Patient Health Questionnaire-9 and 17-item Hamilton Depression Rating Scale for detecting post-stroke depression in cerebral infarction

Abstract

BACKGROUND

The prevalence of post-stroke depression (PSD) is approximately 30%-50% among patients with cerebral infarction, which hinders recovery and increases mortality: However the diagnosis rate is low as its symptoms overlap with other conditions. Although 17-item Hamilton Depression Rating Scale (HAMD-17) is the clinical reference standard, its complexity limits its use for screening in routine practice. Patient Health Questionnaire-9 (PHQ-9) is easily accessible but has variable validity in stroke cohorts. This study assessed the diagnostic accuracy of both scales against International Classification of Diseases, 10^th Revision criteria in order to guide PSD screening practices.

AIM

To compare the diagnostic efficacy of the HAMD-17 and the PHQ-9 in screening for PSD in cerebral infarction patients, and to explore the optimal diagnostic cutoff values for both scales.

METHODS

Retrospective analysis was performed on clinical data of 345 cerebral infarction patients who were admitted to our institution between February 2020 and March 2025. Psychiatrists performed structured clinical interviews based on the International Classification of Diseases, 10^th Revision criteria to make the gold-standard diagnosis. HAMD-17 and PHQ-9 were completed by all patients. Diagnostic efficacy of both scales was assessed using receiver operating characteristic curve analysis. Difference in area under the curve (AUC) was compared using the DeLong method. Best diagnostic cutoff values were based on the principle of maximum value of Youden index, and Kappa consistency test was calculated to assess the agreement between scale diagnoses and gold standard.

RESULTS

PSD was found in 112 of 345 patients (incidence: 32.46%). The AUC of HAMD-17 for detecting PSD was 0.891 (95%CI: 0.854-0.928), and the AUC of PHQ-9 was 0.867 (95%CI: 0.826-0.908), and there was no significant difference between them (Z = 1.426, P = 0.154). The optimal diagnostic cutoff value for HAMD-17 was 7 scores; its sensitivity was 88.39%, specificity was 79.83%, Youden index was 0.682, and Kappa value was 0.632. The best cutoff value of PHQ-9 for diagnosis is 7 points, at which sensitivity equals is 83.04%, specificity equals is 81.12%, Youden index equals is 0.642, and Kappa value is 0.606.

CONCLUSION

Both HAMD-17 and PHQ-9 have good diagnostic efficacy for PSD screening without significant difference in AUC. As a simple and easy-to-use self-rating tool, PHQ-9 can be the desired scale for early screening of PSD in neurology wards. We suggest a score of ≥ 7 as the most useful positive screening cutoff and recommend that patients screening positive be assessed by a specialist for confirmation of the diagnosis.

Key Words: Post-stroke depression; Hamilton Depression Rating Scale; Patient Health Questionnaire-9; Diagnostic efficacy; Receiver operating characteristic curve

Core Tip: Post-stroke depression is prevalent and often underrecognized in neurological environments. This study directly compares the diagnostic utility of the 17-item Hamilton Depression Rating Scale clinician-rated and self-administered Patient Health Questionnaire-9 (PHQ-9) in patients with cerebral infarction, using International Classification of Diseases, 10^th Revision structured interviews as reference standard. The diagnostic accuracy of both scales was good, and there was no significant difference in the area under the curve values. The best sensitivity and specificity was at a PHQ-9 cutoff score of ≥ 7. Because it is simple and feasible, PHQ-9 might be preferred to assess post-stroke depression in neurology wards for screening alone and then require confirmation by a specialist assessment.

INTRODUCTION

Cerebral infarction is a common acute cerebrovascular disease in the field of neurology, with high incidence and disability rates, which poses a serious threat to patient health and quality of life[1]. Post-stroke depression (PSD) is a common neuropsychiatric complication after cerebral infarction, with the incidence reported to be about 30%-50% in home and abroad[2]. PSD not only prolongs neurological functional recovery and diminishes the patients’ scales of activities of daily living and treatment compliance but is also significantly correlated with heightened risk of recurrence post-stroke and raised long-term mortality[3]. And, given that the clinical manifestations of PSD frequently overlap with the somatic symptoms brought by cerebral infarction itself[4], and some patients may have speech disorders or emotional apathy, PSD is often overlooked or misdiagnosed in clinical practice[4]. Thus, the selection of appropriate screening tools for early identification of PSD is clinically significant.

