Validation of the Arabic Version of Patient Health Questionnaire‐9 in Omani and Yemeni Patients With Chronic Physical Diseases Authors: Amira Mohammed Ali, Saeed A Al‐Dossary, Carlos Laranjeira, Sameer A Alkubati, Maryam Alharrasi, Heba Emad El‐Gazar, Shawana Masad Al Harrasi, Rasmieh Alamer, Mohamed Ali Zoromba, Ahmad M A Zayed, Abdalla S M Gaballa, Annamaria Pakai, Murat Yıldırım, Ahmad H Abu Raddaha, Haitham Khatatbeh Published: 2026 ABSTRACT ABSTRACTBackgroundSecondary depression frequently develops in people with chronic diseases, and it may hinder disease progress and treatment outcomes. This study aimed to examine the construct validity and measurement invariance of the Arabic version of the Patient Health Questionnaire‐9 (PHQ‐9) among patients with chronic physical conditions.MethodsThis cross‐sectional study used confirmatory factor analysis (CFA) and multigroup CFA to evaluate the Arabic PHQ‐9 in a convenience sample comprising patients with cancer, heart failure (HF), and renal failure (RF) from Oman and Yemen.ResultsOut of eight models, the unidimensional PHQ‐9 (depression) displayed satisfactory fit when three pairs of error were correlated whereas a similar fit was exhibited by a three‐factor (somatic, cognitive/affective, and concentration/motor) solution, which was modified into (somatic, affective, and cognitive–psychomotor) solution based on partial scalar analysis tests. This structure was invariant across all sociodemographic characteristics, with better values of reliability coefficients, average variance extracted, and heterotrait‐monotrait ratios of correlation.ConclusionThe Arabic PHQ‐9 is a reliable and psychometrically stable tool that may differentiate depressive symptoms in people with chronic disorders into three symptom groups. Meaningful comparisons between countries support higher psychological burden of physical disorders in Yemen than Oman.Keywords: cancer, depression, heart failure, measurement invariance, omani, patient health questionnaire‐9 (PHQ‐9), reliability, renal dysfunction, validity, yemeni Abbreviations Average Variance Extracted Confirmatory Factor Analysis Comparative Fit Index Confidence Interval Degrees of Freedom Heart Failure Heterotrait‐Monotrait Patient Health Questionnaire Primary Care Evaluation of Mental Disorders Patient Questionnaire Renal Failure Root Mean Square Error of Approximation Standardized Root Mean Residual Tucker–Lewis Index Chi‐square Introduction Globally, more than 332 million people suffer from depressive disorders, accounting for the largest proportion of disability‐adjusted life years (DALYs: 36.24%) (Chen et al. 2025). The highest prevalence of depression is noticed in younger groups (≥ 12–25 years) (Goodwin et al. 2022). Depression is also the primary cause of premature death in 727,000 people who lost their lives due to suicide in 2021 (World Health Organization 2025). The MENA region, which comprises 19 out of 22 Arab countries, records higher incidence and prevalence rates of depression than the global averages, with prevalence considerably varying across countries (Moradinazar et al. 2022), particularly among special groups such as medical students (40%–77.9%) (Al‐Amer et al. 2025). The prevalence of depression in Oman considerably varies among different population groups: 45.7% in parents of children with cancer (Al‐Saadi et al. 2025), 33.7% in university students (Al Salmani and et al. 2025), 23.4% in patients with inflammatory bowel diseases (Al‐Aamri et al. 2022), and 13.6% in emergency nurses (Al Mawaali et al. 2024). Struck by war, political instability, and related unfavorable living conditions, the prevalence of depression in Yemen is generally higher: 70.3% in infertile women (Gubran et al. 2026), 63% in patients on hemodialysis (Alkubati et al. 2024), and 73.7% in medical students (Al‐naib et al. 2025). It is associated with unemployment, low income, lack of medical insurance, living in rural areas, family problems, comorbidities, and a history of depression (Gubran et al. 2026; Alkubati et al. 2024; Al‐naib et al. 2025). Nearly half the adult population endures one or more chronic diseases, necessitating adherence to treatment regimen and coping with numerous daily challenges (e.g., loss of independence and symptom self‐management) (Scott et al. 2023). Depression is highly prevalent among people with chronic physical conditions such as heart failure (HF), cancer, and renal failure (RF), with documented negative effects on self‐care capability, disease progression and prognosis as well as patients' quality of life (Loosen et al. 2021; Shang et al. 2019; Adejumo et al. 2024). Aggregate data indicate a pooled depression prevalence of up to 40% in RF (Adejumo et al. 2024) while a bidirectional relationship has been revealed between depression and RF (Zheng et al. 2022). Follow up data (5 years) note accelerated development of depression in HF and cancer survivors, with HF patients exhibiting a significantly higher likelihood of developing depression than oncology patients (Loosen et al. 2021). Despite the increasing prevalence of depression, help seeking is globally low. Less than 6% of community‐dwelling depressed individuals reported seeking professional help or pharmacological treatment for depression between 2015 and 2019 (Goodwin et al. 2022). Unfortunately, Arabs struggling with depression refrain from professional help seeking due to negative social views that prevail in the predominantly collectivistic Arabic culture (Ali et al. 2025). Treatment seeking for depression is even lower in people with chronic physical diseases as they do not recognize depressive symptoms and associate them with the physical disease (Alderson