Development and validation of a nomogram model for identifying post-stroke anxiety symptoms

Abstract

BACKGROUND

Currently, no established nomogram models exist for identifying post-stroke anxiety symptoms (PSAS).

AIM

To develop and validate a nomogram for the identification of PSAS.

METHODS

This retrospective cross-sectional study included 947 stroke patients. Demographics, the Generalized Anxiety Disorder 7-item, the Patient Health Questionnaire-9 (PHQ-9), clinical biomarkers, and stroke-related scale assessments were collected. Participants were randomly divided into the training and the internal validation cohorts in a 7:3 ratio. Multivariate logistic regression was employed to identify predictors for constructing the nomogram. The performance of the nomogram was evaluated using receiver operating characteristic curves, calibration plots, and decision curve analysis.

RESULTS

Five independent predictors were identified: Age, stroke type, central post-stroke pain, family history of stroke, and depression. The C-statistic (equivalent to the area under the receiver operating characteristic curve) was 0.86 (95% confidence interval: 0.83-0.90) for the training cohort and 0.82 (95% confidence interval: 0.74-0.90) for the split-sample internal validation cohort, indicating a strong ability to identify patients with PSAS. Calibration curves demonstrated a good agreement between predicted and observed risks in both cohorts. Decision curve analysis confirmed the clinical utility of the nomogram.

CONCLUSION

The nomogram demonstrated good discrimination and calibration in a single-center sample. It may help identify patients at elevated risk for current anxiety symptoms, though external validation is needed.

Key Words: Stroke; Anxiety; Nomogram; Independent associated factors; Retrospective cross-sectional study

Core Tip: This retrospective cross-sectional study sought to develop a clinical recognition model for post-stroke anxiety symptoms. The nomogram exhibited robust discrimination and calibration in a single-center sample, indicating its potential to identify patients at high risk for current anxiety symptoms, pending external validation.

INTRODUCTION

Globally, stroke is the second leading cause of mortality worldwide. In 2020, the estimated prevalence, incidence, and mortality rates of stroke in China were 2.6%, 505.2 per 100000 person-years, and 343.4 per 100000 person-years, respectively[1]. It is projected that by 2025, there will be approximately 2.2 million individuals in China living with stroke-related disabilities[2]. Therefore, stroke has become a leading cause of mortality and disability in China, exerting a substantial impact on public health. Post-stroke anxiety (PSA) is a prevalent complication among stroke survivors, often resulting from motor dysfunction and a significant decline in quality of life, which can lead to a sustained decline in quality of life[3,4]. Current studies have shown that the risk factors for PSA include, sleep duration[5,6], blood lipid[7,8], gender, Neurological Institute of Health Stroke Scale (NIHSS) score[7,9], age, hypertension, diabetes, high-density lipoprotein, alcohol consumption[7], and thyroid-stimulating hormone (TSH) level[10]. However, patients with PSA frequently exhibit heightened concern regarding their prognosis, including the risk of stroke recurrence, return to work, falls, and the ability to maintain independence. These concerns are considered typical responses to stroke, rendering the accurate diagnosis of PSA a significant challenge[7].

Nomograms are widely used in medical prediction models across various fields, including oncology, as they are highly accurate and easy to interpret, which assist clinicians in making informed decisions[11-14]. A nomogram serves as a graphical representation of the results derived from regression analysis. It typically begins with a “points” line at the top, which assigns scores to each prognostic factor. Upon integrating and analyzing all included factors, each variable is plotted proportionally[13,15]. Currently, no nomogram is available for identifying post-stroke anxiety symptoms (PSAS).

The objective of this study was to develop such a model. This tool is intended to assist clinicians in early identification of PSAS using readily available clinical factors, thereby facilitating timely intervention.

MATERIALS AND METHODS

Study design and patients

This retrospective cross-sectional study aimed to develop and validate a nomogram for PSAS. A total of 947 stroke patients, who met the inclusion criteria, were consecutively enrolled between March and September 2023 (Figure 1). Participants were randomly allocated into a training set (n = 662, 70%) and an internal validation set (n = 285, 30%). Baseline demographic and clinical characteristics were comparable between the two groups. Model construction, variable selection, and multivariate logistic regression analysis were conducted exclusively within the training set, which included 99 patients with PSAS (positive outcome events).

Figure 1  Flow chart of the study.

