Development and validation of a risk assessment model for fear of falling in elderly stroke patients

Abstract

BACKGROUND

Fear of falling (FOF) is a common psychological complication after stroke, yet simple assessment tools based on inpatient clinical data remain limited.

AIM

To develop and validate a nomogram risk assessment model for FOF in elderly stroke patients using clinical characteristics collected during hospitalization.

METHODS

This cross-sectional study included elderly stroke patients admitted to Hejiang People’s Hospital from January 2021 to December 2025. Based on admission date, patients were assigned into a training set (n = 197, 20212023) and a temporal validation set (n = 101, 20242025). Least absolute shrinkage and selection operator regression was used for variable selection. Independent risk factors were identified by multivariate logistic regression, and a nomogram was constructed. Model performance was evaluated using receiver operating characteristic curves, calibration curves, and decision curve analysis (DCA).

RESULTS

Multivariate analysis identified seven independent risk factors for FOF: History of falls in the past year [odds ratio (OR) = 6.156, 95% confidence interval (CI): 2.275-16.659], use of walking aids (OR = 5.361, 95%CI: 2.198-13.076), Fall Risk Questionnaire score (OR = 1.353, 95%CI: 1.046-1.751), National Institutes of Health Stroke Scale score (OR = 1.480, 95%CI: 1.214-1.805), Barthel Index (OR = 0.953, 95%CI: 0.923-0.983), Five Times Sit-to-Stand Test (OR = 1.210, 95%CI: 1.061-1.379), and Four-Stage Balance Test (OR = 0.424, 95%CI: 0.280-0.642). The nomogram achieved an area under the curve (AUC) of 0.880 (95%CI: 0.834-0.927) in the training set, and an AUC of 0.889 (95%CI: 0.836-0.962) in the validation set. Tenfold crossvalidation yielded a mean AUC of 0.848 ± 0.077. Calibration curves showed good agreement, and DCA indicated a clinical net benefit.

CONCLUSION

This risk assessment model incorporates seven clinically accessible indicators, performs well in both cohorts, and can support early identification and individualized intervention of FOF in elderly stroke patients.

Key Words: Stroke; Elderly; Fear of falling; Risk assessment model; Nomogram

Core Tip: This cross-sectional study constructed and validated a risk assessment nomogram for fear of falling in elderly stroke patients based on seven readily available clinical variables including history of falls, walking aid use, Fall Risk Questionnaire score, National Institutes of Health Stroke Scale score, Barthel Index, Five-Times Sit-to-Stand Test, and Four-Stage Balance Test. The model demonstrated good discrimination ability with a training area under the curve (AUC) of 0.880 and a validation AUC of 0.889, and a 10-fold cross-validated AUC of 0.848 ± 0.077. It provides a practical tool for early identification of high-risk individuals.

INTRODUCTION

Stroke is one of the most dangerous diseases threatening human life and health worldwide. With over 12 million new cases annually, it is currently the second leading cause of mortality and the third most common cause of disability[1]. The aging population and changing lifestyles in China contribute to the country’s increasing stroke burden. The elderly are most frequently affected, with an estimated 1.5 million new cases occurring each year[2,3]. Survivors frequently have varied degrees of physical disability, despite the fact that medical treatment advancements have greatly increased survival rates[4]. These patients require long-term rehabilitation, which places substantial burdens on families and society.

Functional impairments following a stroke, including motor, sensory, and cognitive deficits, are common consequences of the disease. Between 40% and 70% of stroke patients fall within the first year after stroke, which can have serious repercussions such as fractures, reduced mobility, and increased mortality[5]. In addition to causing physical harm, falls frequently result in the debilitating psychological disorder known as fear of falling (FOF). FOF is a persistent fear that prevents people from engaging in activities that they are still capable of performing[6]. FOF affects 40% to 60% of stroke patients and may trigger a vicious cycle, in which fear leads to activity restriction, muscle weakness, balance deterioration, and further physical decline[7]. This cycle severely hinders the rehabilitation process, reduces quality of life, and adds to the burden on caregivers. Although the incidence of FOF is high and its consequences are far-reaching, healthcare often overlooks it while prioritizing physical recovery.

However, most existing studies on FOF in stroke patients have focused on chronic populations or used cross-sectional designs to identify associated factors[8,9]. There are currently few easy-to-use and efficient methods for identifying hospitalized patients who are at high risk of FOF based on readily available clinical characteristics. The fear-avoidance cycle may be broken by early detection of high-risk individuals, which would allow for targeted psychological and rehabilitative interventions. Accordingly, the aim of this study was to develop and validate a risk assessment model for FOF in elderly stroke patients using clinical information accessible during hospitalization. By evaluating patients’ fear related to daily activities, we sought to provide clinicians with a practical tool to identify at-risk patients and facilitate tailored intervention.