et al. 2014). The literature harbors multiple self‐assessment measures that exhibit strong depression screening utility in people with chronic diseases. Nonetheless, most measures comprise items that capture cognitive‐affective and somatic symptoms, with a possible overlap of depressive somatic symptoms with physical disease (Mehmetoglu et al. 2025). The Patient Health Questionnaire (PHQ‐9) is a nine‐item measure that has been derived from the Primary Care Evaluation of Mental Disorders Patient Questionnaire (PRIME‐MD PQ) (Alkubati et al. 2024). It is a comprehensive, brief, and easy to use, score, and interpret screening tool that complies with the diagnostic criteria of Diagnostic and Statistical Manual of Mental Disorders, Fourth Edition (DSM‐IV) for major depressive disorders (Alkubati et al. 2024). It has been as effective in depression screening among chronically ill people (systemic sclerosis and HF) as longer measures such as Center of Epidemiologic Studies Depression Scale (Mehmetoglu et al. 2025) and Beck Depression Inventory‐II (Hammash et al. 2013). While the PHQ‐9 is generally used as a unidimensional measure (Chae et al. 2025), Western validations reported two‐factor structures that have varying item combinations in patient populations (Krause et al. 2008; Richardson and Richards 2008; Krause et al. 2010) as well as healthy university students (Rufino et al. 2024). Similar results were reported in healthy populations and university students in countries from the far East such as Vietnam and China (Wang et al. 2023; Vu et al. 2022). However, a bifactor solution expressed a better fit than the unidimensional and two‐factor solutions in Japanese clinical and non‐clinical respondents (Doi et al. 2018). The Arabic PHQ‐9 has been extensively used to capture depression as a single domain in various populations from both Oman (Al‐Saadi et al. 2025; Al Salmani and et al. 2025; Al‐Aamri et al. 2022; Al Mawaali et al. 2024; Alharrasi et al. 2025) and Yemen (Gubran et al. 2026; Alkubati et al. 2024). However, in the light of inconsistency of its structure (Chae et al. 2025; Doi et al. 2018), checking its cultural sensitivity in the Arab context is necessary to ensure adequate measurement. The psychometrics of the PHQ‐9 have been reported among Arab patients with chronic physical conditions such as HF, cancer, and spinal cord injury. Nonetheless, testing approaches have been inconsistent (Alharrasi et al. 2025; Summaka et al. 2019; Alkalbani et al. 2025), with only one study evaluating its construct validity and measurement invariance through confirmatory factor analysis (CFA) (Alharrasi et al. 2025). While the PHQ‐9 was unidimensional among Saudi caregivers (Alreshidi 2024), different dimensional structures (two‐ and three‐factors) have been reported in patient samples (Alharrasi et al. 2025; Summaka et al. 2019), including those with psychiatric problems (Sawaya et al. 2016). Therefore, the current study aimed to evaluate the construct validity and measurement invariance of the Arabic version of the PHQ‐9 among patients with cancer, HF, and RF from Oman and Yemen—two neighboring Arab countries with distinct demographic and economic profiles. Comparing the levels of depressive symptoms between the groups of country and physical disease came as a secondary aim. Methods Study Design, Participants, and Procedure This study is a secondary analysis based on a cross‐sectional design and four convenience samples of patients with cancer, HF, and RF who were at least 18 years old and agreed to provide free and voluntary consent to participate in the study. Nurse practitioners working at the recruitment sites referred eligible patients to the study. Sociodemographic and clinical data were extracted from the medical record. Patients with a history of psychiatric diseases were excluded from the study. The survey was carried out at varying time intervals during the COVID‐19 pandemic. Data collection methods also varied. Questionnaires were handed to two samples of Yemeni patients in Hodeida city: (N = 200) at the Dialysis Center and (N = 334) at Al Amal Cancer Center of Al‐Thawra Hospital. A researcher was available to assist by reading questionnaires and explaining items if there were any questions (e.g., providing help for illiterate patients). Two samples were collected from Oman. The first sample comprised female breast cancer survivors (N = 154) who were recruited from the oncology clinic at Sultan Qaboos University Hospital and the National Oncology Center at Royal Hospital. They completed an anonymous online survey conducted through the SurveyMonkey platform (San Mateo, CA). Patients with HF (N = 180) were recruited from five tertiary care settings in north and south Oman. They were interviewed by trained research assistants either on site or on the phone according to participants' preference. Measures Sociodemographic variables were extracted from the medical record (e.g., age, gender, marital status, and employment). The participants were administered the PHQ‐9. It comprises nine items which inquire about the extent to which respondents were bothered during the preceding 2 weeks by problems pertaining to loss of interest “Little interest or pleasure in doing things”, sadness/low mood “Feeling down, depressed, or hopeless”, dysfunctional sleep “Trouble falling or staying asleep, or sleeping too much”, disrupted energy and fatigue “Feeling tired or having little energy”, appetite changes “Poor appetite or overeating”, negative self‐view “Feeling bad about yourself”, poor concentration “Trouble concentrating on things”, obvious psychomotor changes “Moving or speaking so slowly…or the opposite — being so fidgety or