Prior to regression analysis, 21 candidate variables - covering demographics, stroke characteristics, laboratory findings, clinical scales, lifestyle factors, and medical history - were examined using univariate analysis. Ultimately, 10 statistically significant variables were incorporated into the final multivariate regression model. The events per variable (EPV) was calculated to be 9.9, which is close to the widely accepted threshold of EPV ≥ 10, and falls within the acceptable range for stable logistic regression and nomogram development.

The validation dataset was solely used to assess the generalizability, discriminative ability and calibration of the developed prediction model. The C-statistic for the training set was 0.86, while that for the validation set was 0.82. The calibration curve demonstrated a high degree of consistency between predicted and observed risks, with a mean absolute error of 0.019. Overall, the sample size of the training set was adequate to support robust multivariate predictive modeling.

Inclusion criteria: (1) According to the 10^th revision of the International Classification of Diseases and Related Health Problems, patients were diagnosed with cerebral infarction or hemorrhage, or had a history of stroke; and (2) Confirmation of hemorrhage or infarction location via computed tomography or magnetic resonance imaging.

Exclusion criteria comprised: (1) A history of psychiatric disorders; (2) An inability to complete the questionnaire due to disturbances in consciousness or language impairment; and (3) Refusal to participate in the study.

Ethical approval was obtained from the appropriate ethics committee. Written informed consent was obtained from all participants or their families, with family members providing consent on behalf of stroke patients who were unable to do so. The initial page of the questionnaire outlined the study's objectives and significance. Participants were also informed of their right to withdraw from the survey at any point.

Evaluations

Each patient underwent the following evaluations: (1) Laboratory analysis: Upon admission, standard blood tests were conducted to assess serum levels of TSH, homocysteine (Hcy), and C-reactive protein (CRP); (2) Stroke-specific assessments: These assessments included evaluations of the modified Rankin Scale and NIHSS scores, as well as determinations of bleeding volume and stroke location. Additionally, the Trial of ORG 10172 in Acute Stroke Treatment (TOAST) subtypes were categorized based on comprehensive imaging and clinical data; and (3) Questionnaire survey: Questionnaires were developed using the Questionnaire Star platform, and professionally trained physicians conducted face-to-face interviews with patients during their hospitalization. Patients provided responses directly. The survey gathered data on a range of variables, including sex, age, height, weight, employment status, marital status, educational attainment, presence of central post-stroke pain, smoking and alcohol consumption habits, past medical history, and family history of psychiatric disorders and stroke. Furthermore, assessments were conducted using the Patient Health Questionnaire-9 (PHQ-9) and Generalized Anxiety Disorder 7-item (GAD-7).

Measurements

Laboratory biomarkers: Fasting blood samples were collected at 6:00 am on the day following admission to determine levels of TSH, CRP, and Hcy. TSH levels were measured using a standardized radioimmunoassay kit, with a reference range of 0.35-4.94 μIU/mL. Hcy levels were assessed through the enzyme cycling method, with a reference range of 0-15 μmol/L. CRP levels were quantitatively measured by immunoturbidimetry, with a detection range of 0-10 mg/L.

Stroke-related clinical assessments: Stroke severity and disability were evaluated using the NIHSS and modified Rankin Scale, two validated stroke rating scales[16]. Ischemic stroke subtypes were classified according to the TOAST criteria, based on imaging and clinical data[17]. All scale scoring and etiological classifications were performed by experienced neurologists.

Psychological scale assessment: Anxiety and depressive symptoms were evaluated using the Chinese versions of the GAD-7 scale and the PHQ-9 scale, respectively[18-21]. Consistent with previous studies, a total score of ≥ 5 on the GAD-7 and PHQ-9 was used to identify anxiety and depression, respectively. The Chinese versions of these scales have been validated for reliability and validity in prior studies. In this investigation, a GAD-7 score of ≥ 5 was employed to indicate significant anxiety symptoms. Although this threshold has high sensitivity for detecting generalized anxiety disorder, it does not equate to a formal clinical diagnosis. Consequently, the outcome variable in our predictive model is anxiety symptoms rather than a diagnosis of PSA disorder.

