Abstract
This study aimed to develop a prediction model based on nomograms and support vector machines (SVM) to assess frailty risk in ischemic stroke patients. Clinical information of ischemic stroke patients admitted to our hospital from January 2023 to December 2024 was retrospectively collected. First, independent risk factors associated with frailty in ischemic stroke patients were identified through univariate and multivariate logistic regression analyses. Subsequently, a nomogram was constructed based on regression analysis results and validated using 10-fold cross-validation. Data were divided into training (70%) and validation (30%) sets. A SVM predictive model was constructed on the training set. The predictive performance of both models was compared using receiver operating characteristic curves. This study included 867 ischemic stroke patients, among whom 296 (34.14%) were classified as frail. The entire cohort was randomly divided into a training set (n = 607) and a validation set (n = 260) at a 7:3 ratio. Logistic regression analysis identified hypertension, impaired self-care ability, physical inactivity, reduced prognosis nutrition index, and depressive status as independent risk factors for frailty in ischemic stroke patients (P < .05). The nomogram demonstrated an area under the receiver operating characteristic curve (AUC) of 0.814 for frailty prediction, while the SVM model achieved an AUC of 0.842, indicating superior predictive capability. Further comparison revealed that while the nomogram offers greater intuitiveness and operational convenience in clinical practice, the SVM model demonstrated superior predictive accuracy. Hypertension, impaired self-care ability, physical inactivity, reduced prognosis nutrition index, and depressive states are independent risk factors for frailty in ischemic stroke patients. Both predictive models (nomogram and SVM) demonstrated high predictive accuracy; however, the SVM model outperformed the nomogram in predictive capability, while the latter retains unique advantages due to its simplicity and clinical applicability.
Keywords: frailty, ischemic stroke, nomogram, prediction model, support vector machine
1. Introduction
Ischemic stroke is one of the leading causes of disability and death worldwide.[1] With the acceleration of population aging, its incidence and prevalence have increased annually, making it a major disease posing a serious threat to public health. Extensive research indicates that poststroke patients not only face neurological impairments such as motor and cognitive dysfunction but also frequently experience varying degrees of frailty.[2] Frailty is a syndrome characterized by multisystem functional decline, manifested as reduced capacity to respond to external stressors. It is closely associated with adverse outcomes including falls, rehospitalization, disability, and mortality.[3] In recent years, the association between stroke and frailty has garnered increasing attention, particularly among ischemic stroke patients, where frailty exhibits a high prevalence, significantly impacting rehabilitation outcomes and quality of life.[4]
Frailty is not only a common health issue among the elderly but also, when present in stroke patients, may exacerbate disease progression and rehabilitation burdens. Researchers have noted that stroke survivors who develop frailty often face higher risks of recurrent stroke, experience slower functional recovery, and exhibit poorer adherence to rehabilitation interventions.[5] Additionally, frailty increases nursing costs and consumes healthcare resources. Therefore, early identification of frailty risk factors in ischemic stroke patients and implementation of targeted interventions are crucial for improving clinical outcomes and optimizing healthcare resource allocation.
Currently, frailty risk factors have been explored in some elderly populations and chronic disease patients, including malnutrition, chronic inflammation, comorbidities, and reduced physical activity.[6] However, research in ischemic stroke populations remains limited, particularly regarding systematic risk factor assessments and the development of quantitative prediction tools.[7] In clinical practice, commonly used frailty assessment methods rely heavily on scale tools such as the Fried frailty phenotype (FFP) and the FRAIL scale. However, these tools present limitations when applied to stroke patients: on 1 hand, functional impairments from stroke may mask or exaggerate frailty manifestations; on the other hand, scale-based assessments require additional human resources and time for clinical implementation.[8] Therefore, there is an urgent need to establish a comprehensive predictive model that integrates clinical characteristics, laboratory indicators, and lifestyle factors, providing clinicians with a simple, intuitive, and actionable tool.
Currently, nomograms and support vector machines (SVMs) represent 2 widely used and effective predictive models.[9] Nomograms integrate multiple risk factors to provide intuitive, personalized risk assessments for patients, demonstrating high practical value in clinical decision-making.[10] SVMs, an advanced machine learning approach, capture complex nonlinear relationships by learning patterns from large datasets and have achieved significant results across multiple biomedical fields.[11]
Based on this, the present study utilized ischemic stroke patients admitted to our hospital from January 2023 to December 2024. Through logistic regression analysis, to identify independent risk factors associated with frailty in ischemic stroke patients. Based on these factors, we constructed a nomogram and a SVM model to predict frailty risk. Additionally, we compared the predictive performance of these 2 models to evaluate their potential and practical value in clinical applications.