restless”, and suicidal ideation “Thoughts that you would be better off dead”. Each item is rated on four alternative responses, which range from 0 to 3 (0 = not at all, 1 = a few days, 2 = more than half of the days, and 3 = almost every day). The minimum and maximum total PHQ‐9 scores range from 0 to 27. Higher scores signify greater levels of depression (Kroenke et al. 2001). The Arabic PHQ‐9 displayed adequate reliability in former investigations (alpha = 0.88 and 0.85) (Alharrasi et al. 2025; Sawaya et al. 2016). Ethical Consideration Study procedures followed local regulations and guidelines regulating ethical approvals, which were granted by the Research Committee of Omani Ministry of Health (MOH/CSR/20/11,605) and Ethical Review Committee at Sultan Qaboos University Hospital (CSR# 35/2019; SQU‐EC/113/19). The study complies with the guidelines of the Declaration of Helsinki. Before data collection, potential respondents were informed about the study and assured about voluntary participation. Those who agreed to participate signed informed. Statistical Analysis We computed the mean and standard deviations of the total scores of the PHQ‐9 and its subscales. We compared the differences in the raw scores of the scale and its subscales between the sociodemographic, disease, and country groups. To determine the structure of the PHQ‐9, seven competing models were tested through confirmatory factor analysis (CFA). Model 1 is the standard unidimensional structure. The following three 3 models are based on Western studies capturing two‐factor (somatic and non‐somatic) solutions of the PHQ‐9 in patients with spinal cord injury. They slightly vary according to differences in items on both dimensions. Model 2 is a bidimensional structure in which the somatic dimension comprises items 3, 4, and 5, which reflect sleep difficulty, fatigue, and appetite change while the rest of the items load on a non‐somatic domain (Krause et al. 2008). Model 3 is similar to Model 2 with a slight change that involves the loading of items 7 “trouble concentrating” and item 8 “slow or restless” on the somatic domain (items 3, 4, 5, 7, and 8) (Richardson and Richards 2008). Model 4 is the same as Model 2 with a single change—item 1, which represents anhedonia loads on the somatic domain rather than the cognitive‐affective domain (items 1, 3, 4, and 5) (Krause et al. 2010). Model 5 is based on a two‐factor structure reported in Lebanese psychiatric patients in which items 1, 2, 6, 8 and 9 loaded on the cognitive‐affective factor while items 3, 4, 5 and 7 loaded on the somatic factor (Sawaya et al. 2016). Model 6 is a three‐factor structure reported in Lebanese patients with spinal cord injury (Summaka et al. 2019): Factor 1 comprised the first three items (1–3) “anhedonia, low mood, and poor sleep”; Factor 2 comprised items 4 to 7 “tired, appetite, bad about self, and poor concentration”; and Factor 3 comprised the last two items (8, 9) slow or restless and suicidal ideation). Model 7 is an adaptation of three‐factor structure reported in the testing of the Spanish PHQ‐9 among Ecuadorian college students involving the constructs of somatic (items 3, 4, 5), cognitive‐affective (items 1, 2, 6, 9), and concentration/motor (items 7, 8) symptoms (López‐Guerra et al. 2022). The criteria used to address the fit of the model as acceptable or good were non‐significant χ2 test in addition to Comparative Fit Index (CFI) and Tucker–Lewis Index (TLI) greater than 0.90 and 0.95, with root mean square error of approximation (RMSEA) and standardized root mean residual (SRMR) lesser than 0.08 and 0.06, respectively (Ali et al. 2022, 2024). Akaike information criterion (AIC) was also included; the best fitting factor solution would also be indicated by the lowest AIC. Multigroup CFA was conducted to examine invariance of the PHQ‐9 across groups of gender, age, marital status, employment, disease condition, and country (see Table 1 for groups). Across each characteristic, a hierarchical structure of four nested models was conducted to test measurement invariance: (1) a configural model with freely estimated factor loadings across groups; (2) a metric/weak invariance model with equal loadings (λ); (3) a scalar/strong invariance model with equal loadings and intercepts (τ), allowing for latent mean comparisons; and (4) a strict invariance model with equal residuals (δ) across groups (Ali et al. 2023, 2024). To evaluate whether the added constraints significantly impacted model fit, differences in chi‐square (Δχ2) values between models were examined. Ideally, χ2 should be at significance level below 0.05, but it can be sensitive to sample size. Therefore, variance is judged according to additional fit indices. Specifically, thresholds of less than 0.020 for ΔCFI and ΔTLI and less than 0.015 for ΔRMSEA would reflect non‐invariance (El‐Gazar et al. 2024). TABLE 1Sociodemographic characteristics of the subjects.CharacteristicsAll samples (N = 868)Omani cancer (n = 154)Omani HF (n = 180)Yemeni cancer (n = 334)Yemeni RF (n = 200)Age (mean SD)51.8 (15.9)41.7 (13.9)70.3 (9.7)52.6 (13.9)41.7 (12.8)GenderMales381 (43.9%)—87 (48.3%)145 (43.4%)149 (74.5%)Females487 (56.1%)154 (100%)93 (51.7%)189 (56.6%)51 (25.5%)Marital statusMarried658 (75.8%)126 (81.8%)122 (67.8%)259 (77.5%)151 (75.5%)Unmarried210 (24.2%)28 (18.2%)58 (32.2%)75 (22.5%)49 (24.5%)EmploymentEmployed257 (29.6%)70 (45.5%)50 (27.8%)76 (22.8%)61 (30.5%)Unemployed611 (70.4%)84 (54.5%)130 (72.2%)258 (77.2%)139 (69.5%)PHQ‐99.8 (5.7)5.8 (4.4)8.9 (5.7)11.2 (5.5)11.3 (5.6)α0.8160.8450.8450.7510.789α if item deleted0.786–0.8080.819–0.8430.812–0.8510.709–0.7400.753–0.782Item total correlations0.428–0.6040.434–0.6430.333–0.6980.383–0.5350.379–0.584Abbreviations: α, alpha coefficient; HF, heart failure; RF, renal failure. Sociodemographic characteristics of the subjects. Abbreviations: α, alpha coefficient; HF, heart failure; RF, renal failure. Discriminant validity was evaluated in models with two or more factors using the Heterotrait‐Monotrait (HTMT) ratio of correlations. HTMT values below 0.85 or 0.90—based on stricter or more lenient criteria—indicate that the constructs are sufficiently distinct. Higher values suggest construct overlapping (Ali, Hori, et al. 2022). Additionally, squared correlations between constructs were calculated, and the Fornell–Larcker criterion was applied. According to this criterion, discriminant validity is supported when the Average Variance Extracted (AVE) for each construct is greater than the squared correlation shared with other constructs (Hilkenmeier et al. 2020). Two indicators were used to assess the convergent validity of the PHQ‐9 and its subscales: (1) AVE, which reflects the extent to which items represent their underlying latent construct—an AVE value greater than 0.50 is generally considered indicative of adequate convergent validity; and (2) item‐total correlations, with values above 0.20 suggesting acceptable item consistency (Alvarenga et al. 2024). The reliability of the PHQ‐9 and its subscales was determined in the whole sample through coefficient alpha and coefficient omega (Alvarenga et al. 2024). The analyses were executed in SPSS version 25 (IBM Corp., 2017), JASP Version 0.17.2 (JASP Team, 2023), and Amos version 22 (IBM Corp., Armonk, N.Y., USA). Results were considered significant at a probability of < 0.05 in two‐tailed tests. Results Sociodemographic Characteristics of the Participants The sociodemographic characteristics of the participants are shown in Table 1. Most participants were females, married, and unemployed. According to age median 50 (40–65) years, less than half the participants were aged above 50 years (n = 409, 47.1%) while more participants were aged 50 years or below (n = 459, 52.9%). Welch's ANOVA based on median with adjusted DF shows significant differences in the scores of age and PHQ‐9 between the subsamples F(3) = 32,535.16, 1246.93; both p values = 0.001. Post hoc comparisons revealed no significant age differences between Yemeni patients with RF and Omani oncology patients, nor differences in PHQ‐9 scores between Yemeni patients with cancer and RF. Factor Structure of the Patient Health Questionnaire‐9 The unidimensional PHQ‐9 displayed poor fit (Table 2), but it approached satisfactory fit after correlating three pairs of residuals (Figure 1, b). Likewise, the fit of all two‐factor somatic and cognitive‐emotional structures was acceptable only after correlating error terms. The previously reported Lebanese three‐factor structure (Summaka et al. 2019) converged only when item 7 was moved from the second to the third factor, and it displayed suboptimal fit, which improved after correlating the error terms of item 3 with 4 and 6 with 7 (Figure 1, l). Among all structures, the Spanish three‐factor PHQ‐9 structure (somatic, cognitive‐affective and concentration/motor subcomponents) expressed satisfactory fit in the absence of correlated errors (Figure 1, m) (López‐Guerra et al. 2022), and it slightly improved when the error terms of items 6 and 7. A second order structure based on this model expressed a similar fit while the bifactor structure did not converge. The German two‐factor solution previously reported in oncology patients (Hinz et al. 2016) displayed poor fit, and modification indices suggested correlating several error terms to improve the fit (Figure 1, p). In our investigation of scalar variance of model 7 across countries, we consulted modification indices, which indicated better loading of item 6 on the third factor. Nonetheless, it was still correlated with the error of item 9, which also displayed higher loading on the third factor. The final model retained items 6 and 9 on the third factor instead of the second factor, which comprised only the first two items. This model (somatic, affective, and cognitive–psychomotor) displayed excellent fit on all fit indices in the absence of any error correlations (Table 2; Figure 1, q). TABLE 2Goodness‐of‐fit of the confirmatory factor analysis models representing the Arabic Patient Health Questionnaire‐9 (PHQ‐9) among patients with chronic diseases (N = 868).Modelsχ2 p DFCMIN/DFAICCFITLIRMSEARMSEA 90% CISRMRModel 1 (1F)Crude207.590.001277.6920,443.310.9050.8740.0880.077 to 0.0990.051Error110.050.001244.5920,351.760.9550.9320.0640.052 to 0.0770.038Model 2 (2F) (Chae et al. 2025)Crude148.850.001265.7320,386.570.9360.9110.0740.063 to 0.0860.043Error104.160.001244.3420,345.870.9580.9370.0620.050 to 0.0740.036Model 3 (2F) (Krause et al. 2008)Crude194.430.001267.4820,432.150.9120.8780.0860.075 to 0.0980.051Error127.910.001245.3320,369.630.9460.9180.0710.059 to 0.0830.043Model 4 (2F) (Richardson and Richards 2008)Crude165.590.001266.3720,403.310.9270.8990.0790.067 to 0.0900.046Error119.450.001244.9820,361.170.9500.9250.0680.56 to 0.0800.039Model 5 (2F) (Mehmetoglu et al. 2025)Crude194.280.001267.4720,432.000.9120.8780.0860.075 to 0.0980.051Error102.540.001234.4620,346.260.9580.9350.0630.051 to 0.0760.038Model 6 (3F) (Zheng et al. 2022)Crude203.720.001248.4920,445.440.9060.8590.0930.081 to 0.1050.051Error131.870.001225.9920,377.580.9420.9060.0760.064 to 0.0860.041Model 7 (3F) (Krause et al. 2010)Crude140.200.001245.8420,381.920.9340.9090.0750.063 to 0.0870.042Error110.470.001234.8020,354.190.9540.9280.0660.054 to 0.0790.037Model 8 (2F) (Sawaya et al. 2016)Crude197.500.001267.6020,440.000.9100.8760.0870.076 