Statistical analysis

Statistical analyses were conducted using SPSS version 26 and R software. For variables not normally distributed, data were summarized as median (interquartile range). Non-parametric tests were applied to compare differences between the two groups. Categorical data were expressed as frequencies and n (%), and χ² tests were employed to evaluate group differences. Univariate and multivariate logistic regression analyses were used for variable selection and model development. Candidate predictors for the multivariate model were selected based on clinical plausibility and findings from previous literature, in addition to a univariate screening threshold of P < 0.05. Specifically, 21 baseline variables encompassing demographics, stroke characteristics, laboratory indicators, clinical scales, lifestyle factors, and medical history were first examined using univariate logistic regression. Variables with P < 0.05 in the univariate analysis were then entered into a multivariate logistic regression model using forced entry (enter method) to adjust for potential confounding and to identify independent associated factors. This approach ensured that all clinically relevant candidates were considered while controlling for the risk of overfitting through the use of a relatively conservative univariate screening threshold and a final model with an adequate events-per-variable ratio (EPV = 19.8 for the five retained predictors). All variables included in the final multivariate model - namely age, stroke type, post-stroke pain, family history of stroke, depression, and the GAD-7 outcome - had complete data with no missing observations. Missing data were not imputed for baseline covariates that were not retained in the final model, as these variables did not enter the multivariable regression analysis. We comprehensively evaluated three dimensions of model performance: Discrimination, calibration, and clinical net benefit.

Discriminative ability: The discriminative ability of the model was assessed using receiver operating characteristic (ROC) curves and the corresponding concordance index (C-index), which is equivalent to the area under the curve, in both the training and validation datasets. The C-index measures the model’s capability to differentiate between patients who develop PSAS and those who do not, with values ranging from 0.5 (indicating no discrimination) to 1.0 (indicating perfect discrimination).

Calibration performance: Calibration curves, derived from 1000 bootstrap resamples, were constructed to assess the alignment between model-predicted probabilities and the actual observed incidence of PSAS. The fitted curves closely aligned with the ideal 45° reference line, indicating a high level of calibration performance for the model.

Clinical utility: Decision curve analysis (DCA) was conducted to determine the net clinical benefit across various risk thresholds, thereby evaluating the clinical utility of the nomogram in decision-making processes.

RESULTS

Predictors screening via regression analysis

The dataset was randomly partitioned into a training set and a validation set in a 7:3 ratio. The analysis revealed no statistically significant differences in the indicators between the training and validation sets, indicating that the baseline characteristics were well balanced between the two groups (Tables 1 and 2).

Table 1 Comparison of sociodemographic characteristics between the training and validation sets, n (%).

Category	Validation set (n = 285)	Training set (n = 662)	t/Z/χ2	P value
Sex	-	-	1.78	0.182
Male	192 (67.37)	416 (62.84)	-	-
Female	93 (32.63)	246 (37.16)	-	-
Age (years)	66.00 (59.00, 74.00)	66.00 (58.00, 74.00)	-0.50	0.615
Employment status	-	-	6.49	0.039a
Employed	39 (13.68)	101 (15.26)	-	-
Retired	107 (37.54)	193 (29.15)	-	-
Unemployed	139 (48.77)	368 (55.59)	-	-
Educational status	-	-	1.23	0.755
Junior high school and below	225 (78.95)	538 (81.39)	-	-
College or university	21 (7.37)	37 (5.60)	-	-
High school or technical secondary school	39 (13.68)	86 (13.01)	-	-

^aP < 0.05.

Table 2 Comparison of clinical characteristics between the training and validation sets, n (%).

Variables	Validation set (n = 285)	Training set (n = 662)	Z/χ2	P value
NIHSS score	2.00 (1.00, 4.00)	2.00 (1.00, 4.00)	-0.27	0.787
mRS score	2.00 (1.00, 3.00)	2.00 (1.00, 3.00)	-0.26	0.793
Hcy	14.46 (11.18, 17.30)	13.85 (10.73, 17.78)	-0.91	0.363
TSH	2.03 (1.29, 3.36)	1.97 (1.19, 3.05)	-1.19	0.235
CRP	2.73 (1.00, 7.90)	2.70 (1.00, 7.75)	-0.09	0.932
Type of stroke	-	-	2.04	0.154
Ischemic stroke	253 (88.77)	607 (91.69)	-	-
Hemorrhagic stroke	32 (11.23)	55 (8.31)	-	-
TOAST classification	-	-	0.50	0.918
Undetermined etiologies	7 (2.77)	13 (2.15)	-	-
Large vessel atherosclerosis	173 (68.38)	417 (68.93)	-	-
Small vessel occlusion	61 (24.11)	142 (23.47)	-	-
Cardio-embolic source	12 (4.74)	33 (5.45)	-	-
Time since stroke	-	-	0.86	0.652
> 6 months	5 (1.75)	9 (1.36)	-	-
> 7 days ≤ 6 months	155 (54.39)	380 (57.40)	-	-
≤ 7 days	125 (43.86)	273 (41.24)	-	-
Feeling pain after stroke	-	-	0.10	0.754
Yes	25 (8.77)	54 (8.16)	-	-
No	260 (91.23)	608 (91.84)	-	-
Pre-stroke sleep duration	-	-	2.72	0.437
< 5 hours	46 (16.14)	131 (19.79)	-	-
5-6 hours	41 (14.39)	94 (14.20)	-	-
6-7 hours	26 (9.12)	46 (6.95)	-	-
> 7 hours	172 (60.35)	391 (59.06)	-	-
History of hyperlipidemia	-	-	5.79	0.016a
No	262 (91.93)	572 (86.40)	-	-
Yes	23 (8.07)	90 (13.60)	-	-
History of hemicrania	-	-	0.03	0.862
No	281 (98.60)	650 (98.19)	-	-
Yes	4 (1.40)	12 (1.81)	-	-
Family history of stroke	-	-	1.26	0.262
No	279 (97.89)	639 (96.53)	-	-
Yes	6 (2.11)	23 (3.47)	-	-
Depression	-	-	2.22	0.136
No	252 (88.42)	561 (84.74)	-	-
Yes	33 (11.58)	101 (15.26)	-	-
Anxiety	-	-	2.29	0.13
No	251 (88.07)	558 (84.29)	-	-
Yes	34 (11.93)	104 (15.71)	-	-