The Hamilton Depression Rating Scale (HAMD) is the most commonly used clinician-administered scale in a psychiatric clinical practice, and its 17-item version has been identified as the reference standard for grading depression severity[5]. This scale offers great insight into evaluation of depressive symptoms but is time-consuming and inherently limited for large-scale screening in neurology wards as it requires trained professionals to undertake the interview-based assessments[6]. The Patient Health Questionnaire-9 (PHQ-9) is a self-report scale developed according to the internationally accepted diagnostic criteria for depression. The advantages of this method are concise items, easy operation and high patient acceptance, which has been widely used for depression screening in communities and general hospitals[7,8]. Previous studies have confirmed the diagnostic performance of PHQ-9 in stroke and post-stroke populations, but the optimal cutoff remains controversial[9,10].

In China, few studies directly compare the diagnostic and evaluation effectiveness of HAMD-17 and PHQ-9 for PSD screening, and there is no consensus on the optimal diagnostic cut-off values of these two scales[11]. Taking the diagnosis by psychiatrists via structured clinical interviews based on International Classification of Diseases, 10^th Revision (ICD-10) as gold standard, this study compared the diagnostic performance of HAMD-17 and PHQ-9 in screening for PSD systemically and explored the optimal cutoffs for both scales to screen PSD patients in this population, providing evidence-based references for establishing rational PSD screening methods[12].

MATERIALS AND METHODS

Study subjects

Clinical data of patients with cerebral infarction who were hospitalized in our hospital from February 2020 to March 2025 were retrospectively collected. The Medical Ethics Committee of our hospital approved this study protocol. Both HAMD-17 and PHQ-9 assessments were performed as a part of a standardized clinical evaluation protocol used for all stroke inpatients at our institution during the study period, not based on research intent. This context provides ecological validity to the findings of our study, albeit with the inherent limitations of a non-interventional, retrospective design.

Inclusion criteria: (1) Diagnosis of cerebral infarction met the “Chinese Guidelines for Diagnosis and Treatment of Acute Ischemic Stroke”[7] guidelines, assisted by head computed tomography or magnetic resonance imaging examination; (2) Age ≥ 18 years; (3) Depression assessed within 2 weeks to 6 months post-onset; (4) Clear consciousness with ability to cooperate with scale assessment; and (5) Complete clinical data including both HAMD-17 and PHQ-9 assessments as well as a structured clinical interview conducted by psychiatrists.

Exclusion criteria: (1) Patients diagnosed with depression or other psychiatric disorders before cerebral infarction; (2) Severe aphasia or cognitive impairment (unable to cooperate with scale assessment based on clinical assessment); (3) Severe cardiac, hepatic, or renal insufficiency or malignant tumors; (4) Other central nervous system diseases (such as Parkinson’s disease, Alzheimer’s disease etc.); (5) History of alcohol or drug dependence; and (6) Incomplete clinical data.

In total, 512 cases were initially screened according to the above criteria: (1) 42 cases were excluded due to a history of psychiatric disorders; (2) 58 cases had severe aphasia or cognitive impairment; (3) 31 had a severe organic disease; and (4) 36 patients with incomplete data were also excluded. Finally, a total of 345 patients met the inclusion criteria.

Study methods

General data collection: (1) General demographic and clinical data were drawn from the electronic medical record system, including: Sex, age, years of education, marital status, cerebral infarction location (e.g., PCA region), National Institutes of Health Stroke Scale (NIHSS score), Barthel Index (BI) (an assessment tool for activities of daily living), comorbidities (hypertension, diabetes mellitus, coronary heart disease); and (2) Time from onset to assessment.