to 0.0990.050Error105.040.001234.5720,350.000.9570.9330.0640.052 to 0.0770.037Model 9 (3F)Crude85.900.001243.5820,330.000.9680.9510.550.042 to 0.0670.030 Note: Model selection is based on CFI/TLI ≥ 0.95, RMSEA and SRMR ≤ 0.06, and the lowest AIC values.Abbreviations: AIC, Akaike information criterion; CFI, comparative fit index; CI, confidence interval; DF, degrees of freedom; RMSEA, root mean square error of approximation; SRMR, standardized root mean residual; TLI, Tucker–Lewis index; χ2, chi‐square. Goodness‐of‐fit of the confirmatory factor analysis models representing the Arabic Patient Health Questionnaire‐9 (PHQ‐9) among patients with chronic diseases (N = 868). Note: Model selection is based on CFI/TLI ≥ 0.95, RMSEA and SRMR ≤ 0.06, and the lowest AIC values. Abbreviations: AIC, Akaike information criterion; CFI, comparative fit index; CI, confidence interval; DF, degrees of freedom; RMSEA, root mean square error of approximation; SRMR, standardized root mean residual; TLI, Tucker–Lewis index; χ2, chi‐square. Various structures of the Arabic Patient Health Questionnaire‐9 (PHQ‐9) that were tested in confirmatory factor analysis. Model 1 (a, b), model 2 (c, d), model 3 (e, f), model 4 (g, h), model 5 (i, j), model 6 (k, l), model 7 (m, n), model 8 (o, p), and model 9 (q). Internal Consistency, Convergent Validity, and Discriminant Validity of the Patient Health Questionnaire‐9 The PHQ‐9 demonstrated adequate internal consistency, with no considerable increase in scale reliability upon item removal in all samples, along with adequate levels of item correlations (Table 1). The reliability of the domains of all bidimensional structures (model 2 to model 5 and model 8) was within acceptable range while the reliability of one of the three domains of the three‐factor structures (model 6, model 7, and model 9) was low (Table 3). AVE expressed by all factors of the PHQ‐9 in all models was suboptimal; however, some subscales of the bi‐ and tri‐dimensional structures, especially model 9, displayed higher AVE than the unidimensional PHQ‐9. In all multidimensional structures, AVE values were smaller than squared inter‐construct correlations between items comprising those models. Meanwhile, at least one value of HTMT value in all those models was above acceptable thresholds (Table 3), indicating suboptimal discriminant validity, albeit model 9 displayed the most acceptable HTMT values among all. In support of its internal structure, the PHQ‐9 had strong positive correlations with its somatic, affective, and cognitive–psychomotor domains (r = 0.84, 0.77, and 0.83, p values < 0.01). The median (Q1‐Q3) expressed by the three subscales in the entire samples were 4 (2–6), 2 (1–3), and 3 (1–5), respectively. TABLE 3Heterotrait‐monotrait ratio of correlations, average variance explained, and internal consistency of the Arabic Patient Health Questionnaire‐9 (PHQ‐9) among patients with chronic diseases (N = 868).ModelsSquared inter‐construct correlationHTMTAVEω α Model 1 (1F)——0.3400.8140.816Model 2 (2F)0.690.8040.483, 0.3200.738, 0.7270.732, 0.733Model 3 (2F)0.770.9500.375, 0.3270.753, 0.6490.743, 0.634Model 4 (2F)0.720.8380.407, 0.3390.736, 0.7040.714, 0.717Model 5 (2F)0.770.9570.315, 0.4000.686, 0.7390.683, 0.710Model 6 (3F)0.90, 0.90, 0.810.951 0.930 0.9420.388, 0.345, 0.2700.650, 0.672, 0.4120.642, 0.671, 0.438Model 7 (3F)0.56, 0.71, 0.720.819 0.740 0.9460.483, 0.327, 0.3620.738, 0.649, 0.5230.732, 0.634, 0.546Model 8 (2F)0.850.9000.365, 0.3490.611, 0.7630.609, 0.755Model 9 (3F)0.74, 0.58, 0.500.855, 0.705, 0.7780.483, 0.408, 0.3520.738, 0.571, 0.6810.732, 0.549, 0.674Abbreviations: AVE, average variance extracted; F, factor; HTMT, heterotrait‐monotrait ratio of correlations; α, alpha coefficient, ω, omega coefficient. Heterotrait‐monotrait ratio of correlations, average variance explained, and internal consistency of the Arabic Patient Health Questionnaire‐9 (PHQ‐9) among patients with chronic diseases (N = 868). Abbreviations: AVE, average variance extracted; F, factor; HTMT, heterotrait‐monotrait ratio of correlations; α, alpha coefficient, ω, omega coefficient. Measurement Invariance of the Patient Health Questionnaire‐9 According to fit indices shown in Table 2, along with convergent and discriminant validity tests in Table 3 the three‐factor structures (model 7 and model 9) were evaluated for measurement invariance across groups. As shown in Tables 4 and 5, both structures demonstrated measurement invariance at the configural, metric, and scalar levels across all sociodemographic characteristics (gender, age, marital status, and employment). However, model 7 exhibited partial scalar non‐invariance across countries (Table 5). TABLE 4Invariance of the three‐factor (model 7) structure of the Patient Health Questionnaire‐9 (PHQ‐9) among Arab patients with chronic disorders (N = 868).GroupsInvariance levelsχ2 DF p Δχ2 ΔDF p(Δχ2)CFIΔCFITLIΔTLIRMSEAΔRMSEASRMRGender (single vs. married)Configural192.78480.0010.9280.8920.0590.053Metric220.80540.00128.0160.0010.9170.0110.8890.0030.060−0.0010.056Scalar230.21600.0019.4260.1520.9150.0020.8980.0090.0570.0030.058Strict301.82690.00171.6190.0010.8840.0300.8790.0170.062−0.0050.070Age (> 50 years vs. =< 50 years)Configural200.05480.0010.9220.8840.0600.058Metric211.69540.00111.6360.0710.9200.0020.893−0.0090.0580.0010.062Scalar222.97600.00111.2960.0800.9170.0030.900−0.0070.0560.0000.066Strict241.99690.00119.0290.0250.9120.0050.908−0.0080.0540.0010.069Marital relation (single vs. married)Configural171.94480.0010.9040.9360.0550.061Metric203.83540.00131.8960.0010.8970.0100.9220.0020.057−0.0020.075Scalar207.42600.0013.5960.7320.9080.0000.924−0.0090.0530.0050.077Strict213.40690.0015.9990.7410.922−0.0020.925−0.0130.0490.0060.080Employment (employed vs. unemployed)Configural212.01480.0010.9150.8730.0630.046Metric221.69540.0019.6860.1390.9130.0020.885−0.0120.0600.0030.053Scalar228.35600.0016.6660.3530.9130.0000.896−0.0110.0570.0030.056Strict296.55690.00168.2090.0010.8830.0300.8770.0290.062−0.0050.056Conditions (cancer, RF, HF)Configural350.59930.0010.8750.8550.0570.053Metric373.78990.00123.1960.0010.8670.0080.8550.0000.0570.0000.057Scalar375.411050.0011.6360.9510.869−0.0030.865−0.0100.0550.0020.057Strict444.181140.00168.7790.0010.8400.0290.8490.0160.058−0.0030.055Countries (Oman vs. Yemen)Configural156.11480.0010.9410.9080.0740.045Metric176.18540.00120.0760.0030.9340.0070.9080.0000.0740.0000.058Scalar219.74600.00143.5820.0010.9140.0200.8930.00150.080−0.0060.060Strict506.17690.001286.43110.0010.7660.1480.7560.1370.086−0.0060.104 Note: Model 7 did not converge at the scalar level across conditions or at the configural level across countries; TLI, Tucker–Lewis index; χ2, chi‐square.Abbreviations: CI, confidence interval; CFI, comparative fit index; DF, degrees of freedom; RMSEA, root mean square error of approximation; SRMR, standardized root mean residual. Invariance of the three‐factor (model 7) structure of the Patient Health Questionnaire‐9 (PHQ‐9) among Arab patients with chronic disorders (N = 868). Note: Model 7 did not converge at the scalar level across conditions or at the configural level across countries; TLI, Tucker–Lewis index; χ2, chi‐square. Abbreviations: CI, confidence interval; CFI, comparative fit index; DF, degrees of freedom; RMSEA, root mean square error of approximation; SRMR, standardized root mean residual. TABLE 5Invariance of the three‐factor (model 9) structure of the Patient Health Questionnaire‐9 (PHQ‐9) among Arab patients with chronic disorders (N = 868).GroupsInvariance levelsχ2 DF p Δχ2 ΔDF p(Δχ2)CFIΔCFITLIΔTLIRMSEAΔRMSEASRMRGender (single vs. married)Configural147.75480.0010.9500.9250.0490.033Metric159.75540.00112.0160.0620.9470.0030.930−0.0050.0480.0010.039Scalar191.42600.00131.6760.0010.9350.0120.9210.0090.050−0.0020.048Strict259.41690.00167.9890.0010.9050.0300.9010.0200.056−0.0060.060Age (> 50 years vs. =< 50 years)Configural150.59480.0010.9480.9220.0500.038Metric160.71540.00119.3060.1200.9460.0020.9270.0050.0480.0020.040Scalar180.01600.00144.3560.0040.9390.0070.9270.0000.0480.0000.047Strict194.93690.00110.1290.0930.9360.0030.9330.0060.0460.0020.050Marital relation (single vs. married)Configural128.15480.0010.9590.9380.0440.029Metric148.34540.00120.1980.0030.951−0.0020.9350.0030.045−0.0010.031Scalar158.76600.00110.4280.1080.9490.0020.939−0.0040.0440.0010.034Strict164.11690.0015.36120.8020.951−0.0020.949−0.0100.0400.0040.033Employment (employed vs. unemployed)Configural152.99480.0010.9460.9190.0500.044Metric163.28540.00110.2960.1790.9440.0020.925−0.0060.0480.0020.051Scalar183.13600.00119.8560.0030.9360.0080.9240.0010.049−0.0010.056Strict248.19690.00165.0690.0010.9080.0280.9030.0210.055−0.0060.057Conditions (cancer, RF, HF)Configural324.26910.0010.8870.9290.0460.048Metric341.38990.00117.1280.1130.8830.0040.934−0.0050.0440.0020.053Scalar345.811050.0014.4360.6190.884−0.0010.9290.0050.046−0.0020.051Strict410.541140.00164.7390.0010.8570.0270.7870.1420.080 −0.034 0.049Countries (Oman vs. Yemen)Configural136.87480.0010.9520.9280.0650.037Metric146.33540.0019.4660.1490.9510.0010.934−0.0060.0630.0020.042Scalar169.73600.00123.4060.0010.9410.0100.9150.0190.071−0.0080.049Strict449.66690.001279.9390.0010.7960.1450.7540.1610.121 −0.050 0.108Abbreviations: CI, confidence interval; CFI, comparative fit index; DF, degrees of freedom; RMSEA, root mean square error of approximation; SRMR, standardized root mean residual; TLI, Tucker–Lewis index; χ2, chi‐square. Invariance of the three‐factor (model 9) structure of the Patient Health Questionnaire‐9 (PHQ‐9) among Arab patients with chronic disorders (N = 868). Abbreviations: CI, confidence interval; CFI, comparative fit index; DF, degrees of freedom; RMSEA, root mean square error of approximation; SRMR, standardized root mean residual; TLI, Tucker–Lewis index; χ2, chi‐square. Discussion Although the psychometrics of the PHQ‐9 have been evaluated in up to 13 studies involving Arab participants, its structure has been examined in four studies with inconsistent results including one (Alreshidi 2024), two (Alharrasi et al. 2025; Sawaya et al. 2016), and three dimensions (Summaka et al. 2019). Only one study evaluated the scale for measurement invariance (Alharrasi et al. 2025). To the best of authors' knowledge, this study provides the first comprehensive psychometric evaluation of the Arabic PHQ‐9 among patients with chronic illnesses. Using CFA and multi‐group CFA, it examined the construct validity, measurement invariance, convergent and discriminant validities as well as internal consistency. Eight competing models were tested, including the conventional unidimensional model and alternative multidimensional solutions. Particularly a three‐factor structure representing somatic, affective, and cognitive–psychomotor symptom domains, exhibited excellent fit in the absence of error correlations, invariance across clinical samples from Oman and Yemen, higher subscale internal consistency as well as better convergent and discriminant validities relative to all other structures. Collectively, the findings show that depressive symptoms assessed by the PHQ‐9 are more appropriately conceptualized as multiple interrelated domains, rather than a single homogeneous factor. In line, poor fit was recorded for the unidimensional PHQ‐9 in 2059 oncology German patients while it expressed acceptable fit in 2693 community‐dwelling controls. Alternatively, a two‐factor