^aP < 0.05.

NIHSS: Neurological Institute of Health Stroke Scale; mRS: Modified Rankin Scale; Hcy: Homocysteine; TSH: Thyroid-stimulating hormone; CRP: C-reactive protein; TOAST: Trial of ORG 10172 in Acute Stroke Treatment.

Further analysis of the training set data identified statistically significant differences in variables such as age, stroke type, stroke onset time, central post-stroke pain, pre-stroke sleep duration, employment status, history of hyperlipidemia, history of migraine, family history of stroke, and depression (Table 3).

Table 3 Differences between patients with anxiety and those without anxiety within the training set, n (%).

Variables	No anxiety (n = 563)	Anxiety (n = 99)	Z/χ2	P value
NIHSS score	2.00 (1.00, 4.00)	2.00 (1.00, 4.25)	-0.63	0.53
mRS score	2.00 (1.00, 3.00)	2.00 (2.00, 3.00)	-0.69	0.493
Hcy	13.80 (10.70, 17.72)	14.05 (11.25, 18.20)	-0.53	0.597
TSH	2.00 (1.19, 3.07)	1.82 (1.17, 3.01)	-0.66	0.508
CRP	2.70 (1.00, 7.40)	2.76 (1.00, 11.92)	-1.59	0.111
Age (years)	68.00 (59.00, 75.00)	58.50 (52.00, 66.50)	-5.95	0.01b
Type of stroke	-	-	6.60	0.01b
Ischemic stroke	505 (90.50)	102 (98.08)	-	-
Hemorrhagic stroke	53 (9.50)	2 (1.92)	-	-
TOAST classification	-	-	3.19	0.364
Stroke of undetermined etiology	11 (2.19)	2 (1.96)	-	-
Large vessel atherosclerosis	342 (67.99)	75 (73.53)	-	-
Small vessel occlusion	119 (23.66)	23 (22.55)	-	-
Cardio-embolic source	31 (6.16)	2 (1.96)	-	-
Time since stroke	-	-	6.68	0.035a
> 6 months	6 (1.08)	3 (2.88)	-	-
> 7 days ≤ 6 months	331 (59.32)	49 (47.12)	-	-
≤ 7 days	221 (39.61)	52 (50.00)	-	-
Sex	-	-	0.01	0.938
Male	351 (62.90)	65 (62.50)	-	-
Female	207 (37.10)	39 (37.50)	-	-
Employment status	-	-	16.59	0.01b
Employed	72 (12.90)	29 (27.88)	-	-
Retired	172 (30.82)	21 (20.19)	-	-
Unemployed	314 (56.27)	54 (51.92)	-	-
Educational attainment	-	-	1.79	0.409
Junior high school and below	459 (82.26)	79 (76.70)	-	-
College or university	30 (5.38)	7 (6.80)	-	-
High school or technical secondary school	69 (12.37)	17 (16.50)	-	-
Feeling pain after stroke	-	-	23.86	0.01b
Yes	33 (5.91)	21 (20.19)	-	-
No	525 (94.09)	83 (79.81)	-	-
Pre-stroke sleep duration	-	-	12.45	0.006b
< 5 hours	101 (18.10)	30 (28.85)	-	-
5-6 hours	77 (13.80)	17 (16.35)	-	-
6-7 hours	35 (6.27)	11 (10.58)	-	-
> 7 hours	345 (61.83)	46 (44.23)	-	-
History of hyperlipidemia	-	-	9.44	0.002b
No	492 (88.17)	80 (76.92)	-	-
Yes	66 (11.83)	24 (23.08)	-	-
History of migraine	-	-	8.37	0.004b
No	552 (98.92)	98 (94.23)	-	-
Yes	6 (1.08)	6 (5.77)	-	-
Family history of stroke	-	-	21.16	0.01b
No	547 (98.03)	92 (88.46)	-	-
Yes	11 (1.97)	12 (11.54)	-	-
Depression	-	-	102.79	0.01b
No	507 (90.86)	54 (51.92)	-	-
Yes	51 (9.14)	50 (48.08)	-	-