2. Materials and methods
2.1. Data sources and collection
This retrospective study systematically collected and organized clinical data from patients hospitalized for ischemic stroke at our hospital between January 2023 and December 2024. The research design involved retrospective case analysis, with all data derived from authentic inpatient medical records, thereby comprehensively reflecting the actual clinical status of this population. The research protocol strictly adhered to relevant medical research standards and fully complied with medical ethics requirements.
2.2. Inclusion and exclusion criteria
Inclusion criteria: Patients with a confirmed diagnosis of ischemic stroke based on the diagnostic principles outlined in the “Chinese Guidelines for Secondary Prevention of Ischemic Stroke and Transient Ischemic Attack (2014 Edition)”; patients were conscious and possessed basic cognitive and communication abilities at admission. Exclusion criteria: individuals with severe concomitant diseases of major organs (heart, brain, lungs, liver, kidneys, etc), or those with psychiatric disorders or malignancies; Subjects unable to complete questionnaires or assessments due to severe visual, auditory, or language impairments; patients with incomplete data.
2.3. Collection of relevant variables
This study systematically collected multidimensional clinical and sociological data from ischemic stroke patients, covering: gender, age, body mass index (BMI), smoking and drinking habits, history of hypertension, diabetes status, educational attainment, marital status, daily labor patterns, prognostic nutritional index (PNI), exercise habits, depressive status, and activities of daily living capabilities. PNI was calculated as the sum of serum albumin concentration (g/L) and 5 times the total peripheral blood lymphocyte count (×109/L), serving as a composite indicator of nutritional and immune status.
The patient’s level of independence in daily living activities was assessed using the Barthel Index (BI). This 10-item scale evaluates abilities including eating, bathing, grooming, dressing, bowel and bladder control, toileting, transferring between bed and chair, walking, and stair climbing, comprehensively reflecting basic functional capacity. The total score ranges from 0 to 100, with higher scores indicating greater independence. Scores are categorized as follows: 100 indicates complete independence; 61 to 99 denotes mild dependence; 41 to 60 signifies moderate dependence; and ≤40 indicates severe dependence.
Concurrently, this study assessed patients’ mental and psychological status using the Geriatric Depression Scale-15 (GDS-15) as the evaluation tool. This scale comprises 15 items, each scored as 1 point, yielding a total score ranging from 0 to 15. Higher scores indicate more pronounced depressive symptoms. Following established literature and clinical practice standards, this study defined a GDS-15 total score ≥5 as the cutoff for depressive symptoms.
2.4. Outcome definition
Frailty was assessed using the FFP scale, which is one of the most widely used tools for frailty evaluation. The FFP includes 5 components: unintentional weight loss, exhaustion, low physical activity, slow walking speed, and weak grip strength. Each component is scored as 0 (absent) or 1 (present), resulting in a total score ranging from 0 to 5.
According to the standard criteria, individuals with 3 or more positive components (score ≥ 3) are classified as frail, those with 1 or 2 components (score 1–2) are considered pre-frail, and those with no components (score = 0) are regarded as robust.
In this study, participants with FFP scores ≥ 3 were defined as the frailty group, while those with scores < 3 were classified as the non-frailty group for subsequent analyses.
2.5. Statistical analysis
All statistical analyses were performed using IBM SPSS 27.0 software. Continuous variables are expressed as mean ± standard deviation (x ± s), while categorical variables are presented as frequencies and percentages. Comparisons between 2 groups of continuous variables were conducted using the independent samples t-test, and differences between categorical variables were analyzed using the chi-square (χ2) test. To identify risk factors associated with frailty in ischemic stroke patients, univariate logistic regression analysis was first performed. Variables with P < .05 in univariate analysis were further evaluated in multivariate logistic regression. The predictive accuracy of each variable for frailty risk was assessed using receiver operating characteristic (ROC) curves. The optimal cutoff value was determined using the Youden index (sensitivity + specificity − 1), which identifies the point that maximizes the combined sensitivity and specificity. A P value < .05 was considered statistically significant.
2.6. Nomogram construction
Statistically significant risk factors identified through logistic regression analysis were incorporated into nomogram development. The nomogram was validated using 10-fold cross-validation, and its average area under the curve (AUC) was calculated. Nomogram development and validation were performed using R software (version 4.21).