Depression assessment tools: (1) HAMD-17: The 17-item version was used, with administration performed by neurology physicians trained to a common standard. All administering physicians underwent a standardized training curriculum consisting of didactic education on anchor-point criteria and supervised practice assessments before enrolling in the study. However, the study protocol did not include formal inter-rater reliability testing which is a methodological limitation. The 17 items on the scale are as follows: Depressed mood, guilt feelings, suicide, initial insomnia (problems falling asleep), middle insomnia (sleep maintaining problems) sleep emerged in the early morning), work and activities, retardation, agitation and psychic anxiety somatic anxiety gastrointestinal symptoms general somatic symptom genital symptom hypochondriasis weight loss insight. The total score was from 0 points to 54 points and the scoring criteria were: < 7 points were defined as no depression; 7-17 points were defined as mild depression; 18-24 points were defined as moderate depression; > 24 points were defined as severe depression. In this analysis, HAMD-17 ≥ 7 points was considered the positive screening cutoff value; and (2) PHQ-9: Patients completed the Chinese version of PHQ-9, either on their own or with help from medical staff. The scale comprises 9 items based on the key symptoms of depression for diagnosis, each item scored according to how common was the symptom experience in that last 2 weeks (0 = not at all; I would be lying if I said ’not at all’; or ‘neither are’, 1 = several days; could make this day okay; 2 = more than half the days - meaning quite a lot, and almost everyday: This describes your life)[6]. The total score is between 0 points and 27 points where scoring cut-off values are: No depression (0-4), mild depression (5-9), moderate depression (10-14), moderately severe depression (15-19) and severe depression (20-27). PHQ-9 ≥ 5 points, ≥ 7 points and ≥ 10 points were analyzed as positive screening cutoff values in this study.

Diagnostic gold standard: The gold standard was the diagnosis of PSD made by psychiatric specialists using structured clinical interviews according to ICD-10 criteria. To reduce review bias, the structured clinical interviews were performed blind to the HAMD-17 and PHQ-9 results. Scale scores were recorded and saved independently from the documentation of the clinical interview, and the psychiatrists who performed diagnostic interviews did not have access to quantitative scale outputs before or during these interviews. This separation was preserved as institutional practice, and although a formal pre-registration of a blinding protocol was not available it is recognised that this is a retroactive design limitation.

Statistical analysis

Statistical analysis was performed using SPSS 26.0 and MedCalc 20.0 software. Continuous variables were evaluated for normality by Shapiro-Wilk, complied with a normal distribution expressed as mean ± SD, and the inter-group comparison used independent sample t-test; not contributed to a normal distribution were expressed as median (interquartile range), and the comparisons between groups used Mann-Whitney U test. Results were presented as n (%) for categorical variables, with inter-group comparison using χ² test. Using the ICD-10 diagnosis as the gold standard, we calculated sensitivity, specificity, positive predictive value (PPV), negative predictive value, positive likelihood ratio, negative likelihood ratio and Youden index for both scales at cutoff values. Receiver operating characteristic (ROC) curves were constructed, followed by computation of the area under the curve (AUC), including 95%CI. Differences in AUC between the two scales were compared using the DeLong method. The best diagnostic cutoff value of each scale was obtained according to the maximum Youden index. The two scale scores and gold standard diagnosis were tested for correlation using Spearman's rank order analysis. Inter-scale agreement was assessed using Cohen’s kappa measure, and poor (< 0.40), moderate (0.40-0.75) and good (> 0.75) agreement between both scales and the gold standard diagnosis were considered significant according to Kappa value. P < 0.05 was considered statistically significant.

RESULTS

Occurrence of PSD and comparison of general data between two groups

PSD was diagnosed in 112 (32.46%) and non-PSD in 233 (67.54%) among 345 cerebral infarction patients according to ICD-10 gold standard. There were substantial differences between the groups in age, NIHSS score, BI and the proportion of diabetes mellitus (P < 0.05), but not in sex, years of education, marital status, cerebral infarction location, assessment time and proportions of hypertension and coronary heart disease (P > 0.05; Table 1).

Table 1 Comparison of general data between two groups, n (%)/mean ± SD/median (interquartile range).