^aP < 0.05.

^bP < 0.01.

NIHSS: Neurological Institute of Health Stroke Scale; mRS: Modified Rankin Scale; Hcy: Homocysteine; TSH: Thyroid-stimulating hormone; CRP: C-reactive protein; TOAST: Trial of ORG 10172 in Acute Stroke Treatment.

We conducted a univariate regression analysis and identified age, stroke type, employment status, sleep duration prior to stroke, central post-stroke pain, history of hyperlipidemia, history of migraine, family history of stroke, and depression as significant factors (P < 0.05) (Table 4).

Table 4 Univariate regression analysis of post-stroke anxiety symptoms.

Variables	β	SE	Z	P value	OR (95%CI)
Age	-0.06	0.01	-5.67	< 0.001c	0.94 (0.93-0.96)
Type of stroke
Ischemic stroke	-	-	-	-	1.00 (reference)
Hemorrhagic stroke	-1.68	0.73	-2.3	0.021	0.19 (0.04-0.78)
Time since stroke
> 6 months	-	-	-	-	1.00 (reference)
> 7 days ≤ 6 months	-1.22	0.72	-1.68	0.093	0.30 (0.07-1.22)
≤ 7 days	-0.75	0.72	-1.04	0.298	0.47 (0.11-1.94)
Employment status
Employed	-	-	-	-	1.00 (reference)
Retired	-1.19	0.32	-3.74	< 0.001c	0.30 (0.16-0.57)
Unemployed	-0.85	0.26	-3.21	0.001c	0.43 (0.25-0.72)
Feeling pain after stroke
Yes	-	-	-	-	1.00 (reference)
No	-1.39	0.3	-4.59	< 0.001c	0.25 (0.14-0.45)
Pre-stroke sleep duration
< 5 hours	-	-	-	-	1.00 (reference)
5-6 hours	-0.3	0.34	-0.87	0.382	0.74 (0.38-1.45)
6-7 hours	0.06	0.4	0.14	0.889	1.06 (0.48-2.33)
> 7 hours	-0.8	0.26	-3.07	0.002b	0.45 (0.27-0.75)
History of hyperlipidemia
No	-	-	-	-	1.00 (reference)
Yes	0.8	0.27	3.01	0.003b	2.24 (1.32-3.77)
History of migraine
No	-	-	-	-	1.00 (reference)
Yes	1.73	0.59	2.94	0.003b	5.63 (1.78-17.82)
Family history of stroke
No	-	-	-	-	1.00 (reference)
Yes	1.87	0.43	4.32	< 0.001c	6.49 (2.78-15.14)
Depression
No	-	-	-	-	1.00 (reference)
Yes	2.22	0.25	9.05	< 0.001c	9.20 (5.69-14.88)

^bP < 0.01.

^cP < 0.001.

All categorical variables were treated as dummy variables in logistic regression models. The reference group for each categorical predictor is marked as 1.00 (reference) within the table rows, and summarized as follows: (1) Type of stroke: Ischemic stroke; (2) Time since stroke: > 6 months; and (3) Employment status: Employed. Feeling pain after stroke: Yes. β: Regression coefficient; SE: Standard error; Z: Statistical quantity; OR: Odds ratio; CI: Confidence interval.

