Validity (n=225)DomainCRAVE12341. Perceived susceptibility0.9080.6490.8062. Perceived severity0.9390.6300.6230.7943. Perceived benefits0.9400.6350.5650.6570.7974. Self-efficacy0.8710.6940.2160.3530.3830.833Notes: Bold diagonal values represent the square roots of AVE; off-diagonal values represent latent factor correlations. Only the lower triangle is shown.Abbreviations: AVE, average variance extracted; CR, composite reliability. Convergent and Discriminant Validity (n=225) Notes: Bold diagonal values represent the square roots of AVE; off-diagonal values represent latent factor correlations. Only the lower triangle is shown. Abbreviations: AVE, average variance extracted; CR, composite reliability. After Holm correction, three of six prespecified known-groups hypotheses were supported. Participants reporting an IDH-related event in the previous 3 months scored higher on perceived susceptibility and severity, and those receiving IDH-related education scored higher on perceived benefits; all effect sizes were small. Among participants reporting an IDH-related event, perceived susceptibility and severity did not differ by reported event frequency, and self-efficacy did not differ according to receipt of IDH-related education (Supplementary Table S17). Reliability and Score Distributions Across the full sample, Cronbach’s α ranged from 0.794 to 0.922 and McDonald’s ω from 0.805 to 0.961. Thirty participants completed the 2-week retest; domain ICC (A,1) values were 0.839–0.894 (Table 5). For the three-item self-efficacy domain, α=0.794, ω=0.805, and ICC=0.894 (95% CI, 0.789–0.948). Table 5Internal Consistency (n=451) and Test–Retest Reliability (n=30) of the Four DomainsDomainItemsCronbach’s α(95% CI)McDonald’s ωICC (95% CI)Perceived susceptibilityA1–A80.922 (0.910–0.932)0.8910.870 (0.585–0.949)Perceived severityB1–B90.905 (0.888–0.919)0.9530.839 (0.627–0.927)Perceived benefitsC1–C90.909 (0.892–0.923)0.9610.850 (0.452–0.945)Self-efficacyD1–D30.794 (0.746–0.833)0.8050.894 (0.789–0.948)Notes: Cronbach’s α and McDonald’s ω were estimated in the total sample (n=451); 95% CIs for α were based on 2,000 bootstrap samples. ICCs were estimated in the retest sample (n=30) using a two-way mixed-effects, absolute-agreement, single-measurement model [ICC(A,1)].Abbreviations: CI, confidence interval; ICC, intraclass correlation coefficient. Internal Consistency (n=451) and Test–Retest Reliability (n=30) of the Four Domains Notes: Cronbach’s α and McDonald’s ω were estimated in the total sample (n=451); 95% CIs for α were based on 2,000 bootstrap samples. ICCs were estimated in the retest sample (n=30) using a two-way mixed-effects, absolute-agreement, single-measurement model [ICC(A,1)]. Abbreviations: CI, confidence interval; ICC, intraclass correlation coefficient. All domain-score distributions were negatively skewed (−0.270 to −0.915). No participant attained a theoretical minimum, and no floor effect was observed. The proportions of participants attaining the maximum possible score were 8.20% for perceived susceptibility, 5.99% for perceived severity, 4.66% for perceived benefits, and 34.37% for self-efficacy. Only the self-efficacy domain exceeded the prespecified threshold for a ceiling effect. Supplementary Table S18 presents full score and self-efficacy item-response distributions. Discussion Main Findings and Construct Interpretation This study developed and preliminarily evaluated a 29-item IDH-Related Health Beliefs Scale comprising perceived susceptibility, perceived severity, perceived benefits, and self-efficacy. EFA and CFA provided preliminary support for the four-factor structure, and the findings generally supported content validity, convergent and discriminant validity, internal consistency, and test–retest reliability. However, known-groups validity received only partial support, self-efficacy showed a substantial ceiling effect, and the final CFA model included post hoc residual correlations. The findings therefore provide preliminary rather than definitive evidence of structural stability or clinical utility. The scale assesses IDH-related health beliefs in patients undergoing MHD. Perceived susceptibility and severity represent judgments about IDH likelihood and consequences, whereas perceived benefits and self-efficacy represent judgments about the value of preventive behaviors and confidence in performing them.14 Because these related domains have distinct theoretical meanings, they should be scored and interpreted separately. Higher-order and bifactor models were not tested, so current evidence does not support a total score.33 Identical totals could reflect different domain profiles and obscure specific beliefs. Higher scores are not uniformly favorable: susceptibility and severity require interpretation against objective clinical risk, whereas low benefits or self-efficacy may indicate educational or behavioral-support needs. Item Development and Theoretical Structure Items were developed and refined through the HBM, literature and guideline review, patient interviews, Delphi consultation, and pretesting. The framework defined candidate domains; literature and existing instruments informed item generation; and interviews contributed target-population experiences. Expert consultation assessed relevance, importance, wording, and theoretical assignment, providing multisource content-validity evidence. However, formal cognitive interviews were not conducted, so respondent interpretation of items, instructions, and response options requires further evaluation.17,34–36 Perceived barriers are central to the HBM, but the candidate barrier items were not retained as an independent domain during item refinement. Expert feedback indicated unclear conceptual boundaries between some barrier and self-efficacy items. Perceived barriers concern difficulties or costs that impede behavior, whereas self-efficacy concerns confidence in performing it; the constructs are related but distinct.13,14 Thus, the findings