Variable	PSD group (n = 112)	Non-PSD group (n = 233)	t/χ2/Z value	P value
Sex			3.72	0.054
Male	54 (48.21)	138 (59.23)
Female	58 (51.79)	95 (40.77)
Age (years)	65.73 ± 10.24	62.18 ± 11.35	2.816	0.005
Years of education (years)	8.42 ± 3.67	8.89 ± 3.82	1.087	0.278
Marital status			1.03	0.598
Married	89 (79.46)	195 (83.69)
Divorced/widowed	18 (16.07)	31 (13.30)
Single	5 (4.46)	7 (3.00)
Cerebral infarction location			2.15	0.341
Anterior circulation	52 (46.43)	124 (53.22)
Posterior circulation	38 (33.93)	76 (32.62)
Multiple	22 (19.64)	33 (14.16)
NIHSS score (points)	6 (3, 9)	4 (2, 7)	-3.624	< 0.001
BI (points)	65 (45, 80)	80 (60, 95)	-4.518	< 0.001
Assessment time (days)	42 (21, 78)	38 (18, 72)	-1.236	0.216
Comorbidities
Hypertension	71 (63.39)	134 (57.51)	1.103	0.294
Diabetes mellitus	48 (42.86)	72 (30.90)	4.762	0.029
Coronary heart disease	29 (25.89)	51 (21.89)	0.689	0.407

BI: Barthel Index; NIHSS: National Institutes of Health Stroke Scale; PSD: Post-stroke depression.

Comparison of scale scores between two groups

According to the data collected (HAMD-17 and PHQ-9), the PSD group had higher HAMD-17 and PHQ-9 scores than the non-PSD group, with statistically significant differences (P < 0.001; Table 2).

Table 2 Comparison of scale scores between two groups, median (interquartile range).

Scale	PSD group (n = 112)	Non-PSD group (n = 233)	Z value	P value
HAMD-17	14 (10, 19)	4 (2, 6)	-12.347	< 0.001
PHQ-9	12 (8, 16)	3 (1, 5)	-11.862	< 0.001

HAMD-17: 17-item Hamilton Depression Rating Scale; PHQ-9: Patient Health Questionnaire-9; PSD: Post-stroke depression.

Correlation analysis with gold standard diagnosis between two scales

Spearman rank correlation analisis was conducted to analyse the relation between HAMD-17 and PHQ-9 scale scores and structured clinical interview diagnoses based on ICD-10 diagnostic criteria. The results indicated strong positive correlation between HAMD-17 scores and ICD-10 diagnosis (rs = 0.762, P < 0.001), while PHQ-9 scores showed similarly strong positive correlation with ICD-10 diagnosis (rs = 0.718, P < 0.001). Good concurrent validity with the gold standard diagnosis was demonstrated by both scales, which achieved correlation coefficients above 0.7 (considered as strong correlation) and were capable of reflecting both presence and severity of PSD. The threshold values for different correlation strengths (strong ≥ 0.7; moderate = 0.5-0.7, weak = 0.3-0.5) are defined by dashed horizontal lines (Figure 1).

Figure 1 Spearman correlation analysis between depression screening scales and International Classification of Diseases, 10^th Revision diagnosis. Both 17-item Hamilton Depression Rating Scale and Patient Health Questionnaire-9 showed strong positive correlations with gold standard diagnosis (rs = 0.762 and 0.718, both P < 0.001). HAMD-17: 17-item Hamilton Depression Rating Scale; ICD-10: International Classification of Diseases, 10^th Revision; PHQ-9: Patient Health Questionnaire-9.

Diagnostic performance of the two scales at various cutoff points

Using a cut-off value of HAMD-17 ≥ 7 points, the sensitivity was 88.39%, specificity was 79.83%, Youden index was found to be 0.682, and Kappa value was computed at 0.632. Using the cutoff value of PHQ-9 ≥ 5 points, sensitivity was found to be 91.07%, specificity was 68.67%, Youden index was 0.597 and Kappa value was 0.521. When the cutoff value was set to PHQ-9 ≥ 7 points, the sensitivity was 83.04%, specificity was 81.12%, Youden index was 0.642, and Kappa value was reported as 0.606. Using PHQ-9 ≥ 10 points as the cutoff value, sensitivity of 73.21%, specificity of 89.27%, Youden index of 0.625 and Kappa value was calculated at 0.632 (Table 3).