Subsequent multivariate logistic regression (Table 5) identified five independent predictors. Regarding stroke type, with ischemic stroke as the reference, hemorrhagic stroke was significantly associated with lower odds of anxiety symptoms [odds ratio (OR) = 0.12, 95% confidence interval (CI): 0.02-0.57], indicating that ischemic stroke patients had higher odds of anxiety symptoms compared with hemorrhagic stroke patients. With the presence of central post-stroke pain as the reference, the absence of pain was associated with lower odds of anxiety symptoms (OR = 0.21, 95%CI: 0.10-0.44), indicating that patients with central post-stroke pain had higher odds of anxiety symptoms compared with those without pain. The remaining independent predictors were age (OR = 0.92, 95%CI: 0.89-0.95), family history of stroke (OR = 3.47, 95%CI: 1.16-10.37), and depression (OR = 11.03, 95%CI: 6.07-20.07). These predictors were utilized to construct the ROC curve (Figure 2).

Figure 2 Receiver operating characteristic curves for assessing discriminative ability of the nomogram. AUC: Area under the curve; CI: Confidence interval; PPV: Positive predictive value; NPV: Negative predictive value.

Table 5 Multivariate regression analysis of post-stroke anxiety symptoms.

Variables	β	SE	Z	P value	OR (95%CI)
Age	-0.08	0.02	-5.24	< 0.001c	0.92 (0.89-0.95)
Type of stroke
Ischemic stroke	-	-	-	-	1.00 (reference)
Hemorrhagic stroke	-2.15	0.81	-2.66	0.008b	0.12 (0.02-0.57)
Time since stroke
> 6 months	-	-	-	-	1.00 (reference)
> 7 days ≤ 6 months	-1.46	0.94	-1.54	0.122	0.23 (0.04-1.48)
≤ 7 days	-0.85	0.95	-0.89	0.371	0.43 (0.07-2.74)
Employment status
Employed	-	-	-	-	1.00 (reference)
Retired	0.14	0.48	0.29	0.769	1.15 (0.45-2.96)
Unemployed	0.44	0.38	1.15	0.249	1.55 (0.74-3.26)
Central post-stroke pain
Yes	-	-	-	-	1.00 (reference)
No	-1.56	0.38	-4.12	< 0.001c	0.21 (0.10-0.44)
Pre-stroke sleep duration
< 5 hours	-	-	-	-	1.00 (reference)
5-6 hours	-0.32	0.42	-0.75	0.452	0.73 (0.32-1.67)
6-7 hours	-0.04	0.53	-0.07	0.947	0.97 (0.34-2.73)
> 7 hours	-0.49	0.33	-1.46	0.143	0.62 (0.32-1.18)
History of hyperlipidemia
No	-	-	-	-	1.00 (reference)
Yes	0.6	0.34	1.8	0.071	1.83 (0.95-3.53)
History of migraine
No	-	-	-	-	1.00 (reference)
Yes	0.21	0.74	0.28	0.777	1.23 (0.29-5.21)
Family history of stroke
No	-	-	-	-	1.00 (reference)
Yes	1.24	0.56	2.22	0.026a	3.47 (1.16-10.37)
Depression
No	-	-	-	-	1.00 (reference)
Yes	2.4	0.31	7.86	< 0.001c	11.03 (6.07-20.07)

^aP < 0.05.

^bP < 0.01.

^cP < 0.001.

All categorical variables were treated as dummy variables in logistic regression models. The reference group for each categorical predictor is marked as 1.00 (reference) within the table rows, and summarized as follows: (1) Type of stroke: Ischemic stroke; (2) Time since stroke: > 6 months; and (3) Employment status: Employed. Feeling pain after stroke: Yes. β: Regression coefficient; SE: Standard error; Z: Statistical quantity; OR: Odds ratio; CI: Confidence interval.

Using these independent associated factors, we developed a nomogram for the identification of PSAS (Figure 3).

Figure 3 Nomograph for identifying post-stroke anxiety symptoms. The nomogram incorporates five independent associated factors: Age, stroke type (ischemic vs hemorrhagic), central post-stroke pain (yes vs no), family history of stroke (yes vs no), and depression (Patient Health Questionnaire-9 ≥ 5). To utilize the nomogram, the points corresponding to each factor are summed to derive the total score, which is then mapped onto the bottom scale to estimate the probability of current anxiety symptoms (Generalized Anxiety Disorder-7 ≥ 5).

Discriminative performance of the nomogram

The dataset was randomly divided into a training set (n = 662) and a validation set (n = 285). ROC curves were plotted to calculate the C-index, which was used to evaluate the model’s discriminative capacity (Figure 2).

In the training cohort, the C-index was 0.86 (95%CI: 0.83-0.90), while in the validation cohort, it was 0.82 (95%CI: 0.74-0.90). These favorable C-index values suggest that the nomogram possesses strong discriminative capability for identifying patients with PSAS.