2.7. Support vector machine (SVM) model
A linear kernel function was employed to construct the SVM model. A total of 867 patients were randomly divided into a training cohort (n = 607) and a validation cohort (n = 260) at a ratio of 7:3. Both cohorts contained frail and non-frail patients. The training set was used for model development, while the validation set was used for independent performance evaluation. Using the training set data, important features were selected via recursive feature elimination (RFE). All input feature data were recorded in “xlsx” or “txt” formats and uploaded for processing within the SVM model. Model construction and coding were performed using Python version 2.7 and the open-source data mining tool Scikit-Learn 0.20.1. SVM accuracy was expressed as the proportion of cases correctly predicted by the model.
2.8. Model performance evaluation
Model discrimination was evaluated using the area under the receiver operating characteristic curve (AUC). Calibration performance was assessed using calibration plots generated by 1000 bootstrap resamples and the Hosmer–Lemeshow goodness-of-fit test, with good calibration defined as no significant difference between predicted and observed probabilities (P > .05). Clinical utility was evaluated using decision curve analysis, which quantifies the net benefit of the prediction models across a range of threshold probabilities. Clinically interpretable classification metrics, including sensitivity, specificity, positive predictive value (PPV), and negative predictive value (NPV), were calculated at the optimal cutoff determined by the Youden index.
Furthermore, the areas under the ROC curves (AUCs) of the nomogram and SVM models were statistically compared using DeLong’s test for correlated ROC curves. A 2-sided P value < .05 was considered statistically significant.
2.9. Missing data, sample size, and bias control
Missing data were assessed for all variables before analysis. The overall proportion of missing values was <5% for all variables. Multiple imputation using chained equations was applied to handle missing data, assuming data were missing at random. Five imputed datasets were generated, and pooled estimates were used for final analyses.
Regarding sample size, all eligible patients who met the inclusion criteria during the study period were consecutively enrolled. Given the retrospective design, a formal a priori sample size calculation was not performed. However, the final sample size of 867 patients, including 296 frail cases, provided sufficient events per variable for multivariate logistic regression and predictive model development.
To minimize potential bias, several strategies were implemented. First, consecutive patient inclusion was adopted to reduce selection bias. Second, standardized and validated instruments (BI, GDS-15, and FFP) were used to ensure measurement reliability. Third, multivariate regression analysis was performed to adjust for potential confounders. Finally, model validation was conducted using an independent validation cohort and cross-validation procedures to reduce overfitting and enhance model generalizability.
3. Results
3.1. Patient characteristics
This study included 867 hospitalized patients diagnosed with ischemic stroke. Among the 867 patients, 296 were classified as frail and 571 as non-frail, with an overall frailty prevalence of 34.14%. Baseline characteristics for both the modeling group (training cohort, N = 607) and the validation group (N = 260) are shown in Table 1. Statistical analysis revealed no significant differences between the 2 groups, as indicated in the table. The patient selection and screening process is illustrated in Figure 1.
Table 1.
Baseline characteristics of the training and validation cohorts.
| Variables | Training cohort (N = 607) | Validation cohort (N = 260) | P |
|---|---|---|---|
| Gender | .923 | ||
| Male | 294 | 125 | |
| Female | 313 | 135 | |
| Age | .671 | ||
| ≤65 yr | 172 | 70 | |
| >65 yr | 435 | 190 | |
| BMI | .881 | ||
| 18.5 ≤ BMI ≤ 24 kg/m2 | 389 | 168 | |
| Other | 218 | 92 | |
| Education level | .374 | ||
| Junior high school or below | 258 | 119 | |
| Senior high school or above | 349 | 141 | |
| Marital status | .336 | ||
| Married | 436 | 195 | |
| Others | 171 | 65 | |
| Type of work | .562 | ||
| Manual labor | 244 | 110 | |
| Mental labor | 363 | 150 | |
| Smoking | .335 | ||
| Yes | 138 | 67 | |
| No | 469 | 193 | |
| Drinking | .483 | ||
| Yes | 177 | 82 | |
| No | 430 | 178 | |
| Hypertension | .409 | ||
| Yes | 320 | 145 | |
| No | 287 | 115 | |
| Diabetes | .574 | ||
| Yes | 189 | 86 | |
| No | 418 | 174 | |
| Physical exercise | .394 | ||
| Yes | 289 | 132 | |
| No | 318 | 128 | |
| PNI | .591 | ||
| <45 | 162 | 74 | |
| ≥45 | 445 | 186 | |
| Depression | .540 | ||
| Yes | 234 | 106 | |
| No | 373 | 154 | |
| Self-care ability | .454 | ||
| ≤60 | 142 | 67 | |
| >60 | 465 | 193 |
BMI = body mass index, PNI = prognosis nutrition index.
Figure 1.