solution in which items 2 “loss of interest”, 6 “self‐blame”, and 9 “suicidal ideation” loaded on one factor expressed best fit (Hinz et al. 2016). While the unidimensional Chinese PHQ‐9 expressed poor fit among university students, the two‐factor solution expressed superior fit when two items were omitted (Wang et al. 2023). Likewise, CFA indicated good fit of a two‐factor structure among Omani patients with HF (Alharrasi et al. 2025) while exploratory factor analysis yielded two and three factors among Lebanese patients with psychiatric disorders (Sawaya et al. 2016) and spinal cord injury (Summaka et al. 2019), respectively. Unlike the Lebanese three‐factor structure (Summaka et al. 2019), which expressed suboptimal fit despite multiple residual correlations (Figure 1, l), the Spanish three‐factor structure (López‐Guerra et al. 2022) expressed satisfactory fit, which was maximized by correlating the errors of items 6 and 7. Interestingly, investigation of scalar variance of this model across countries indicated a preferrable loading of item 6 (which was freed in partial invariance test) on the third factor. This was associated with another violation of local independence (item 6 correlation with item 9 on the second factor), which was addressed by loading item 9 on the third factor as well. Error correlations involving items 6 and 9 have been consistently noticed (Figure 1). Loading those items on one factor resulted in acceptable fit among German oncology patients (Hinz et al. 2016), suggesting that the shared variance between them may be due to systematic errors. The first two items, which comprise the ultra‐short form of the scale (PHQ‐2), came as a distinct dimension while the somatic dimension remained intact. Despite being sufficiently distinct as noted by HTMT ratios, the three subscales strongly correlated with each other and with the whole scale. This result in addition to higher AVE values demonstrated by its domains support structural coherence and convergence of the three dimensions of model 9 with the parent scale. Thus, it may be inferred from the clean structure of model 9 (absence of error correlations and best discriminant validity among all models) that it involves minimal systematic errors. Scalar invariance is a prerequisite for meaningful comparison of factor means as it ensures that group mean differences represent actual differences in the underlying latent construct rather than systemic errors that pertain to group differences in the interpretation of scale items (Tse et al. 2024; Ali et al. 2026). Model 9 displayed scalar invariance across all characteristics, making PHQ‐9 comparisons empirically informative. This is because scalar invariance is sufficient in the current case given that PHQ‐9 items are ordered‐polytomous whereas strict or partial strict invariance would be required when testing measures with dichotomous items (Tse et al. 2024). The scores of the PHQ‐9 and its subscales (Supplementary materials) spotted the highest depressive and cognitive–psychomotor symptoms among patients with RF, along with equal psychological burden among Yemeni patients with cancer and RF who exhibited significantly higher levels of depression relative to Omani patients with HF and cancer. This pattern aligns with previous reports of substantially higher depression levels in Yemeni populations than in Omani populations (Al‐Saadi et al. 2025; Al Salmani and et al. 2025; Al‐Aamri et al. 2022; Al Mawaali et al. 2024; Gubran et al. 2026; Alkubati et al. 2024; Alharrasi et al. 2025). It also mirrors psychosocial challenges of living with a chronic or life‐threatening illness in the light of economic and political instability, which may influence the severity of depressive symptoms more than the specific diagnosis or age. The latter considerably varied among the subsamples and was positively associated with somatic symptoms only. Particularly, Oman is a politically stable country with a strong economy that involves oil industry, agriculture, overseas trading, and fishery. Such economic and political characteristics are reflected in a state‐funded healthcare system and a nationwide integrated program launched by Ministry of Health for tackling non‐communicable diseases, with a special focus on cancer care and control services (Mehdi and et al.et al. 2022). On the other hand, Yemen has been under the dark shades of war and armed conflicts for many years, which is associated with collapse of the healthcare infrastructure and scarcity of diagnostic and treatment services for many serious conditions such as cancer and RF. Therefore, Yemeni patients usually get diagnosed at later stages of the disease, and they have a limited chance of receiving optimal medical management, eventually leading to low quality of life and higher mortality rates (Al‐Ibraheem et al. 2022). Thus, the findings support known‐group validity of the PHQ‐9. The reliability of the PHQ‐9 in our samples (Table 1) is consistent with the literature. For instance, a systematic review of 98 psychometric studies reports a range of 0.76–0.92 for Cronbach's alpha of the whole scale (Chae et al. 2025). According to Table 3, most two‐factor structures reported fair (< 0.70) to acceptable (> 0.70) internal consistency while all three‐factor PHQ‐9 structures produced one domain with poor reliability (< 0.60) and another one with fair reliability (< 0.70). Not all previous studies reported on the reliability of its subscales. In this respect, the internal consistency of the two factors was high in the validation of the Vietnamese (0.84 and 0.80) (Vu et al. 2022) and Japanese (0.84, and 0.91) versions, and acceptable for the Chinese version (0.788 and 