Calibration performance of the nomogram

Calibration curves were constructed using 1000 bootstrap resamples in both the training and validation cohorts (Figure 4). These curves demonstrated a high degree of consistency with the ideal 45° reference line, with a mean absolute error of 0.019 in the training set. These findings confirm that the nomogram exhibits satisfactory calibration performance, with predicted risks aligning closely with actual clinical outcomes.

Figure 4 Calibration curve evaluating the consistency between predicted and observed post-stroke anxiety symptoms probabilities. The dashed diagonal line denotes the ideal reference line, indicating perfect concordance between predicted and observed probabilities. In contrast, the solid line illustrates the bias-corrected calibration curve, which is obtained through bootstrap resampling with 1000 iterations. A closer alignment with the dashed line signifies superior calibration. A: Training set; B: Validation set.

Clinical utility evaluation via DCA

DCA was performed to assess the clinical application value of the nomogram, and the results of the training and validation sets are shown in Figure 5. Across a wide range of risk thresholds, the nomogram provided greater net clinical benefit than the “treat all” and “treat none” strategies. The results of the DCA corroborated that the prediction model consistently provided a stable positive net benefit within the threshold probability range of 0.75 to 0.95, thereby demonstrating significant clinical applicability within this interval.

Figure 5 Decision curve analysis of the nomogram. A: Training set; B: Validation set. The Y-axis depicts net benefit, while the X-axis denotes threshold probability. The blue curve represents the nomogram, whereas the gray descending line illustrates the “treat all” strategy. The gray horizontal line at a net benefit of zero symbolizes the “treat none” strategy.

DISCUSSION

Application potential of nomogram prediction model

Nomograms are now widely used in medical prediction models, particularly in oncology. Their primary advantages include high accuracy and easily interpretable results, which assist clinicians in making more informed clinical decisions[11-14]. A nomogram is essentially a visual representation of regression analysis outcomes. It begins with a “points” line at the top, representing scores for each prognostic factor. After integrating and analyzing all included factors, each variable is proportionally plotted[13,15]. However, there are currently no nomogram models available for PSAS.

Predictive performance of the established nomogram

We performed two types of validation: Discrimination and calibration. Firstly, the C-index values exceeding 0.8 in both the training and validation cohorts validated the model's robust discriminative capacity, indicating its ability to distinguish between high- and low-risk patients for PSAS. Secondly, the calibration curves indicated a strong concordance between predicted and observed event probabilities, which is crucial for clinical risk assessment, as it ensures that clinicians can rely on the risk estimates provided by this nomogram to guide intervention decisions. DCA further supported these findings, demonstrating the model's clinical utility beyond statistical performance.

Analysis of independent influencing factors for PSAS

Our research identified associations between PSAS and factors such as age, stroke type, central post-stroke pain, family history of stroke, and depression. Notably, we observed a negative correlation between PSAS and age, aligning with previous studies on anxiety and aging[22,23]. This correlation may be attributed to a decreased sensitivity to negative emotional stimuli with advancing age[22,24], and a reduced encoding of negative emotions in the middle temporal gyrus[22]. Additionally, younger patients may experience post-stroke impairments more acutely due to their active professional and social lives. However, some studies have reported no significant association between age and PSA[7,25], which could be due to a lower prevalence of anxiety in older adults or the increased risk of stroke obscuring anxiety rates.

Our sample, which covered a wide age range (20-91 years), minimized age-related bias and highlighted the need to pay special attention to younger stroke patients. Additionally, this study established a correlation between PSAS and stroke type, revealing that patients with ischemic stroke exhibit a higher risk of anxiety. This contrasts with hemorrhagic stroke patients, who often experience superior functional recovery[26], potentially alleviating anxiety. In contrast, ischemic stroke patients encounter extended challenges in functional adaptation. Previous research has not identified a connection between PSA and stroke type[25,27]. However, a study conducted in Africa indicated that patients with hemorrhagic stroke were more susceptible to PSA[28]. These discrepancies may be due to differences in sample size, variable selection, or ethnic background. Consequently, further research is warranted.

Furthermore, central post-stroke pain has been linked to PSAS, consistent with previous research findings[29,30]. This association may arise from patients' lack of understanding regarding the etiology of their pain[31], as some patients may perceive pain as a sign of worsening, leading to fear, anxiety, catastrophizing, and other maladaptive cognitive responses[30]. Additionally, hemorrhagic strokes occurring in the thalamic region may activate glial cells and inflammatory cytokines via the hypoxiainducible factor1α/NODlike receptor family pyrin domaincontaining 3 signaling pathway, leading to pain and comorbid anxiety[32]. Early psychological intervention for central post-stroke pain is therefore crucial. This study indicates that a family history of stroke correlates with PSAS, potentially due to the increased incidence, mortality, and recurrence rates associated with stroke. These factors impose a significant burden on caregivers of stroke patients, contributing to heightened levels of anxiety and depression among them[33]. Such individuals may also encounter greater socioeconomic pressures, further increasing their vulnerability to anxiety.