MATERIALS AND METHODS

Study design

Elderly stroke patients admitted to Hejiang People’s Hospital between January 2021 and December 2025 were enrolled in this cross-sectional, single-center study. Initially, 315 patients were screened. Following the application of the inclusion and exclusion criteria, 298 patients comprised the final sample, with 10 patients excluded for not meeting the inclusion criteria and 7 for meeting the exclusion criteria. For temporal validation, patients were divided chronologically by admission date: Those admitted between January 2021 and December 2023 constituted the training set (n = 197), and those admitted between January 2024 and December 2025 constituted the temporal validation set (n = 101). Among the training set, 124 had high FOF and 73 had low FOF. In the temporal validation set, 61 had high FOF and 40 had low FOF (Figure 1). The Hejiang People’s Hospital Ethics Committee approved the study (Ethics No. 202536). Every procedure was carried out in compliance with the Declaration of Helsinki's ethical standards.

Figure 1 Flowchart of patient selection. FES-I: Falls Efficacy Scale-International.

Sample size calculation

According to the events-per-variable rule, a minimum of 10 outcome events per candidate predictor is recommended for logistic regression model development as suggested by Peduzzi et al[10]. With 12 candidate variables, at least 120 high FOF events were required. For least absolute shrinkage and selection operator (LASSO) regression specifically, simulation studies have shown that stable predictions can be achieved with an events per variable (EPV) as low as 5 in the presence of noise predictors, as demonstrated by Pavlou et al[11]. The training set contained 197 patients, of whom 124 had high FOF, yielding an EPV of approximately 10.3 with 124 events divided by 12 variables. This exceeds both the conventional EPV threshold of 10 and the LASSO-specific recommendation of 5, confirming that the sample size was adequate for model development and internal validation

Diagnostic criteria for stroke

According to the “Chinese Guidelines for Endovascular Treatment of Acute Ischemic Stroke 2018”[12] and the “Chinese Intracerebral Hemorrhage Diagnosis and Treatment Guidelines 2019”[13], stroke was diagnosed as having an acute onset, with newly developed ischemic or hemorrhagic lesions verified by cranial magnetic resonance imaging or head computed tomography.

Inclusion and exclusion criteria

Inclusion criteria: (1) Age ≥ 65 years; (2) Fulfillment of the diagnostic criteria for ischemic or hemorrhagic stroke; (3) First onset, with a lesion in a single cerebral hemisphere; (4) Completion of the Falls Efficacy Scale-International (FES-I) assessment during hospitalization; and (5) Medical records being at least 90% complete, with no essential missing variables.

Exclusion criteria: (1) Clinical diagnosis of dementia or a score below 10 on the Chinese version of the Mini-Mental State Examination; (2) Hospital stay of less than 72 hours; (3) Presence of severe heart, liver, or kidney dysfunction; (4) Presence of severe mental illness or visual and auditory impairments affecting the scale assessment; and (5) Missing more than 30% of the key assessment records in the medical record.

Data collection

Two uniformly trained researchers collected all data using standardized questionnaires, structured data collection forms, face-to-face interviews, and medical record review to ensure consistency across all assessments. The collected data were categorized as follows.

General demographic data: These included age, sex, body mass index, educational level, marital status, living situation, stroke type, motor deficit side, disease duration, smoking history, drinking history, and chronic disease history (diabetes, hypertension, and hyperlipidemia).

History and behavior related to falls: This category comprised history of falls in the past 1 year, use of walking aids, number of medications during hospitalization, and nocturia frequency. History of falls in the past 1 year was defined as any self-reported unintentional fall to the ground that occurred within the 12 months prior to enrollment. Use of walking aids was defined as regular use of any assistive device such as a cane, walker, or crutch for ambulation during daily activities, recorded as a binary variable (yes/no). Number of medications during hospitalization was recorded as the total count of prescription medications taken by the patient at the time of admission. Nocturia frequency was defined as the number of times that the patient woke up to urinate during the night, categorized as ≤ 3 times per night or > 3 times per night.

Outcome indicator: During the patient’s hospital stay, the rehabilitation therapist guided the patient to complete the FES-I assessment to evaluate their concern about falling during daily activities[14]. This scale consists of 16 items assessing the patient’s concern about falling during various daily activities, including getting dressed, preparing meals, bathing, going up and down stairs, reaching above the head or picking up objects from the ground, walking on slippery or uneven surfaces, and participating in social activities. The total scores range from 16 to 64. Each item is scored from 1 to 4 (1 = not at all concerned, 4 = highly concerned). Based on the established cut-off value for stroke populations, patients with a total score > 28 were classified as having high FOF[15].

Neurological function: During the patient’s hospital stay, we collected the total score on the National Institute of Health Stroke Scale (NIHSS) as assessed by neurologists or stroke nurses[16]. Consciousness, visual fields, facial movement, upper/lower limb movement, ataxia, feeling, speech, dysphasia/neglect, and other items are all assessed by the 15 neurological function items in the NIHSS. Each item is scored on a predefined scale, with 0 representing normal function and 42 representing severe impairment. A higher total score indicates more severe neurological deficit.

