Depressive disorder (hereafter referred to as depression) is a highly prevalent mental health condition. The World Health Organization (1) notes that it affects approximately 3.8% of the global population and that over 700,000 suicide deaths occur annually and are attributable to this disorder. Effective screening and diagnosis are critical to identifying individuals requiring intervention. The 9-item Patient Health Questionnaire (PHQ-9) stands as one of the most widely validated and utilized depression screening instruments in both research and clinical contexts (2). Derived from the Primary Care Evaluation of Mental Disorders (PRIME-MD; 3), which was originally anchored in DSM-III-R and DSM-IV diagnostic criteria, the PHQ-9 operationalizes these criteria via a brief self-report format (3, 4). Comprising nine items rated on a 4-point Likert scale, the PHQ-9 assesses the frequency of affective, cognitive, and somatic depressive symptoms over a two-week period. A cutoff score of 10 is conventionally used to denote clinically significant depression, demonstrating 88% sensitivity for major depressive disorder and 93% specificity for non-depressed individuals (4). Subsequent meta-analytic evidence (5, 6) has confirmed that this threshold optimizes the balance between sensitivity and specificity.

The PHQ-9 has exhibited robust psychometric properties across multicultural samples and clinical contexts (4, 7, 8), with cumulative evidence supporting its measurement invariance across linguistic, cultural backgrounds, gender, and age-related dimensions.

In Chinese-speaking populations, the PHQ-9 has also been extensively employed (9), with prior investigations confirming its psychometric stability in diverse samples, including adolescents (10), university students (11), general hospital inpatients (12), and psychiatric patients diagnosed with depression (13, 14). Notwithstanding, the majority of these studies have focused on community samples or diagnostic subgroups, often treating adolescents and adults as homogeneous populations or independent cohorts. Given the increasing recognition of age-related heterogeneity in depression symptomatology (15), whether the PHQ-9 maintains consistent structural and clinical properties across adolescent and adult psychiatric inpatients remains unaddressed—a critical gap for refining age-specific diagnostic protocols. Given the developmental disparities in symptom presentation and clinical needs, a comprehensive psychometric evaluation of the PHQ-9 across these age groups is essential to clarify its structural validity and inform evidence-based screening practices in psychiatric settings.

The factor structure of the PHQ-9 remains a topic of enduring scientific debate (16). Central to this discourse is the controversy regarding whether the scale is best conceptualized as unidimensional or two-factor, with additional heterogeneity in item allocation within two-factor models. Accumulating empirical evidence supports a unidimensional structure: Shevlin et al. (17) demonstrated optimal fit of a single-factor model across four European populations, and Bianchi et al. (18) reported congruent findings in a large community sample. Two-factor models typically aim to discriminate between cognitive/emotional and somatic symptom domains; conversely, studies such as Shin et al. (19) proposed alternative structures, including an “affective-somatic”/”cognitive” two-factor model in a Korean general population. A comprehensive review by Lamela et al. (16) concluded that although two-factor models occasionally exhibit marginally better model fit, unidimensional solutions are supported by more consistent cross-sample evidence. Notably, researchers (18, 20) have highlighted that even in studies favoring two-factor solutions, substantial cross-sample variability in item loadings and high inter-factor correlations challenge the discriminant validity of these factors.

Given the established influence of cultural contexts on the conceptualization and expression of mental health symptoms (21), the factor structure of the PHQ-9 is hypothesized to exhibit population-level variability. In a cross-cultural study of antenatal women across eight low-resource countries, a unidimensional model provided the optimal fit in four populations, while a cognitive-motivational/somatic two-factor model demonstrated superior fit in Jamaica. Notably, no optimal model fit was achieved in the remaining three countries. Network analysis within the same study revealed that although the pattern of associations between PHQ-9 items was consistent across countries—and correlations with external measures aligned with theoretical expectations (22)—the magnitude of these associations varied significantly.

Collectively, these findings suggest that no single factor model can be universally optimal across cultural and clinical contexts; instead, factor structural solutions should be evaluated and validated within specific populations. In the Chinese context, Sun et al. (14) demonstrated that a unidimensional model provided the best fit for data from hospitalized and outpatient patients with depression, while Feng et al. (13) reported congruent results in a sample of depressed inpatients. However, the generalizability and robustness of these findings across broader psychiatric inpatient populations—encompassing diverse diagnostic categories—remain underexplored, creating a critical gap with implications for the diagnostic process.