Flowchart of patient selection and model development. DCA = decision curve analysis, ROC = receiver operating characteristic, SVM = support vector machines.
3.2. Univariate and multivariate logistic regression analysis
Univariate and multivariate logistic regression analyses were performed on all 867 patients. Univariate logistic regression was first conducted on 14 candidate variables. The results indicated that 8 indicators were closely associated with the occurrence of frailty in ischemic stroke patients and may serve as potential risk factors. These included age, smoking, drinking, hypertension, physical exercise, PNI, depression, and self-care ability (see Table 2). Subsequently, statistically significant variables from the univariate analysis were incorporated into a multivariate logistic regression model. Results indicated that hypertension, poor self-care ability, lack of exercise, reduced PNI, and depressive status constitute independent risk factors for frailty in ischemic stroke patients (see Table 3).
Table 2.
Univariate analysis of frailty in patients with ischemic stroke.
| Variables | Frailty group (N = 207) | Control group (N = 400) | P |
|---|---|---|---|
| Gender | .699 | ||
| Male | 98 | 196 | |
| Female | 109 | 204 | |
| Age | .027 | ||
| ≤65 yr | 47 | 125 | |
| >65 yr | 160 | 275 | |
| BMI | .095 | ||
| 18.5 ≤ BMI ≤ 24 kg/m2 | 142 | 247 | |
| Other | 65 | 153 | |
| Education level | .165 | ||
| Junior high school or below | 96 | 162 | |
| Senior high school or above | 111 | 238 | |
| Marital status | .164 | ||
| Married | 156 | 280 | |
| Others | 51 | 120 | |
| Type of work | .754 | ||
| Manual labor | 85 | 159 | |
| Mental labor | 122 | 241 | |
| Smoking | .042 | ||
| Yes | 57 | 81 | |
| No | 150 | 319 | |
| Drinking | .028 | ||
| Yes | 72 | 105 | |
| No | 135 | 295 | |
| Hypertension | .023 | ||
| Yes | 122 | 197 | |
| No | 85 | 203 | |
| Diabetes | .775 | ||
| Yes | 66 | 123 | |
| No | 141 | 277 | |
| Physical exercise | <.001 | ||
| Yes | 73 | 216 | |
| No | 134 | 184 | |
| PNI | .004 | ||
| <45 | 70 | 92 | |
| ≥45 | 137 | 308 | |
| Depression | .001 | ||
| Yes | 99 | 135 | |
| No | 108 | 265 | |
| Self-care ability | .006 | ||
| ≤60 | 62 | 80 | |
| >60 | 145 | 320 |
BMI = body mass index, PNI = prognosis nutrition index.
Table 3.
Multivariate analysis of frailty in patients with ischemic stroke.
| Factors | P | OR | 95% CI |
|---|---|---|---|
| Hypertension | .045 | 1.412 | 1.125–2.288 |
| Self-care ability ≤ 60 | <.001 | 2.499 | 1.266–3.352 |
| No physical exercise | .028 | 1.592 | 1.075–2.549 |
| PNI < 45 | .014 | 3.611 | 1.513–3.987 |
| Depression | .046 | 1.498 | 1.033–1.856 |
CI = confidence interval, OR = odds ratio, PNI = prognosis nutrition index.
3.3. Nomogram development and validation
Based on the logistic regression results and incorporating the weights of each predictor, a nomogram was constructed to assess frailty risk in ischemic stroke patients (Fig. 2). This nomogram assigns corresponding scores to each patient’s predictors, ultimately calculating the total score to evaluate the probability of frailty occurrence.
Figure 2.
Nomogram prediction model for frailty development in ischemic stroke patients. PNI = prognosis nutrition index.
3.4. Model performance evaluation
The nomogram was validated using 10-fold cross-validation. ROC curve analysis demonstrated an AUC of 0.814 (Fig. 3), indicating good discriminatory ability in predicting frailty risk among ischemic stroke patients. The optimal cutoff value was determined using the Youden index, representing the threshold that best distinguishes patients at higher risk of frailty from those at lower risk.
Figure 3.
Nomogram regression model to predict ROC curve for frailty development in ischemic stroke patients. ROC = receiver operating characteristic.
3.5. Support vector machine model
A SVM model was constructed to predict frailty risk. ROC analysis revealed an AUC of 0.842 (Fig. 4), with an overall classification accuracy of 88.7%, suggesting strong predictive performance. The optimal cutoff value for classification was also determined using the Youden index.
Figure 4.
SVM model to predict the ROC curve for frailty development in ischemic stroke patients. ROC = receiver operating characteristic, SVM = support vector machines.