Table 3 Comparison of diagnostic efficacy of two scales at different cutoff values.

Scale and cutoff value	Sensitivity (%)	Specificity (%)	PPV (%)	NPV (%)	+LR	-LR	Youden Index	Kappa value
HAMD-17 ≥ 7 points	88.39	79.83	67.81	93.47	4.38	0.15	0.682	0.632
PHQ-9 ≥ 5 points	91.07	68.67	58.29	94.12	2.91	0.13	0.597	0.521
PHQ-9 ≥ 7 points	83.04	81.12	67.88	90.87	4.4	0.21	0.642	0.606
PHQ-9 ≥ 10 points	73.21	89.27	76.64	87.39	6.83	0.3	0.625	0.632

HAMD-17: 17-item Hamilton Depression Rating Scale; NPV: Negative predictive value; PHQ-9: Patient Health Questionnaire-9; PPV: Positive predictive value; +LR: Positive likelihood ratio; -LR: Negative likelihood ratio.

Analysis of ROC curve and optimal diagnostic cut-off values

ROC curve analysis showed that the AUC of HAMD-17 for diagnosing PSD was 0.891 (95%CI: 0.854-0.928), and PHQ-9 AUC was 0.867 (95%CI: 0.826-0.908). Comparing AUC of both scales using DeLong showed no statistically significant difference (Z = 1.426, P = 0.154; Table 4, Figure 2).

Figure 2 Receiver operating characteristic curves of 17-item Hamilton Depression Rating Scale and Patient Health Questionnaire-9 for diagnosing post-stroke depression. Using International Classification of Diseases, 10^th Revision structured clinical interviews as the gold standard, the diagnostic performance of both scales was evaluated. The area under the curve for 17-item Hamilton Depression Rating Scale (HAMD-17) was 0.891 (95%CI: 0.854-0.928) and for Patient Health Questionnaire-9 (PHQ-9) was 0.867 (95%CI: 0.826-0.908), with no significant difference (Z = 1.426, P = 0.154). The optimal cutoff value for HAMD-17 was 7 points (sensitivity is 88.4%, specificity is 79.8%), and for PHQ-9 was 7 points (sensitivity is 83.0%, specificity is 81.1%). Purple solid line represents HAMD-17, yellow dashed line represents PHQ-9, and black dashed line is the reference line. HAMD-17: 17-item Hamilton Depression Rating Scale; PHQ-9: Patient Health Questionnaire-9.

Table 4 Receiver operating characteristic curve analysis results of two scales.

Scale	AUC	95%CI	Optimal cutoff (points)	Sensitivity (%)	Specificity (%)	PPV (%)	NPV (%)	+LR	-LR	Youden Index
HAMD-17	0.891	0.854-0.928	7	88.39	79.83	67.81	93.47	4.38	0.15	0.682
PHQ-9	0.867	0.826-0.908	7	83.04	81.12	67.88	90.87	4.4	0.21	0.642

Comparison of area under the curve between two scales, Z = 1.426, P = 0.154. AUC: Area under the curve; HAMD-17: 17-item Hamilton Depression Rating Scale; NPV: Negative predictive value; PHQ-9: Patient Health Questionnaire-9; PPV: Positive predictive value; +LR: Positive likelihood ratio; -LR: Negative likelihood ratio.

Using the maximum Youden index (which was derived from sensitivity and specificity) principle, the optimal diagnostic cutoff value of HAMD-17 was 7 points (sensitivity is 88.39%, specificity is 79.83%, Youden index is 0.682), while the optimal diagnostic cutoff value of PHQ-9 was also be enacted as 7 points (sensitivity is 83.04%, specificity is 81.12% by Youden index0 factor = 0.642)(Table 4).