This study identified a significant association between depression and PSAS, aligning with previous research findings[9,34]. Numerous studies have demonstrated that depression and anxiety may share common pathophysiological mechanisms, such as neuroinflammation[35-37], dysregulated hypothalamic-pituitary-adrenal axis signaling[38] and the microbiota-gut-brain axis[39-41]. Consequently, it is imperative to address the comorbidity of PSA and post-stroke depression, and to provide timely and appropriate treatment.

Clinical use of the nomogram

This model can identify patients at high risk of PSAS early, which is essential for the implementation of personalized treatment plans. In addition, the model enhances patient management by stratifying patients according to their risk level, ensuring that patients with the highest risk receive the most intensive management and follow-up. Further research is required to ascertain whether the routine application of this nomogram can enhance clinical decision-making and patient outcomes.

Clinical application orientation

Given that the outcome was determined using a screening scale rather than a structured diagnostic interview, this nomogram should be employed for preliminary risk stratification and to identify patients who may require further psychiatric evaluation. It is not intended to substitute for a clinical diagnosis of PSA disorder. For patients identified as high risk by the nomogram, a comprehensive psychiatric assessment, such as the Structured Clinical Interview for Diagnostic and Statistical Manual of Mental Disorders, Fifth Edition or a clinical interview, is recommended.

Limitations

Several limitations of this study should be acknowledged. Firstly, PSA was defined based on the cutoff value of the GAD-7 scale, which only screens for mild to moderate anxiety symptoms rather than providing a confirmed clinical psychiatric diagnosis. Secondly, all psychological scale assessments (PHQ-9 and GAD-7) were conducted at a single time point during hospitalization. Due to the overlapping symptom items between the two scales, the observed correlation between depressive and anxiety symptoms primarily reflects concurrent psychiatric comorbidity, rather than indicating a definitive temporal predictive relationship for the onset of anxiety following discharge. Consequently, this model is limited to screening in-hospital patients for current anxiety symptoms and lacks the capability to accurately predict the future onset of anxiety during long-term follow-up. To establish more reliable prognostic models, longitudinal cohort studies with distinct baseline exposure assessment and delayed outcome observation are necessary. Thirdly, several advanced internal validation metrics, such as the Brier score and bootstrapped calibration slope, were not calculated in this study. Although the minor difference in the C-index between training and validation sets, along with adequate EPV, ensured basic model stability, further prospective multicenter external validation incorporating comprehensive statistical indicators is essential to enhance the clinical applicability and generalizability of the nomogram. Fourth, the original statistical code and individual-level predicted probabilities for the validation cohort were lost due to an unforeseen system failure. Rebuilding the model from scratch would alter the random training/validation split and model coefficients, which would compromise the validity of the primary results. Consequently, performance metrics derived from the confusion matrix, such as sensitivity, specificity, positive predictive value, and negative predictive value at a fixed cut-off, cannot be reported for the validation cohort. The model's performance is therefore assessed using threshold-independent measures, including the area under the curve, calibration curves, and DCA, which remain robust despite this limitation. Fifth, the absence of true external validation represents the primary limitation of this study. Future multicenter or registry-based studies are needed to externally validate our nomogram before clinical implementation. Furthermore, we did not stratify stroke patients by time period. Future studies with larger sample sizes, particularly in the chronic phase (greater than six months), should conduct formal stratified or interaction analyses to verify the model's consistency across all post-stroke phases.

This study lays the groundwork for further investigations, and we anticipate that our findings will inspire researchers to explore this critical question within prospective, multicenter cohorts.

CONCLUSION

Our nomogram, based on five readily available clinical predictors - age, stroke type, central post-stroke pain, family history of stroke, and depression - showed good discrimination and calibration in this single-center retrospective cross-sectional cohort. This tool may help clinicians identify stroke patients at higher risk of current anxiety symptoms. Nonetheless, external validation in independent cohorts and prospective studies is needed to confirm its generalizability and assess its impact on clinical outcomes.

ACKNOWLEDGEMENTS

The authors wish to thank all participants for their contributions to the study.