Assessment of activities of daily living: We collected the total score of the Barthel Index (BI) evaluated by the responsible nurse during the patient’s hospitalization[17]. Eating, bathing, grooming, clothing, controlling urine and feces, using the restroom, moving from bed to chair, walking on level ground, and stair climbing are included. A higher score indicates a greater capacity for self-care in daily living. The total score runs from 0 to 100.

Special function test: We collected the following functional test results performed by the rehabilitation therapists during the patient’s hospital stay: The Five-Times Sit-to-Stand Test (FTSST) measures the time in seconds required for a patient to stand up and sit down five times, assessing lower limb muscle strength and dynamic balance[18]; the Timed Up & Go Test measures the time in seconds required for a patient to stand up from a chair, walk 3 meters, turn around, return, and sit down, assessing overall mobility and dynamic balance[19]; the Four-Stage Balance Test (FSBT) consists of four progressively more difficult postures—standing on tiptoes with feet together, standing on tiptoes with feet apart, standing with feet together, and standing with feet apart. The highest level (1-4) that the patient can achieve is recorded to assess the patient’s static balance ability[20].

Fall risk assessment: The self-rated Fall Risk Questionnaire (FRQ) is used to assess the fall risk of the patients[21]. This questionnaire contains 12 items and has no subscales. It uses binary scoring. The total score ranges from 0 to 14. A fall risk is indicated when the total score is ≥ 4, and the higher the score, the higher the fall risk for older adults.

Quality control

All data collectors underwent standardized training on study protocols, questionnaire administration, and data entry prior to study initiation. Data were collected using a pre-designed, structured data collection form to minimize missing data and transcription errors. Two researchers independently extracted all data from medical records and interviews, and cross-checked all entries for consistency; any discrepancies were resolved by consensus with a third senior researcher. Cases with > 10% missing data on key variables were excluded from the final analysis.

Statistical analysis

SPSS 27.0 (IBM Corp., Armonk, NY, United States) was used to analyze the data. Normality was assessed using the Shapiro-Wilk test. Normally distributed continuous variables, presented as the mean ± SD, were compared using the independent samples t-test; non-normally distributed variables, presented as median (Q₂₅, Q₇₅), were compared using the Wilcoxon rank-sum test. Categorical variables were compared using the χ² or Fisher’s exact test. For variable selection, univariate analysis was first performed with a threshold of P < 0.05 to identify candidate variables. To avoid premature exclusion of clinically important factors, the key selection step was LASSO regression with 10-fold cross-validation. Variables with non-zero coefficients at the optimal λ were retained. Before performing multivariate logistic regression, we assessed multicollinearity using the variance inflation factor (VIF). A VIF > 5 was considered indicative of collinearity. Subsequently, the selected variables were entered into multivariate logistic regression to identify independent risk factors for high FOF, with odds ratio (OR) and 95% confidence interval (CI) calculated. Only variables with P < 0.05 in the multivariate model were retained in the final model. A nomogram was constructed based on the final model. The internal validity of the final model was assessed using 10-fold crossvalidation on the training set. Model performance was evaluated in both the training and validation cohorts using receiver operating characteristic (ROC) curves, with an area under the curve (AUC) > 0.8 indicating good discrimination. Calibration plots with the Hosmer-Lemeshow test (P > 0.05 indicated good calibration) were used to assess model calibration, along with decision curve analysis (DCA). Internal validation was also performed using bootstrap with 1000 resamples. All tests were two-sided, and P < 0.05 was considered statistically significant. To assess the potential impact of unmeasured confounding on the primary finding, an E-value analysis was performed for the strongest risk factor.

RESULTS

The training and validation cohorts’ baseline characteristics

The baseline characteristics of the training and validation groups are presented in Table 1. No significant differences were observed between the two groups for any of the variables (all P > 0.05), indicating that the temporal split resulted in well-balanced cohorts.

Table 1 Comparison of baseline characteristics between the training and internal test cohorts, n (%).