0.791) (Wang et al. 2023). However, for the latter this result was obtained in a seven‐item model, which excluded the last two items “movement” and “desire to die” as they contributed to poor fit of the CFA structure (Wang et al. 2023). Minimal reliability expressed by some factors in the present study may limit the meaningful interpretation of their scores. Further evaluations of these structures in more diverse Arab populations are warranted. Implications for Research and Practice This study offers several important implications for research, clinical practice, and the application of the PHQ‐9 in diverse populations. The findings support a multifaceted conceptualization of depression (somatic, affective, and cognitive–psychomotor) in medically ill populations, rather than a single underlying construct that is widely expressed in the general populations. Therefore, clinicians should consider symptom profiles when assessing depression in those patients and avoid relying solely on a total PHQ‐9 score. The results also highlight the importance of considering the possible interaction of depressive somatic symptoms with physical symptoms of the disease. Different symptom profiles would require tailored interventions as they entail varying responses to treatment approaches. Future research should re‐evaluate PHQ‐9 cutoff points for determining depression severity in populations with chronic physical illness. Studies should also report subscale scores rather than relying exclusively on total scores. Strengths and Limitations This study is strengthened by being the first to examine nine competing models of the PHQ‐9 among individuals with chronic physical disorders. The findings are consistent with previous research indicating inadequacy of the unidimensional structure for the PHQ‐9. Instead, they suggest that depressive symptoms in this population are better represented as distinct somatic, affective, and cognitive–psychomotor dimensions. This structure was conceptually stable across a range of characteristics including country and disease, which would promote meaningful comparisons. This study has several limitations that may affect the generalizability of the findings. The cross‐sectional design and use of convenience sampling, with data collected during the COVID‐19 pandemic, may introduce bias due to pandemic‐related psychological distress, changes in healthcare access, and altered symptom reporting. Additionally, variations in group sizes and data collection procedures could have influenced the results of multigroup CFA. Finally, the inclusion of samples from only two Arab countries limits the extent to which the findings can be generalized to the broader Arab region. Conclusions The Arabic PHQ‐9 is best described by three distinct domains (somatic, affective, and cognitive–psychomotor), which demonstrated stability at the configural, metric, and scalar levels across numerous group characteristics. This structure expressed better convergent validity and reliability than seven other competing structures. Age was associated only with higher somatic symptoms. Yemeni patients expressed higher depressive and cognitive–psychomotor symptoms than Omani patients regardless the diagnosis, emphasizing an extensive need for comprehensive mental health care in Yemeni clinical settings. Replications of the study are needed to confirm the findings. Author Contributions Amira Mohammed Ali: conceptualization, methodology, formal analysis, software, writing – original draft, supervision. Saeed A. Al‐Dossary: methodology, formal analysis, writing – review and editing. Carlos Laranjeira: methodology, validation, writing – original draft, writing – review and editing. Sameer A. Alkubati: investigation, writing – review and editing. Maryam Alharrasi: investigation, data curation, writing – review and editing. Heba Emad El‐Gazar: resources, writing – review and editing. Shawana Masad Al Harrasi: investigation, writing – review and editing. Rasmieh Alamer: data curation, writing – review and editing. Mohamed Ali Zoromba: resources, data curation, writing – review and editing. Ahmad M. A. Zayed: validation, writing – review and editing. Abdalla S. M. Gaballa: formal analysis, validation, writing – review and editing. Annamaria Pakai: project administration, funding acquisition, writing – review and editing. Murat Yıldırım: methodology, writing – review and editing. Ahmad H. Abu Raddaha: validation, writing – review and editing. Haitham Khatatbeh: conceptualization, supervision, writing – review and editing. All authors have approved the final version. Funding The authors have nothing to report. Ethics Statement The study received ethical approval from the Research Committee of Omani Ministry of Health (MOH/CSR/20/11605) and Ethical Review Committee at Sultan Qaboos University Hospital (CSR# 35/2019; SQU‐EC/113/19). Consent Written informed consent was obtained from all participants prior to their inclusion in the study. Written informed consent was obtained from the participants for the publication of this study and any accompanying data. Conflicts of Interest The authors declare no conflicts of interest. Supporting information Supporting Information S1 Acknowledgments The authors extend their sincere gratitude to all participants who took part in this study. Data Availability Statement The datasets used and/or analyzed during the current study are available from the corresponding author upon reasonable request. References Associated Data Supplementary Materials Supporting Information S1 Data Availability Statement The datasets used and/or analyzed during the current study are available from the corresponding author upon reasonable request.