3.6. Comparison of nomogram and support vector machine models
The SVM model showed a numerically higher AUC than the nomogram (0.842 vs 0.814). To determine whether this difference was statistically significant, DeLong’s test was performed. The results indicated that the difference between the 2 AUCs was not statistically significant (ΔAUC = 0.028, P = .087).
These findings suggest that although the SVM model demonstrated slightly better discrimination, its advantage over the nomogram was not statistically significant.
3.7. Calibration and clinical utility
Calibration curves showed good agreement between predicted and observed frailty probabilities for both the nomogram and SVM models (Fig. 5). The Hosmer–Lemeshow test indicated satisfactory calibration (P > .05).
Figure 5.
Calibration curves of the nomogram and SVM models for predicting frailty risk in ischemic stroke patients. (A) Calibration curve of the nomogram model. (B) Calibration curve of the support vector machine (SVM) model. SVM = support vector machines.
Decision curve analysis demonstrated that both models provided higher net clinical benefit than the treat-all and treat-none strategies across a wide range of threshold probabilities (Fig. 6), indicating potential clinical usefulness.
Figure 6.
Decision curve analysis (DCA) of the nomogram and SVM models for predicting frailty risk in ischemic stroke patients. (A) Decision curve analysis of the nomogram model. (B) Decision curve analysis of the SVM model. DCA = decision curve analysis, SVM = support vector machines.
3.8. Classification performance
At the optimal cutoff value, the nomogram achieved a sensitivity of 77.5%, specificity of 74.3%, PPV of 61.1%, and NPV of 86.4% (validation cohort, n = 260).
For the SVM model, sensitivity, specificity, PPV, and NPV were 82.0%, 77.2%, 65.2%, and 89.2%, respectively (validation cohort, n = 260).
4. Discussion
Based on clinical data from 867 ischemic stroke patients, this study systematically analyzed the prevalence of frailty and its associated risk factors and developed corresponding predictive models. The results showed that 34.14% of patients were classified as frail, indicating that frailty is a common and clinically significant condition in this population. Multivariate logistic regression identified hypertension, impaired self-care ability, physical inactivity, reduced PNI, and depressive status as independent risk factors. Both the nomogram and SVM models demonstrated good predictive performance for frailty risk in ischemic stroke patients.
Hypertension was confirmed as an independent risk factor for frailty (OR = 1.412), which is consistent with previous studies. Hypertension not only represents a major risk factor for stroke but also plays an important role in the development and progression of frailty. Long-term hypertension can lead to arteriosclerosis, cerebral small vessel disease, and impaired cerebral perfusion, thereby accelerating neuronal injury and brain function degeneration and reducing neuroplastic recovery capacity after stroke.[12] Furthermore, hypertension is closely associated with chronic systemic inflammation. Pro-inflammatory cytokines, such as IL-6 and TNF-α, may mediate muscle protein degradation and mitochondrial dysfunction, leading to reduced skeletal muscle mass and strength and forming a “sarcopenia–frailty” pathway.[13] Previous studies have also reported a significantly increased risk of frailty in elderly hypertensive individuals.[14]
Impaired self-care ability (BI ≤ 60), corresponding to moderate-to-severe dependency, was identified as another independent risk factor for frailty. A low BI reflects dependence on others or assistive devices for activities of daily living, such as eating, mobility, bathing, and toileting, indicating diminished overall functional reserve.[15] Functional limitations may reduce physical activity, further triggering muscle disuse atrophy, decreased skeletal muscle strength, and abnormal bone metabolism, thereby promoting frailty development. In addition, functional dependence often coexists with inadequate nutritional intake, impaired digestion and metabolism, and increased infection risk, which further weakens physical capacity at the nutritional–metabolic level. Long-term dependence may also lead to psychosocial consequences, including reduced self-efficacy, depressive mood, and social withdrawal, which significantly contribute to frailty progression.[13]
Physical inactivity was also identified as an independent risk factor for frailty (OR = 1.592), highlighting its crucial role in poststroke frailty development. Regular physical activity is essential for maintaining muscle mass, bone density, and cardiopulmonary function. Conversely, insufficient physical activity reduces skeletal muscle protein synthesis while increasing degradation, leading to sarcopenia and diminished exercise capacity, which are key pathological foundations of frailty.[16] Stroke patients often experience activity limitations due to neurological dysfunction, hemiplegia, and psychological factors, creating a vicious cycle in which reduced activity leads to muscle atrophy and joint stiffness, further increasing dependence in daily activities and frailty risk. Moreover, physical inactivity reduces insulin sensitivity and cardiovascular adaptability, increasing the risk of metabolic syndrome and vascular events and thereby worsening poststroke vulnerability.[17] Studies have shown that individuals with low physical activity levels face a significantly higher risk of frailty than those with adequate activity.[18]