This gap is further widened by two key limitations in existing research on the PHQ-9. First, the majority of studies have focused on community samples or narrow diagnostic subgroups, leaving the measure’s validity in the aforementioned broader psychiatric inpatient populations (with diverse diagnoses) largely unaddressed. Second, even in inpatient-focused research, adolescents and adults are often treated as a homogeneous group, despite growing evidence of age-related heterogeneity in depression symptomatology (23). To address these dual shortcomings, the present study aims to: (1) validate the Chinese version of the PHQ-9 specifically in Chinese psychiatric inpatients—a focus that stands in critical contrast to community-based findings (e.g., 11, who reported a conventional cutoff of 10 for general populations); and (2) compare the measure’s psychometric properties (including factor structure, network characteristics, and diagnostic cutoffs) between adolescent and adult inpatients. This dual objective not only fills the inpatient-specific validity gap identified earlier but also explores age-related differences within this clinical population, addressing a long-overlooked dimension of PHQ-9 validation.

This study systematically evaluated the psychometric properties of the PHQ-9 among adolescent and adult psychiatric inpatients in China, with a specific focus on its factorial structure, symptom network characteristics, and optimal diagnostic threshold determination. The findings revealed both cross-age-group consistencies and developmentally relevant divergences, providing critical insights into the scale’s clinical applicability and construct validity within this specialized population.

Confirmatory factor analysis (CFA) results revealed that the two-factor model—comprising cognitive-affective and somatic dimensions—exhibited superior model fit compared to the unidimensional model in adults (Δχ²=6.49, p=0.011). In adolescents, however, no significant difference in fit was observed between the two models (Δχ²=0.79, p = 0.374), with both models showing marginally acceptable fit (CFI = 0.91) despite inflated RMSEA (likely due to small sample size; 32). Beyond the potential influence of a small sample size, this finding may also reflect greater heterogeneity in symptom presentation or cognitive immaturity in how adolescents endorse symptoms - an observation that aligns with developmental research. This literature highlights that adults typically conceptualize depressive symptoms through distinct cognitive and somatic frameworks, whereas adolescents often exhibit a more diffuse, less differentiated symptom profile (15).

Building on recent work documenting heterogeneity in adolescent emotional dysregulation (24, 27), we propose that adolescent depressive symptoms constitute an undifferentiated syndrome - one that may be more complex than its adult equivalent. This syndrome is embedded within the broader context of emotional susceptibility differentiation: a developmental process intertwined with biological factors, which manifests in the trajectory of personality development (encompassing both pathological and normative domains; 34).

Network analysis further unveiled age-dependent variations in symptom architecture. Although both networks comprised 38 nonzero edges, the adult network exhibited marginally stronger connectivity (mean edge weight = 0.197 vs. 0.177). Centrality analyses revealed that low mood (PHQ2) and anhedonia (PHQ1) served as core symptoms in adults, aligning with prior evidence highlighting affective disturbances as the clinical pivot of adult depression (8). By contrast, adolescents manifested a distinct structural profile, with fatigue (PHQ4) and concentration difficulties (PHQ7) emerging as central nodes. These findings may reflect the somatic presentation of adolescent depression—a pattern commonly documented in East Asian populations and congruent with somatization theory, which posits that emotional distress in this demographic is often expressed via physical complaints (35).

Significantly, the adolescent network demonstrated poor stability (CS-coefficient for strength = 0), thereby limiting the reliability of centrality estimations. This observation is likely attributable to the smaller sample size (n = 105) and heightened symptom heterogeneity. Future investigations should prioritize replication in larger cohorts to validate these findings. Edge-level analyses unveiled both shared and age-specific symptom associations. For example, the robust linkage between low self-worth (PHQ6) and suicidal ideation (PHQ9) across age groups underscores the clinical salience of self-evaluation in suicidal risk assessment. By contrast, adolescents exhibited stronger modularity among sleep disturbance (PHQ3), fatigue (PHQ4), and concentration difficulties (PHQ7), whereas adults showed tighter affective-fatigue symptom couplings. These topological differences may inform the design of age-tailored interventions, targeting symptom clusters that drive depressive severity propagation.