Characteristic	Training cohort (n = 197)	Validation cohort (n = 101)	Statistic	P value
Age (years, mean ± SD)	74.67 ± 4.49	74.30 ± 4.24	0.6861	0.494
Sex			0.9732	0.324
Male	119 (60.41)	55 (54.46)
Female	78 (39.59)	46 (45.54)
BMI (kg/m2, mean ± SD)	24.54 ± 3.39	25.32 ± 2.97	1.9431	0.053
Educational level			1.6042	0.448
Primary school and below	20 (10.15)	7 (6.93)
Middle school	117 (59.39)	67 (66.34)
College and above	60 (30.46)	27 (26.73)
Marital status			0.9442	0.624
Married	163 (82.74)	87 (86.14)
Divorced	13 (6.60)	4 (3.96)
Widowed	21 (10.66)	10 (9.90)
Living situation			0.0832	0.774
Live alone	42 (21.32)	23 (22.77)
Live with family	155 (78.68)	78 (77.23)
Stroke type			0.1012	0.750
Ischemic	169 (85.79)	88 (87.13)
Hemorrhagic	28 (14.21)	13 (12.87)
Motor deficit side			1.6802	0.432
Left	46 (23.35)	22 (21.78)
Right	45 (22.84)	30 (29.70)
Bilateral	106 (53.81)	49 (48.51)
Disease duration			0.3412	0.559
< 3	142 (72.08)	76 (75.25)
≥ 3	55 (27.92)	25 (24.75)
Smoking history			0.0032	0.955
Yes	65 (32.99)	33 (32.67)
No	132 (67.01)	68 (67.33)
Drinking history			2.3222	0.128
Yes	57 (28.93)	38 (37.62)
No	140 (71.07)	63 (62.38)
Combined hypertension			0.5782	0.447
Yes	93 (47.21)	43 (42.57)
No	104 (52.79)	58 (57.43)
Combined diabetes			0.1002	0.752
Yes	66 (33.50)	32 (31.68)
No	131 (66.50)	69 (68.32)
Combined hyperlipidemia			0.1182	0.732
Yes	77 (36.84)	37 (41.57)
No	132 (63.16)	52 (58.43)
History of falls in the past 1 year			0.1602	0.689
Yes	61 (29.19)	26 (29.21)
No	148 (70.81)	63 (70.79)
Walking aid			0.0002	0.988
Yes	74 (37.56)	40 (39.60)
No	123 (62.44)	61 (60.40)
Number of medications during hospitalization [n, M (Q25, Q75)]	5.00 (4.00, 6.00)	5.00 (4.00, 7.00)	-1.1863	0.236
Nocturia frequency			1.3192	0.251
> 3	35 (17.77)	13 (12.87)
≤ 3	162 (82.23)	88 (87.13)
FRQ [points, M (Q25, Q75)]	5.00 (4.00, 6.00)	5.00 (4.00, 6.00)	-0.1053	0.916
NIHSS [points, M (Q25, Q75)]	9.00 (8.00, 11.00)	9.00 (7.00, 12.00)	-0.9473	0.344
BI (points, mean ± SD)	65.53 ± 13.07	65.98 ± 10.98	-0.2981	0.766
FTSST (seconds, mean ± SD)	17.93 ± 3.36	18.51 ± 3.77	-1.3551	0.177
TUGT (seconds, mean ± SD)	18.79 ± 3.83	19.23 ± 3.73	-0.9491	0.343
FSBT [levels, M (Q25, Q75)]	3.00 (2.00, 3.00)	3.00 (2.00, 3.00)	-0.0313	0.975

¹Independent samples t-test.

²χ² test.

³Mann-Whitney U test.

BMI: Body mass index; FRQ: Fall risk questionnaire; NIHSS: National Institute of Health Stroke Scale; BI: Barthel Index; FTSST: Five-Times Sit-to-Stand Test; TUGT: Timed Up & Go Test; FSBT: Four-Stage Balance Test.

Univariate analysis of factors associated with high FOF in elderly stroke patients

Significant differences in age, combined diabetes, history of falls in the past 1 year, use of walking aids, FRQ score, NIHSS score, BI score, FTSST, and FSBT between the high and low FOF groups were found using the univariate analysis (all P < 0.05). For further selection, these variables were then entered into the LASSO regression analysis (Table 2).

Table 2 Univariate analysis of factors associated with high fear of falling in elderly stroke patients, n (%).