Low PNI (<45) was strongly associated with frailty (OR = 3.611), suggesting that nutritional and immune status play a central role in frailty development. PNI is a composite indicator derived from serum albumin levels and peripheral blood lymphocyte counts, reflecting both nutritional reserves and immune function.[19] A decline in PNI often indicates insufficient protein synthesis, chronic wasting, and weakened immune defenses. Low albumin levels suggest persistent malnutrition and inflammation, which accelerate muscle protein breakdown and promote sarcopenia, thereby impairing stress response and functional recovery capacity. Meanwhile, lymphopenia reflects compromised immune function and increased susceptibility to infection and chronic inflammation, further accelerating systemic frailty progression. Moreover, malnutrition and immune dysfunction may impair neural repair and rehabilitation outcomes after stroke, exacerbating dependence in daily activities and promoting frailty syndrome development.[20] Previous studies have demonstrated that low PNI levels significantly predict postoperative complications and poor outcomes.[21]
Depression was also identified as an independent risk factor for frailty (OR = 1.498), indicating a close association between psychological status and frailty development. Poststroke depression may accelerate frailty progression through multiple mechanisms. First, depression is often accompanied by hyperactivation of the hypothalamic–pituitary–adrenal axis. Prolonged cortisol secretion can induce insulin resistance, muscle protein breakdown, and reduced bone density, thereby promoting sarcopenia and frailty.[22] Second, depression is closely associated with chronic inflammation, and elevated serum levels of IL-6, CRP, and TNF-α have been consistently reported in depressed patients. These pro-inflammatory factors overlap with frailty phenotypes, such as decreased physical capacity and reduced activity.[23] Third, depression significantly reduces rehabilitation adherence and physical activity levels, limiting neurological recovery and daily functioning and thereby increasing frailty risk. In addition, depression is closely linked to malnutrition, as decreased appetite and inadequate energy intake further weaken immune and metabolic reserves, exacerbating frailty.[24] Studies have shown that individuals with higher depression symptom scores exhibit a markedly increased risk of frailty during follow-up.[25]
Based on these findings, this study developed practical nomogram and SVM models to predict frailty risk in ischemic stroke patients. Nomograms are widely used clinical tools that provide individualized risk estimation in a simple, intuitive, and cost-effective manner. Using 10-fold cross-validation, the nomogram demonstrated good predictive performance and clinical interpretability.
As a machine learning method, SVM can capture complex nonlinear relationships within large datasets. In this study, the SVM model achieved a numerically higher AUC (0.842) and classification accuracy (88.7%) compared with the nomogram (AUC = 0.814). However, DeLong’s test indicated that the difference between the 2 models was not statistically significant, suggesting comparable discriminatory performance.
From a clinical perspective, the nomogram offers important advantages, including transparency, ease of use, and suitability for bedside application, which may facilitate routine clinical decision-making. In contrast, although the SVM model showed marginally better numerical performance, its implementation requires specialized software and technical expertise, potentially limiting its widespread clinical adoption.
Therefore, the 2 models may be considered complementary. The nomogram is more suitable for routine clinical practice, while the SVM model may serve as a supportive analytical tool in research settings or institutions with advanced computational resources.
This study has several limitations. First, as a single-center retrospective study, the sample was derived exclusively from patients treated at our institution, which may introduce selection bias and limit the generalizability of the findings. Second, although multiple demographic, clinical, nutritional, and psychological variables were included, some potentially important factors may have been omitted, such as social support, medication use, stroke subtype, and stroke severity. In particular, key stroke-related variables, including NIHSS score, infarct volume, and acute treatments (such as intravenous thrombolysis or mechanical thrombectomy), were not available in our dataset. These factors are well-recognized determinants of functional outcomes and frailty status after ischemic stroke, and their absence may have resulted in residual confounding and reduced the comprehensiveness of the predictive models. Third, this study relied primarily on baseline data collected during hospitalization and lacked long-term follow-up, which prevented dynamic assessment of frailty progression and its impact on long-term outcomes. Fourth, certain variables, such as depressive status and physical activity levels, were assessed using self-reported scales, which may have introduced information bias. Taken together, these limitations suggest that future studies should adopt multicenter, large-sample, prospective designs with comprehensive stroke-related variables and long-term follow-up to further validate and refine the predictive models and enhance their clinical applicability.