indicate only that the available items failed to support a clear, relatively independent barriers domain, not that barriers are unimportant in IDH prevention. Further interviews and cognitive interviewing should distinguish external barriers from low self-efficacy and reassess measurability.36,37 Factor Structure and Psychometric Performance Initial parallel analysis suggested retention of five factors. In the corresponding EFA solution, D5 and D6 loaded mainly on an additional factor. D5 concerned developing an individualized plan with healthcare professionals, whereas D6 concerned adjusting fluid intake after professional reminders. Both therefore reflected professional support, collaboration, or external prompting beyond behavioral confidence and may not have represented self-efficacy alone. Because the factor had few items and no clear theoretical boundary, it was not retained. After sequential deletion of D6 and D5 because of non-intended primary loadings and limited theoretical interpretability, the final EFA yielded an interpretable four-factor structure. Parallel analysis and the Pearson-correlation sensitivity analysis provided additional support for this solution. Factor retention therefore reflected statistical evidence, item content, and theoretical interpretability rather than the prespecified framework alone.17,38 In the validation sample, the initial model showed favorable CFI and TLI values but a suboptimal RMSEA. Two post hoc within-domain residual correlations, A1–A2 and A3–A4, were added based on modification indices, item content, and theoretical considerations. These modifications improved model fit, but standardized residual correlations >0.80 indicated substantial local dependence and possible item redundancy. The items were retained because they represented distinct IDH risk cues, and deletion could narrow content coverage. No further parameters were added to limit overfitting. Future studies should compare retention, revision, combination, or deletion of these items.39 These limited, content-justified post hoc modifications require confirmation in an independent sample. Only three self-efficacy items remained after screening. Their standardized loadings, internal consistency, composite reliability, convergent-validity evidence, and test–retest reliability were generally supportive, but the small item set may restrict construct coverage. Moreover, 34.37% of participants attained the maximum score, indicating limited discrimination at high self-efficacy and raising concerns about the ability to detect further improvement.29 This ceiling effect may reflect relatively easily endorsed behaviors, the small number of items, and a narrow range of item difficulty. Cognitive interviews and item response theory could guide refinement of content and difficulty.40 Cronbach’s α, McDonald’s ω, and CR supported internal consistency and composite reliability, whereas ICCs supported temporal stability. AVE supported convergent validity, and the Fornell–Larcker criterion supported discriminant validity.27,28 However, only three of six known-groups hypotheses were supported. Unsupported findings may reflect reliance on patient-reported IDH data and the inability of the IDH-related education variable to capture content, quality, or comprehension. Construct validity should combine prespecified hypotheses with external measures of related constructs. Further validation should therefore include external instruments, objective clinical data, and longitudinal outcomes.41 Potential Clinical Relevance and Relationship to Existing Instruments IDH identification and management should continue to rely on objective information, including intradialytic blood pressure, symptoms, ultrafiltration, interdialytic weight gain, comorbidities, and medications.42 Patient-reported instruments complement these data by capturing experiences, perceptions, and needs not directly reflected in clinical measures.43,44 The present scale should therefore supplement, not replace, blood pressure monitoring, diagnosis, risk assessment, or prescription management and should not be used to predict whether IDH will occur. After further feasibility testing, the scale could be administered before patient education or nursing communication, with domain scores reviewed alongside objective clinical information to identify belief areas that may warrant further discussion. Clinicians should interpret domains separately without calculating a total score. Low susceptibility or severity scores, particularly when discordant with objective clinical risk, may indicate possible underestimation of IDH risk or consequences and warrant further discussion. Low perceived benefits may prompt discussion of preventive value, whereas low self-efficacy may indicate a need for behavioral guidance, skills support, or patient–provider collaboration. Domain scores may inform communication and help identify educational needs but are not clinical thresholds for nursing intervention. Because cutoff scores, predictive validity, and responsiveness are unavailable, the scale should not be used for diagnosis, event prediction, prescription adjustment, or intervention evaluation. The scale differs from the Brief IPQ, HDSMI, and Kidney Disease Quality of Life Short Form (KDQOL-SF), which assess general illness perceptions, self-management behavior, and health-related quality of life, respectively. The present scale was designed to assess IDH-specific threat perceptions, perceived preventive benefits, and behavioral confidence.15,16,45 These complementary instruments provide different patient-reported perspectives. Because this study neither compared them directly nor assessed external criterion validity, the new scale cannot be considered superior and is best positioned as a complement to clinical information and existing patient-reported measures. Limitations and Future Research Several limitations warrant consideration. First, convenience sampling from only four