Characteristic	High FOF group (n = 124)	Low FOF group (n = 73)	Statistic	P value
Age (years, mean ± SD)	75.25 ± 4.37	73.68 ± 4.54	2.3971	0.017
Sex			2.3632	0.124
Male	80 (64.52)	39 (53.42)
Female	44 (35.48)	34 (46.58)
BMI (kg/m2, mean ± SD)	24.68 ± 3.53	24.31 ± 3.14	0.7491	0.455
Educational level			3.0732	0.215
Primary school and below	15 (12.10)	5 (6.85)
Middle school	76 (61.29)	41 (56.16)
College and above	33 (26.61)	27 (36.99)
Marital status			3.6962	0.158
Married	98 (79.03)	65 (89.04)
Divorced	9 (7.26)	4 (5.48)
Widowed	17 (13.71)	4 (5.48)
Living situation			0.8532	0.356
Live alone	29 (23.39)	13 (17.81)
Live with family	95 (76.61)	60 (82.19)
Stroke type			1.0072	0.316
Ischemic	104 (83.87)	65 (89.04)
Hemorrhagic	20 (16.13)	8 (10.96)
Motor deficit side			3.0722	0.215
Left	24 (19.35)	22 (30.14)
Right	29 (23.39)	16 (21.92)
Bilateral	71 (57.26)	35 (47.95)
Disease duration			2.0752	0.150
< 3	85 (68.55)	57 (78.08)
≥ 3	39 (31.45)	16 (21.92)
Smoking history			1.6442	0.200
Yes	45 (36.29)	20 (27.40)
No	79 (63.71)	53 (72.60)
Drinking history			1.3012	0.310
Yes	39 (31.45)	18 (24.66)
No	85 (68.55)	55 (75.34)
Combined hypertension			2.6052	0.107
Yes	64 (51.61)	29 (39.73)
No	60 (48.39)	44 (60.27)
Combined diabetes			5.4322	0.020
Yes	49 (39.52)	17 (23.29)
No	75 (60.48)	56 (76.71)
Combined hyperlipidemia			2.7272	0.099
Yes	52 (41.94)	22 (30.14)
No	72 (58.06)	51 (69.86)
History of falls in the past 1 year			10.0912	0.001
Yes	47 (37.90)	12 (16.44)
No	77 (62.10)	61 (83.56)
Walking aid			10.7052	0.001
Yes	56 (45.16)	16 (21.92)
No	68 (54.84)	57 (78.08)
Number of medications during hospitalization [n, M (Q25, Q75)]	5.00 (4.00, 7.00)	5.0 (4.00, 6.00)	-1.7643	0.078
Nocturia frequency			2.3472	0.126
> 3	26 (20.97)	9 (12.33)
≤ 3	98 (79.03)	64 (87.67)
FRQ [points, M (Q25, Q75)]	5.00 (4.00, 6.00)	4.0 (3.00, 5.00)	-3.6653	< 0.001
NIHSS [points, M (Q25, Q75)]	10.00 (8.00, 12.00)	9.0 (8.00, 10.00)	-4.0193	< 0.001
BI (points, mean ± SD)	62.98 ± 11.56	69.86 ± 14.35	-3.4871	< 0.001
FTSST (seconds, mean ± SD)	18.39 ± 3.48	17.14 ± 3.02	2.5671	0.011
TUGT (seconds, mean ± SD)	19.17 ± 3.96	18.16 ± 3.54	1.7891	0.075
FSBT [levels, M (Q25, Q75)]	2.00 (2.00, 3.00)	3.0 (2.00, 4.00)	-3.8983	< 0.001

¹Independent samples t-test.

²χ² test.

³Mann-Whitney U test.

BMI: Body mass index; FRQ: Fall risk questionnaire; NIHSS: National Institute of Health Stroke Scale; BI: Barthel Index; FTSST: Five-Times Sit-to-Stand Test; TUGT: Timed Up & Go Test; FSBT: Four-Stage Balance Test.

LASSO regression for variable selection

The one standard error criterion (λ = 0.0379) was used to determine the optimal tuning value λ, which produced a more economical model. At this λ value, 8 variables with non-zero coefficients were retained, including combined diabetes, history of falls in the past 1 year, use of walking aids, FRQ score, NIHSS score, BI score, FTSST, and FSBT. Figure 2 displays the LASSO coefficient trajectories along with the cross-validation outcomes. Subsequently, the selected variables were entered into the multivariate logistic regression analysis.

Figure 2 Least absolute shrinkage and selection operator regression analysis for variable selection. A: Least absolute shrinkage and selection operator (LASSO) coefficient profiles of the candidate variables; B: Ten-fold cross-validation for tuning parameter selection in the LASSO model. The dotted vertical lines represent the log(λ) values corresponding to the minimum mean squared error (lambda.min) and the one standard error criterion (lambda.1se).

Multivariable logistic regression analysis

Multivariable logistic regression analysis identified seven independent risk factors for high FOF. Before performing the regression, we assessed multicollinearity using VIF, all VIF values were below 2, ranging from 1.015 to 1.173, indicating no significant collinearity among the risk factors. A history of falls within the past year (OR = 6.156, 95%CI: 2.275-16.659, P < 0.001), use of a walking aid (OR = 5.361, 95%CI: 2.198-13.076, P < 0.001), higher FRQ score (OR = 1.353, 95%CI: 1.046-1.751, P = 0.021), higher NIHSS score (OR = 1.480, 95%CI: 1.214-1.805, P < 0.001), and longer FTSST (OR = 1.210, 95%CI: 1.061-1.379, P = 0.004) were significantly associated with an increased risk of FOF. Conversely, higher BI (OR = 0.953, 95%CI: 0.923-0.983, P = 0.003) and higher FSBT level (OR = 0.424, 95%CI: 0.280-0.642, P < 0.001) were protective factors (Table 3).

Table 3 Multivariable logistic regression analysis of factors associated with high fear of falling in elderly stroke patients.

Characteristic	β	SE	Wald χ2	P value	OR	95%CI
History of falls in the past 1 year	1.817	0.508	12.802	< 0.001	6.156	2.275-16.659
Walking aid	1.679	0.455	13.621	< 0.001	5.361	2.198-13.076
FRQ	0.303	0.131	5.304	0.021	1.353	1.046-1.751
NIHSS score	0.392	0.101	14.985	< 0.001	1.480	1.214-1.805
BI	-0.049	0.016	9.134	0.003	0.953	0.923-0.983
FTSST	0.190	0.067	8.124	0.004	1.210	1.061-1.379
FSBT	-0.857	0.211	16.451	< 0.001	0.424	0.280-0.642

For categorical variables, the reference categories were no history of falls (fall history = 0) and no use of walking aid (walker = 0). OR: Odds ratio; CI: Confidence interval; FRQ: Fall risk questionnaire; NIHSS: National Institute of Health Stroke Scale; BI: Barthel Index; FTSST: Five-Times Sit-to-Stand Test; FSBT: Four-Stage Balance Test.