Agreement between two scales and gold standard diagnosis

Results from consistency tests with common cutoff values as well as optimal cutoff values for each scale were as follows: (1) Kappa value for the HAMD-17 response of ≥ 7 points (optimal cutoff value) was 0.632 (P < 0.001) when compared with ICD-10 diagnosis; (2) Kappa value for PHQ-9 response of ≥ 5 points was 0.521 (P < 0.001) when compared against ICD-10 diagnosis; (3) Kappa value of PHQ-9 response at a cut-off point of ≥ 7 points (optimal cutoff value) was 0.606 (P < 0.001) compared to ICD-10 diagnosis; and (4) The optimal PBH cut-point pressure of ≥ 10 had a kappa co-efficient corresponding to ICD-10 at 0.632 (P < 0001). All cutoff values had moderate agreement with the gold standard diagnosis (Table 5).

Table 5 Agreement analysis between two scales and gold standard diagnosis.

Scale and cutoff value	Positive cases	True positive	False positive	True negative	False negative	Kappa value	P value
HAMD-17 ≥ 7 points (optimal cutoff value)	146	99	47	186	13	0.632	< 0.001
PHQ-9 ≥ 5 points	175	102	73	160	10	0.521	< 0.001
PHQ-9 ≥ 7 points (optimal cutoff value)	137	93	44	189	19	0.606	< 0.001
PHQ-9 ≥ 10 points	107	82	25	208	30	0.632	< 0.001

HAMD-17: 17-item Hamilton Depression Rating Scale; PHQ-9: Patient Health Questionnaire-9.

DISCUSSION

PSD is the most prevalent affective disorder after cerebrovascular disease, which has a serious impact on patients' neurological functional recovery and quality of life. Based on the retrospective analysis of clinical data from 345 patients with cerebral infarction, and selected as gold standard by ICD-10 structured clinical interview, this study systemically compared the diagnostic efficacy in PSD screening between HAMD-17 and PHQ-9, provided evidence-based guidance for rational selection of screening tools in clinical practice.

The incidence of PSD was 32.46%, and this result is generally consistent with a number of domestic and foreign epidemiological studies[13]. The high prevalence of PSD indicates that clinicians must screen regularly to diagnose and treat cerebral infarction patients to achieve early detection and intervention. This study showed that the patients with PSD were older and had more serious neural functional deficiency (higher NIHSS scores), worse activity of daily living (lower BI) and a higher prevalence of diabetes mellitus, which was consistent with previous studies[14,15]. Elderly patients are often faced with more psychosocial stressors, those who have neurological deficits or severe ones lose confidence about their own recovery and loss of activities of daily living can reduce self-efficacy which may contribute to a higher risk of the development of PSD[16]. Diabetes mellitus is also an important risk factor for cerebral infarction that may be involved in the occurrence and development of PSD by affecting brain monoamine neurotransmitter metabolism and aggravating white matter damage[17].

ROC curve analysis demonstrated that the AUC curve for HAMD-17 as a diagnostic marker for PSD was 0.891, while the AUC of PHQ-9 was 0.867, with no statistically significant difference (P = 0.154), suggesting that both measurement scales have good diagnostic efficacy in screening of PSD. Diagnostic accuracy can be classified according to previous studies as follows: (1) AUC = 0.7-0.9 moderate diagnostic accuracy; and (2) AUC > 0.9 high diagnostic accuracy[18]. The AUC values of both scales in this study were close to 0.9, indicating that their clinical application value in PSD screening is very high. It should be underscored, however, that this range of AUC values indicates good population normal-differential discriminative ability but does not imply that either scale is capable of supplanting a structured psychiatric diagnosis in individual patient care. Screening instruments are probabilistic triage tools that stratify risk and inform referral decisions but do not provide definitive diagnoses, and this distinction should be considered when interpreting the present findings.

HAMD-17, a classic clinician-administered scale with extensive usage in diagnosing and assessing the severity of depression, is considered the reference standard for evaluating depression[19,20]. But this scale has some problems by requiring structured evidence based interviews for scoring by trained raters, long time to assess leading to high human resource and time costs for extensive screening in neurology wards[20]. Being a self-rating scale, PHQ-9 has concise items (only 9 items), can be filled out by the patients freely within 5 minutes, easy to conduct and thus fit for promotion and applied as an initial screen in clinical practice[21]. The current study results showed that diagnostic efficiency of PHQ-9 was similar to HAMD-17, indicating its application value in PSD screening.