tertiary hospitals in China may have introduced selection bias, restricted the clinical spectrum represented in the sample, and limited generalizability. With four centers, multilevel modeling and formal estimation of within-center clustering were not performed, so clustering may have affected estimates. Although all responses came from patients, 57.9% of questionnaires used standardized investigator reading and recording. The study therefore evaluated mixed administration rather than fully independent self-administration. Administration mode was not statistically associated with domain scores after adjustment for age, educational level, and recruitment center, but investigator involvement and social-desirability bias remain possible. Second, the cross-sectional design precluded assessment of subsequent IDH events or behavioral change. IDH experience, frequency, and severity were self-reported and unverified against standardized blood pressure criteria, dialysis records, or medical records, creating potential recall bias and misclassification. Only three of six known-groups hypotheses were supported, and no external instrument provided additional construct-validity evidence. Third, the final CFA model included two post hoc residual correlations, and the three-item self-efficacy domain showed a substantial ceiling effect. Formal cognitive interviews were not conducted, and predictive validity, responsiveness, measurement error, clinical cutoff scores, and minimal important change were not evaluated. The scale should therefore not be used to predict IDH, evaluate interventions, or make direct clinical decisions; current applicability is limited to populations and settings similar to those studied. Future studies should reassess the four-factor structure in independent multicenter samples across regions and cultures and verify IDH events using standardized blood pressure monitoring and clinical records. Cognitive interviews should refine locally dependent susceptibility items and self-efficacy items, while external instruments should extend construct-validity testing. Longitudinal studies should evaluate predictive validity, responsiveness, measurement error, and clinical interpretability, and item response theory may help evaluate item information and difficulty and guide further item refinement. Cross-language use requires standardized translation, cultural adaptation, and measurement-invariance testing. Conclusion This study developed the HBM-based IDH-Related Health Beliefs Scale for patients undergoing MHD. The final scale contains 29 items across perceived susceptibility, perceived severity, perceived benefits, and self-efficacy. The current sample provides preliminary evidence for its four-factor structure, internal consistency, test–retest reliability, and selected aspects of validity. As a candidate patient-reported instrument, the scale provides a preliminary means of assessing domain-specific IDH-related health beliefs and identifying priorities for education and risk communication. Domains should be interpreted separately rather than combined into a total score, and results should be considered alongside clinical status. The scale cannot replace objective IDH monitoring or clinical risk assessment. Before routine use, independent multicenter studies across regions and cultures should confirm structural stability, and longitudinal studies should evaluate predictive validity, responsiveness, and clinical feasibility. Acknowledgments The authors would like to thank the experts who participated in the Delphi consultation for their valuable comments and professional guidance. The authors also thank the nurses in the hemodialysis units for their support during participant recruitment and data collection. We are grateful to all patients who participated in this study. Funding Statement This research did not receive any specific grant from funding agencies in the public, commercial, or not-for-profit sectors. Abbreviations AVE, average variance extracted; CFA, confirmatory factor analysis; CFI, comparative fit index; CR, composite reliability; EFA, exploratory factor analysis; HBM, Health Belief Model; ICC, intraclass correlation coefficient; I-CVI, item-level content validity index; IDH, intradialytic hypotension; KMO, Kaiser–Meyer–Olkin; MHD, maintenance hemodialysis; RMSEA, root mean square error of approximation; S-CVI/Ave, scale-level content validity index based on the average method; SRMR, standardized root mean square residual; TLI, Tucker–Lewis index; WLSMV, weighted least squares mean- and variance-adjusted estimation. Data Sharing Statement The data that support the findings of this study are not publicly available due to privacy and ethical restrictions but are available from the corresponding author upon reasonable request and with permission from the relevant ethics committee. Consent for Publication Not applicable. This manuscript does not contain any identifiable individual information, images, videos, or audio recordings. Audio recordings from the semi-structured interviews were used only for transcription and qualitative analysis, and all interview data were anonymized. Author Contributions All authors made a significant contribution to the work reported, whether that is in the conception, study design, execution, acquisition of data, analysis and interpretation, or in all these areas; took part in drafting, revising or critically reviewing the article; gave final approval of the version to be published; have agreed on the journal to which the article has been submitted; and agree to be accountable for all aspects of the work. Disclosure The authors report no conflicts of interest in this work. References Associated Data Data Availability Statement The data that support the findings of this study are not publicly available due to privacy and ethical restrictions but are available from the corresponding author upon reasonable request and with permission from the relevant ethics committee.