5. Conclusion
Based on clinical data from 867 ischemic stroke patients, this study investigated the incidence of poststroke frailty and its risk factors, constructing a visual predictive model. Results indicate a high prevalence of frailty (34.14%) among ischemic stroke patients, influenced by multiple factors. Multivariate logistic regression analysis identified hypertension, self-care ability ≤ 60 points, physical inactivity, PNI < 45, and depressive status as independent risk factors. Both the nomogram and SVM models demonstrated good predictive capabilities.
Author contributions
Conceptualization: Ya-li Kang.
Data curation: Ya-li Kang.
Formal analysis: Jin-tao Kong.
Investigation: Jin-tao Kong.
Methodology: Jin-tao Kong.
Software: Hua Tian.
Validation: Hua Tian.
Visualization: Hua Tian.
Writing – original draft: Ya-li Kang.
Writing – review & editing: Ya-li Kang.
Abbreviations:
- AUC
- area under the curve
- BI
- Barthel Index
- BMI
- body mass index
- FFP
- Fried frailty phenotype
- GDS-15
- Geriatric Depression Scale-15
- PNI
- prognosis nutrition index
- ROC
- receiver operating characteristic
- SVM
- support vector machines.
This retrospective study was reviewed and approved by the Ethics Committee of The First Affiliated Hospital of Xinjiang Medical University (Approval No. K202405-10). Due to the retrospective design and the use of anonymized patient data, the requirement for informed consent was waived by the Ethics Committee. All procedures were conducted in accordance with the Declaration of Helsinki.
The authors have no funding and conflicts of interest to disclose.
The datasets generated during and/or analyzed during the current study are not publicly available, but are available from the corresponding author on reasonable request.
How to cite this article: Kang Y-l, Kong J-t, Tian H. Prediction model for frailty risk in ischemic stroke patients: Application and validation of support vector machines and nomograms. Medicine 2026;105:14(e48270).
Contributor Information
Jin-tao Kong, Email: kongjintao78451@163.com.
Hua Tian, Email: tianhua78435@163.com.
References
- [1].Forti P, Ciani M, Maioli F. Enhancing prognostic models for stroke outcomes in older patients: insights from pre-stroke frailty and disability assessments. J Neurol Sci. 2025;477:123659. [DOI] [PubMed] [Google Scholar]
- [2].Rust M, Küppers C, Nikoubashman O, et al. Blood-based frailty index in patients with acute ischemic stroke undergoing endovascular treatment. Cerebrovasc Dis. 2025;54:518–26. [DOI] [PubMed] [Google Scholar]
- [3].Soler-Espejo E, Zazo-Luengo BA, Rivera-Caravaca JM, et al. Poor clinical outcomes associated with multimorbidity, frailty, and malnutrition in patients with atrial fibrillation. J Nutr Health Aging. 2025;29:100430. [DOI] [PMC free article] [PubMed] [Google Scholar]
- [4].Bahar AR, Bahar Y, Kidess G, et al. Association of frailty with outcomes in patients with large vessel occlusion stroke undergoing mechanical thrombectomy. Neuroradiology. 2025;67:845–54. [DOI] [PubMed] [Google Scholar]
- [5].Huang L, Gong C, Qiu Z, et al. Association between premorbid frailty status and functional independence in acute ischemic stroke patients following endovascular treatment in the late window. Clin Interv Aging. 2025;20:523–35. [DOI] [PMC free article] [PubMed] [Google Scholar]
- [6].Zhang S, Jiang Y, Wang H, et al. Associations between plant-based diets and cardiovascular disease, frailty, and cognitive dysfunction in middle and old age: a systematic review and meta-analysis of cohort studies. Food Funct. 2025;16:4061–84. [DOI] [PubMed] [Google Scholar]
- [7].Schnieder M, Metz H, Baehr M, et al. ; GSR-ET Investigators. Frailty’s influence on older stroke patients: Neurological outcome and mortality after endovascular treatment in stroke: a national German stroke registry analysis. Eur Stroke J. 2025;10:1312–9. [DOI] [PMC free article] [PubMed] [Google Scholar]
- [8].Nguyen NH, Luong TT, Le TN, Nguyen HTT, Nguyen DV. Prevalence of frailty among older patients with acute ischemic stroke and its association with short-term functional outcome: a cohort study in Vietnam. BMC Geriatr. 2025;25:430. [DOI] [PMC free article] [PubMed] [Google Scholar]