Setting appropriate diagnostic thresholds is an essential step in validating screening tools for clinical usage. Association analysis showed that based on the maximum Youden index principle, this study proposed the optimal diagnostic cutoff value for HAMD-17 was 7 points (sensitivity is 52.1%, specificity is 79.4%) and the optimal diagnostic cutoff value for PHQ-9 was 7 points (sensitivity is 46.1%, specificity is 96.7%). At HAMD-17 ≥ 7 points, the sensitivity was 88.39%, and the specificity was 79.83%, similar to cutoff values recommended in previous studies[22] Interestingly, we determined the optimal cutoff value for PHQ-9 was 7 points rather than the clinical cutoffs of 5 points or 10 points.

When PHQ-9 ≥ 5 points as the cutoff value, sensitivity could be up to 91.07% but specificity was only 68.67%, with high false-positive rate which easily leads to over-diagnosis and thus requirements on medical resources[23]. This lower threshold may be appropriate in contexts where the primary goal is to avoid missed cases no matter what the burden on specialist services, as might occur with intended early discharge and when social support is limited. Using ≥ 10 points as the cutoff value, although specificity was increased to 89.27%, sensitivity decreased to 73.21% and led to loss of around one-quarter PSD patients[24]. This elevation aligns better with established primary care thresholds and may be pragmatic in settings of limited resources where capacity to follow up patients with a specialist is limited. A relatively balanced sensitivity of 83.04% and specificity of 81.12%, with the maximal Youden index (0.642) and optimal diagnostic efficacy, was obtained when a cutoff value of ≥ 7 points was used. This balance should not be confused with clinical neutrality: A false-negative rate of approximately 16.96% implies that the first screening would miss about 17 of every 100 true PSD cases, highlighting a continued need for vigilance by clinicians even in patients who screen negative. In addition, the PPV of 67.88% suggests that approximately one in three patients screening positive will not fulfill ICD-10 diagnostic criteria for PSD following formal assessment. Thus, this finding is consistent with and reinforces the proposed two-stage model – PHQ-9 as broad initial filter followed by structured psychiatric evaluation in a confirmation gate-like process – rather than detracting from the utility of PHQ-9 as an initial screen. It should also be noted that PPV directly depends on disease prevalence; in settings where the incidence of PSD is lower than this cohort (32.46%), the proportion of false positives would be anticipated to increase further, resulting in an even greater need for specialist confirmation[15]. This finding is consistent with some studies that have reported a similar threshold among stroke populations[25], while other studies have demonstrated higher or lower cutoffs[26], suggesting that the optimal cutoff value may differ based on study population characteristics, timing of assessment and others.

The Kappa values in the analysis of consistency were as follows: (1) 0.632 between HAMD-17 ≥ 7 points and ICD-10 group diagnosis; (2) 0.606 between PHQ-9 ≥ 7 points and ICD-10 diagnosis; and (3) 0.632 between the HAMD-DR ≥ 10 with a moderate agreement showing good uniformity among these classification methods. These findings suggest that whatever the scale or cut off value, some distance from the gold standard diagnosis remains and scales can never replace clinical diagnosis by psychiatrists[27].

The Kappa value for PHQ-9 ≥ 5 points is only 0.521, with a relatively lower agreement mainly because of the large number of false-positive cases (73 cases). In patients with cerebral infarction, somatic symptoms as fatigue or sleep disorders and/or appetite changes may overlap with depressive symptoms in which applying too low a cutoff value could misclassify these somatic symptoms as manifestations of depression[28]. Thus, reasonable increase of PHQ-9 diagnostic cutoff value is a good way to reduce false positive in the diagnosis limb and improve the specificity for cerebral infarction patients.