Sensitivity analysis for unmeasured confounding

To assess the potential impact of unmeasured confounding on the main finding, we calculated the E-value for the strongest risk factor, history of falls. For the observed OR of 6.156 (95%CI: 2.275-16.659), the E-value for the point estimate was 11.79, and the E-value for the lower confidence limit was 3.47.

Nomogram development and ROC analysis

To provide a visual aid for evaluating the probability of FOF in older stroke patients, a nomogram combining the seven independent risk factors was created (Figure 3). In the training cohort, the discriminative performance of each individual factor and the combined model was assessed using ROC curve analysis. As shown in Table 4 and Figure 4, among the individual risk factors, NIHSS score had an AUC of 0.670 (95%CI: 0.597-0.743), FSBT had an AUC of 0.660 (95%CI: 0.583-0.736), and BI had an AUC of 0.646 (95%CI: 0.563-0.729). The history of falls in the past year yielded an AUC of 0.607 (95%CI: 0.547-0.668). Walking aid achieved an AUC of 0.616 (95%CI: 0.551-0.681). FTSST produced an AUC of 0.604 (95%CI: 0.523-0.684). FRQ had an AUC of 0.654 (95%CI: 0.575-0.734). The optimal cut-off values, sensitivities, specificities, and Youden indices for each risk factor are presented in Table 4. The combined model demonstrated good discrimination, with an AUC of 0.880 (95%CI: 0.834-0.927), a sensitivity of 0.849, and a specificity of 0.750 at the optimal cut-off value of 0.678.

Figure 3 Nomogram for assessing the risk of fear of falling in elderly stroke patients. FRQ: Fall risk questionnaire; NIHSS: National Institute of Health Stroke Scale; BI: Barthel Index; FTSST: Five-Times Sit-to-Stand Test; FSBT: Four-Stage Balance Test.

Figure 4 Receiver operating characteristic curves of individual risk factors and the combined model for assessing fear of falling in elderly stroke patients. FRQ: Fall risk questionnaire; NIHSS: National Institute of Health Stroke Scale; BI: Barthel Index; FTSST: Five-Times Sit-to-Stand Test; AUC: Area under the curve; FSBT: Four-Stage Balance Test; FPR: False positive rate; TPR: True positive rate.

Table 4 Receiver operating characteristic analysis results for risk factors of fear of falling in elderly stroke patients.

Characteristic	AUC	Sensitivity	Specificity	Youden index	Cut-off value	95%CI
History of falls in the past 1 year	0.607	0.836	0.379	0.215	0.500	0.547-0.668
Walking aid	0.616	0.781	0.452	0.232	0.500	0.551-0.681
FRQ	0.654	0.534	0.686	0.220	4.500	0.575-0.734
NIHSS	0.670	0.959	0.363	0.322	11.500	0.597-0.743
BI	0.646	0.452	0.807	0.259	73.500	0.563-0.729
FTSST	0.604	0.562	0.637	0.199	17.550	0.523-0.684
FSBT	0.660	0.740	0.548	0.288	2.500	0.583-0.736
Total	0.880	0.849	0.750	0.599	0.678	0.834-0.927

FRQ: Fall risk questionnaire; AUC: Area under the curve; CI: Confidence interval; NIHSS: National Institute of Health Stroke Scale; BI: Barthel Index; FTSST: Five-Times Sit-to-Stand Test; FSBT: Four-Stage Balance Test.

Model performance evaluation in the training and validation cohorts

The training cohort (Figure 5A) and validation cohort (Figure 5B) had AUCs of 0.880 (95%CI: 0.834-0.927) and 0.889 (95%CI: 0.836-0.962), respectively. Ten-fold cross-validation on the training set yielded a mean AUC of 0.848 ± 0.077, indicating stable internal performance. Figure 6 displays the calibration curves. The apparent and bias-corrected calibration curves in the training cohort only slightly deviated from the ideal reference line in the moderate-to-high probability range (Figure 6A). In the validation cohort, the apparent and bias-corrected curves also demonstrated good agreement with the ideal line, indicating satisfactory calibration of the model (Figure 6B). Low estimation error was reflected by the Brier score of 0.137 in the training cohort and 0.141 in the validation cohort. DCA assessed the model’s clinical utility. In both the training (Figure 7A) and validation (Figure 7B) cohorts, the model curve consistently demonstrated a higher net benefit than both the treat-all and treat-none methods over a broad range of threshold probabilities from 0 to 0.8. The advantage was particularly evident across low-to-moderate risk thresholds, and the model curve remained above the treat-all line across the entire threshold range, with no negative net benefit observed.