ROC analysis furnished additional evidence of age-dependent diagnostic disparities in PHQ-9 performance. Although both adolescent and adult inpatient groups exhibited acceptable levels of discriminative validity, adult participants demonstrated superior overall accuracy (AUC = 0.724 vs. 0.636 for adolescents) and a higher Youden index (0.39 vs. 0.31 for adolescents). Notably, the age-specific optimal diagnostic thresholds identified in this study—15.5 for adolescents and 14.5 for adults—surpassed the internationally recommended PHQ-9 cutoff of 10 (4, 6). This elevation in optimal thresholds may be attributed to three key factors: (1) the heightened clinical severity inherent to psychiatric inpatient populations (distinct from community samples); (2) cultural factors, such as the amplification of somatic symptoms observed in East Asian populations (35); and (3) the high comorbidity of depressive symptoms with non-depressive primary diagnoses (see Table 2 ; 64.7% of participants scored ≥15 on the PHQ-9, compared to only 43.7% who received a diagnosis of primary depressive disorder). Further support for the role of comorbidity comes from the relatively low specificity values observed in both groups (adolescents: 0.47; adults: 0.49), as comorbid depressive symptoms may increase false-positive classifications for primary depressive disorder (ICD-11 6A7).

Consistent with these ROC-derived findings, we argue that the conventional PHQ-9 cutoff of 10—while classifying 92.4% of our psychiatric inpatient sample as “depressed”—fails to distinguish between patients with primary depressive disorder (ICD-11 6A7) and those with depressive symptoms comorbid with other psychiatric conditions. In contrast, the higher age-specific cutoffs (15.5 for adolescents, 14.5 for adults) proposed in this study more effectively target patients with clinically significant depression. By reducing the over inclusion of subthreshold depressive cases (i.e., individuals with mild or transient symptoms not meeting diagnostic criteria for primary depression), these cutoffs correct for the severity underestimation that occurs in the true primary depressive disorder subgroup within Chinese psychiatric inpatient samples. Collectively, these findings establish an empirical foundation for optimizing age-stratified depression screening protocols in Chinese psychiatric inpatients. The higher cutoffs (15.5, 14.5) are not intended to replace the conventional cutoff of 10 for community samples (11) but rather to address the unique severity profile of inpatient populations, where comorbid and subthreshold symptoms are common.

Several study limitations merit acknowledgment. First, the cross-sectional nature of the design restricts causal inference regarding temporal symptom dynamics. Second, the small adolescent sample (n=105) may have inflated RMSEA values in CFA, overstating model misfit. Future studies with larger adolescent samples should validate the PHQ-9’s factor structure in this group. The adolescent network’s poor stability (CS-coefficient=0) invalidates confident comparisons with the adult network. Future studies with larger adolescent samples are needed to clarify age-related differences in symptom network structure. Third, recruitment from a single psychiatric hospital may introduce selection bias, potentially compromising external validity to community populations. Fourth, a potential limitation is the uneven sample size (adolescents: n=105; adults: n=380) and differing primary diagnostic distributions between age groups. Although comorbid depressive symptoms were prevalent across all diagnostic categories (ensuring comparability of the PHQ-9’s target construct), future studies with larger, matched adolescent samples (e.g., stratified by primary diagnosis) would strengthen causal inferences about age-related differences. The last limitation is that the PHQ-9 was not originally designed for children <12, so results for this subgroup should be interpreted cautiously.

In summary, the PHQ-9 exhibits good reliability for assessing depressive symptoms among Chinese psychiatric inpatients, including both adolescent and adult subgroups. However, its latent structure and optimal diagnostic cutoffs differ by age: the conventional PHQ-9 cutoff of 10 fails to align with the unique severity profile of this inpatient population, as it cannot distinguish between patients with primary depressive disorder (ICD-11 6A7) and those presenting with comorbid or subthreshold depressive symptoms—leading specifically to the misclassification of symptom severity (rather than the overestimation of depression prevalence). Age-specific cutoffs (15.5 for adolescents, 14.5 for adults) are recommended exclusively for Chinese psychiatric inpatients and are not intended to replace the conventional cutoff of 10 for community samples; notably, the influence of comorbidity on the study’s findings should be accounted for when interpreting these results.