According to the full results of this study, PHQ-9 has similar diagnostic performance as HAMD-17 for PSD screening, and it also has advantages such as simple operation, short time, and high patient acceptance rate, so that we believe PHQ-9 should be a preferred tool for first-line screening of suspected PSD patients in neurology wards[29]. In light of the findings from this study, we recommend the following clinical application recommendations: (1) PHQ-9 ≥ 7 points is recommended in screening PSD. This cutoff was selected according to the maximum Youden index in the present study, while recognizing that “optimal” cutoffs may vary across populations[30]; and (2) Patients screen positive for PHQ-9 should receive further assessment by HAMD-17 or be referred to a psychiatry consultation for diagnosis confirmation and depression severity evaluation[31]. The “two-step” screening strategy can maximize the benefits of PHQ-9, which is simple and fast, while increasing diagnostic accuracy through secondary evaluation and reasonably distributing medical resources[32].

Furthermore, clinicians can apply the screening at scale by those from A to Z based on individual patient data. For example, for patients with significant neurological deficits, poor verbal expression abilities or lower levels of education, the self-rating scales may be less reliable and clinician-administered rating scales should be preferentially chosen or consultation sought to assist in the assessment[33].

The study has the following limitations: First, this is a single-center retrospective study and all samples are from the same hospital so generalizability may be limited and require further validationby multi-center prospective studies. The retrospective design also has inherent risks of information bias: (1) Implementation of the scales was not controlled strictly relative to the time at which patients underwent psychiatric interview (i.e., administration was asynchronous); (2) Individual differences in proximity between completion of the scale and structured clinical interview could theoretically affect observed agreement; and (3) Day-to-day fluctuations in depressive symptoms that may characterize early post-stroke period. Future prospective studies should pre-specify a narrow, standardized assessment window to mitigate this source of variability. In addition, while the independence of the psychiatric diagnosis from the scale scores was a maintained practice at the institution in question, there was no formally pre-registered blinding protocol; subsequent prospective studies ought to implement and document formal blinding techniques so that no residual uncertainty may remain about whether index tests and reference standard were indeed independent. Second, the interval of depression assessment for enrolled patients was wide (ranging from 2 weeks to 6 months after onset), and subgroup analysis by various time windows was not conducted; as PSD development is a time-variable event it is possible that optimal diagnostic cutoff values may change during different intervals. Early PSD may be more heavily driven by neurobiological processes, and later-onset depression may be more dependent on psychosocial determinants such as functional disability or impaired sense of self-efficacy; this heterogeneity in the underlying mechanisms would therefore likely affect performance characteristics of both scales[18]. Further investigations must explore whether these criteria displays distinctive optimal cutoff values across acute, subacute and chronic stroke phase. Third, in this study patients with severe aphasia and cognitive impairment were excluded, yet those patients represent the very high-risk population for PSD, and how to screen that population effectively remains a clinical challenge. Both conditions are common sequelae of stroke and independently associated with increased risk for depression, suggesting that the current study population may systematically underrepresent some of the most disadvantaged stroke survivors. In the future, there should be a focus on developing and validating observer-rated or proxy-completed instruments to complement self-report or interview-based assessments among patients who lack the ability to provide their own accounts of HRQOL. Finally, this study did not assess the test-retest reliability and inter-rater reliability of both scales as well as analyze the diagnostic contribution of each scale dimension.

Future researches can fall into following aspects: (1) To conduct multi-center large-sample prospective studies so as to verify the importance of good diagnosis between cutoff values through different populations; (2) To explore PSD screening tools with high applicability for patients suffering from aphasia or cognitive impairment; and (3) By using multiple biomarkers (such as inflammatory factors, neuroimaging indicators, etc.) building a comprehensive diagnostic model to further improve diagnostic efficiency.

CONCLUSION

In summary, both HAMD-17 and PHQ-9 show good diagnostic value in screening for PSD, and there is no statistically significant difference between them in terms of AUC. Results: The optimal cutoff value of HAMD-17 for diagnosis is 7, and the optimal cut point for PHQ-9 is also 7. Due to its simplicity and convenience, PHQ-9 can also be used as a self-rating tool and can be considered the preferred scale for PSD screening in neurology wards. A cutoff value of ≥ 7 points is recommended for positive screening, and patients who screen positive should receive further specialist evaluation to confirm the diagnosis.