Figure 5 Receiver operating characteristic curves of the training cohort and the validation cohort. A: Training cohort; B: Validation cohort. AUC: Area under the curve; CI: Confidence interval; FPR: False positive rate; TPR: True positive rate.

Figure 6 Calibration curves of the training cohort and the validation cohort. A: Training cohort; B: Validation cohort.

Figure 7 Decision curve analysis of the training cohort and the validation cohort. A: Training cohort; B: Validation cohort.

DISCUSSION

Stroke remains a primary cause of disability and mortality among older adults worldwide. FOF, as a common psychological complication after stroke, significantly affects patients’ activity ability and quality of life[22]. This work developed a FOF risk assessment nomogram by combining multivariate logistic regression with univariate analysis and LASSO regression to screen variables based on a large sample cohort of elderly stroke patients. The clinical utility, calibration, and discrimination of the model were all systematically verified. The analysis revealed that independent risk factors for FOF included a fall within the previous year, reliance on walking aids, elevated NIHSS scores, and longer FTSST durations. In contrast, higher BI scores and increased FSBT levels acted as protective factors. The constructed nomogram demonstrated excellent discriminative ability, achieving an AUC of 0.880 in the training cohort and 0.889 in the validation cohort. Additionally, it demonstrated substantial clinical net benefit and good calibration, offering a potentially useful tool for the early identification and targeted intervention of FOF in older stroke patients, pending external validation.

In this study, a history of falls in the past year emerged as the strongest risk factor for FOF, with affected patients exhibiting a 6.156-fold higher risk compared to those without such a history. An E-value analysis was conducted to assess robustness against unmeasured confounding. The E-value for the point estimate was 11.79, suggesting that an unmeasured confounder would need to be associated with both fall history and high FOF by an OR of at least 11.79 to fully account for the observed effect, a magnitude that is highly unlikely in clinical research[23]. Gonçalves et al[24] reported that in community-dwelling older adults, both fall history and FOF were associated with reduced mobility, and their interaction amplified the effect, implying that past falls entail not only physical decline but also persistent psychological distress. Barceló et al[25] further demonstrated that prior fall frequency independently predicted long-term mortality in hip fracture patients and also acted as a trigger for FOF, suggesting that the effect of fall history on prognosis may be partly mediated through FOF. The experience of falling strengthens catastrophic beliefs about future falls, initiating a vicious cycle of falling, fear, and activity avoidance to functional decline. This cycle is particularly dangerous in stroke patients, who already have neurological deficits, as activity avoidance further accelerates functional deterioration. In clinical practice, stroke patients with a history of falls should undergo systematic psychological assessment before discharge, and cognitive behavioral interventions should be integrated into rehabilitation plans.

The risk of FOF in patients using walking aids was 5.361 times higher than that in non-users, whereas for every 1-point increase in the BI, the risk decreased by 4.7%. These two seemingly contrasting indicators both reflect patients’ capacity to manage daily activities. Although walking aids provide an objective functional compensation, they may also act as a constant reminder of physical limitations, thereby heightening psychological dependence. Mishra et al[26] found in a study of patients with diabetes and end-stage renal disease that walking aid users had significantly higher FOF and lower physical activity levels than non-users, suggesting that assistive devices do not automatically lead to improved psychological security. In contrast, the BI reflects independence in basic activities such as eating, dressing, and toileting. Higher self-care ability means that patients repeatedly gain success in daily life, with positive feedback gradually building into stable self-efficacy. García-Martínez et al[27] found that living alone was a risk factor for FOF, while patients with strong self-care ability maintained good mobility even when living alone, further underscoring the protective effect of the BI. In clinical practice, for patients using walking aids, attention should be given not only to the provision of devices but also to psychological adaptation, helping them form a realistic perception of their own abilities. Meanwhile, active training in activities of daily living should be promoted to gradually lessen psychological dependence on assistive devices.

Higher FRQ score was also identified as an independent risk factor for FOF. For each one-point increase in the FRQ, the odds of high FOF increased by 35.3%. The FRQ is a self-rated scale that reflects an individual’s perceived risk of falling. Patients who consider themselves at higher risk of falling may develop a persistent FOF, even when their actual physical function is relatively well preserved. The close relationship between self-perceived fall risk and FOF has been documented in stroke populations. A cross-sectional study of 302 older stroke patients revealed that FRQ score was significantly associated with FES-I score in structural equation modeling[28]. A systematic review and meta-analysis of eight studies involving 1597 participants also identified female sex, impaired balance, lower mobility, history of falls, walking aid use, anxiety, and depression as significant risk factors for FOF[7]. These findings suggest that subjective fall risk perception is an important driver of FOF and can be readily assessed in routine clinical practice. The FRQ achieved an AUC of 0.654, suggesting that it provides unique discriminative value beyond objective functional measures.