- [9].Hu Y, Zhang Y, Han P, et al. Machine learning-driven prediction of readmission risk in heart failure patients with diabetes: synergistic assessment of inflammatory and metabolic biomarkers. Int J Cardiol. 2025;441:133743. [DOI] [PubMed] [Google Scholar]
- [10].Zou JH, Wang H-W, Long J-Z, et al. Unveiling the diagnostic and causal role of hyperlipidemia- and lipophagy-associated genes PLAUR, IVNS1ABP, and QKI in acute myocardial infarction. Int J Cardiol. 2025;441:133635. [DOI] [PubMed] [Google Scholar]
- [11].Chen C, Chen X, Gao Y, Deng Y, Li Z. Graded nomograms based on perioperative parameters for predicting New-Onset severe acute kidney injury following liver transplantation in patients with normal preoperative renal function: the SALT scale. Ren Fail. 2025;47:2553809. [DOI] [PMC free article] [PubMed] [Google Scholar]
- [12].Aswi A, Poerwanto B, Hammado N, et al. Spatial Bayesian semi-parametric Cox-Leroux modeling of stroke patient hospitalization: aspects on survival. Geospat Health. 2025;20:23. [DOI] [PubMed] [Google Scholar]
- [13].Luo Y, Hao J, Su Z, et al. Prevalence and related factors of hypokalemia in patients with acute ischemic stroke. Int J Gen Med. 2024;17:5697–705. [DOI] [PMC free article] [PubMed] [Google Scholar]
- [14].Kim BJ, Cho Y-J, Hong K-S, et al. Treatment intensification for elevated blood pressure and risk of recurrent stroke. J Am Heart Assoc. 2021;10:e019457. [DOI] [PMC free article] [PubMed] [Google Scholar]
- [15].Winovich DT, Longstreth WT, Arnold AM, et al. Factors associated with ischemic stroke survival and recovery in older adults. Stroke. 2017;48:1818–26. [DOI] [PMC free article] [PubMed] [Google Scholar]
- [16].Okoye C, Finazzi A, Pagan E, et al. one-year mortality in hospitalized, frail older adults with atrial fibrillation on oral anticoagulant therapy: impact of stroke and bleeding events. Drugs Aging. 2025;42:781–9. [DOI] [PubMed] [Google Scholar]
- [17].Chen J, Wang C, Duan P, et al. Associations of lung function impairment and biological aging with mortality and cardiovascular disease incidence: findings from UK biobank participants. Front Public Health. 2025;13:1635195. [DOI] [PMC free article] [PubMed] [Google Scholar]
- [18].Evans NR, Pinho J, Beishon L, et al. Frailty and stroke: global implications for assessment, research, and clinical care-A WSO scientific statement. Int J Stroke. 2025;20:905–17. [DOI] [PMC free article] [PubMed] [Google Scholar]
- [19].Fawzy AM, Bisson A, Lochon L, Lenormand T, Lip GYH, Fauchier L. Outcomes in atrial fibrillation patients with different clinical phenotypes: insights from the French population. J Clin Med. 2025;14:1044. [DOI] [PMC free article] [PubMed] [Google Scholar]
- [20].Gurková E, Bartoníčková D, Šaňák D, Šaňáková S, Zapletalová J, Štureková L. Relationship between social support, functional outcomes and health-related quality of life in working-aged adults at three months after ischemic stroke: results from the FRAILTY study. Health Qual Life Outcomes. 2025;23:8. [DOI] [PMC free article] [PubMed] [Google Scholar]
- [21].Duan H, Ding Y, Cheng Z, et al. Low serum alanine aminotransferase (ALT) levels are associated with poor outcomes in acute ischemic stroke patients regardless of age. Brain Res. 2024;1842:149130. [DOI] [PubMed] [Google Scholar]
- [22].De Cassai A, Sella N, Pettenuzzo T, et al. Anesthetic management of acute ischemic stroke undergoing mechanical thrombectomy: an overview. Diagnostics (Basel). 2024;14:2113. [DOI] [PMC free article] [PubMed] [Google Scholar]
- [23].Cheng H, Ling Y, Li Q, et al. ICU admission braden score independently predicts delirium in critically ill patients with ischemic stroke. Intensive Crit Care Nurs. 2024;82:103626. [DOI] [PubMed] [Google Scholar]
- [24].Wei J, Wang J, Chen J, Yang K, Liu N. Stroke and frailty index: a two-sample Mendelian randomisation study. Aging Clin Exp Res. 2024;36:114. [DOI] [PMC free article] [PubMed] [Google Scholar]
- [25].Park JW, Kim J-T, Lee JS, et al. Brain frailty and outcomes of acute minor ischemic stroke with large-vessel occlusion. J Clin Neurol. 2024;20:175–85. [DOI] [PMC free article] [PubMed] [Google Scholar]