In this study, NIHSS score, FTSST time, and FSBT level collectively described patients’ physical functional status, reflecting neurological function, lower limb executive function, and static balance, respectively. This combination highlighted a core pattern: The worse the physical functional status, the stronger the psychological FOF. The risk of FOF increased by 1.48 times for every 1-point increase in NIHSS score, by 1.21 times for every 1-second increase in FTSST, whereas it decreased by 57.6% for every 1-level increase in FSBT. Krysiak et al[29] conducted a cross-sectional study in post-stroke inpatients, and their results confirmed our findings. The NIHSS score was positively correlated with the FES-I score, while the Tinetti balance and gait test was strongly negatively correlated with the FES-I score, and the BI also exhibited a significant negative correlation with the FES-I score. These results suggest that patients with more severe neurological deficits, poorer balance, and lower self-care ability have higher levels of FOF. The study also found that 61.5% of post-stroke inpatients reported high FOF, despite most being classified as low fall risk, suggesting that psychological fear may precede objective functional assessment, further supporting the need for early psychological intervention even when functional status appears acceptable.

From a mechanistic perspective, neurological deficits directly impair limb control and sensory integration. This decline in basic function is inevitably reflected in FTSST, a comprehensive functional task, while FSBT further uncovers balance reserve under different postural challenges. Min et al[30] demonstrated at the neural level that prefrontal cortex activation is significantly associated with balance confidence, with more difficult standing postures requiring greater prefrontal engagement, suggesting a cognitive-motor integration basis for the relationship between balance function and FOF. Alissa and Westlake[31] provided psychophysiological evidence, confirming through skin conductance levels that balance confidence is significantly associated with FES-I scores, a relationship measurable through physiological indicators. Polesel et al[32] further suggested that simply providing assistive devices without systematic training fails to effectively improve patients’ psychological state, underscoring the central role of functional training in alleviating FOF. When functional indicators are simultaneously impaired, patients develop persistent doubts about their ability to perform activities, which gradually harden into stable FOF. In clinical practice, rehabilitation for stroke patients should combine neurological recovery, lower limb strengthening, and balance training, gradually restoring activity confidence through functional improvement.

The logistic regression model developed in this study incorporated seven clinically accessible risk factors, yielding an AUC of 0.880 in the training cohort and 0.889 in the validation cohort. DCA confirmed a positive clinical net benefit, and the calibration curve showed strong concordance between predicted and observed probabilities. In comparison with the dynamic nomogram developed by Wu et al[33] for assessing fall risk, our model also demonstrated good discriminative performance for FOF. This difference may be attributable to the distinct targets, as Wu et al[33] focused on the objective outcome of falls while our study focused on the psychological state of FOF, which may have a stronger inherent associations with clinical indicators. Moreover, all risk factors in our model are derived from routine clinical assessments, require no specialized equipment, and could be applicable in similar settings, potentially offering clinical utility, although external validation is needed to establish generalizability.

There are several limitations to this study. First, because this was a single-center study, all participants were recruited from one institution, which may have introduced selection bias. Therefore, the present model should be regarded as an exploratory model rather than a definitive prediction tool. To assess its generalizability, multicenter, large-sample prospective studies are required. We intend to conduct an external validation study in at least two independent hospitals using the same inclusion criteria and data collection protocols; in that study, temporal validation by chronological split based on admission date will also be adopted. Second, the predictive performance may have been influenced by the exclusion of potential confounding variables such the patients’ psychological state, including anxiety and depression, the family support system, and adherence to rehabilitation. Owing to the crosssectional design, causality cannot be inferred. It is possible that patients with higher FOF become less active and consequently more dependent on walking aids, and that lower BI, longer FTSST time, lower FSBT level, and a history of falls may be both causes and effects of FOF. Thus, the observed associations may be bidirectional. Additionally, patients who were unable to complete the FESI assessment due to severe cognitive impairment, aphasia, or critical illness were excluded, which may have introduced selection bias and limited the generalizability of our findings to more severely affected stroke patients. Some variables were not captured in full detail, such as walking aid use which was recorded only as binary, and detailed fall history which was not available in the medical records. Future studies should consider using fall diaries or prospective recording to improve accuracy. Additionally, integrating artificial intelligence and wearable device technologies to incorporate real-time activity data could allow dynamic prediction models for real-time FOF monitoring and early warning. Finally, clinical intervention studies based on this model are justified to develop individualized strategies for high-risk patients and to assess the effectiveness of model-guided interventions in reducing FOF incidence and improving patient outcomes.

CONCLUSION

A nomogram for assessing the risk of FOF in elderly stroke patients was successfully developed and validated in this study. The model is based on seven independent risk factors that are readily available in clinical settings. It demonstrates excellent discrimination, good calibration, and significant clinical net benefit. This may provide a useful tool for early clinical screening and individualized intervention, but further external validation is required to establish its generalizability and clinical utility.

ACKNOWLEDGEMENTS

We are grateful to the medical staff of Hejiang People’s Hospital for their support and help. We would also like to thank all the participants in this study.