Construction and Validation of a Risk Prediction Model Incorporating Temporal Muscle Thickness for Adverse Outcome in Acute Ischemic Stroke Patients

Abstract

Background

Sarcopenia significantly impacts stroke prognosis. Temporal muscle thickness (TMT) is an emerging metric for sarcopenia.

Objectives

To developed a TMT-incorporated model to predict 6-month adverse outcomes in acute ischemic stroke (AIS).

Design

In this retrospective study, 479 AIS patients were divided into training (n=283), test (n=120), and external validation cohorts (n=76).

Methods

A combined model was constructed to predict adverse outcomes in the training and test cohorts using LASSO regression analysis. Model performance was assessed via calculating accuracy, sensitivity, specificity, positive predictive value (PPV), negative predictive value (NPV) and F1 score.

Results

The proportion of patients with an adverse outcomes in the training and test sets was 18.02% vs 17.50%, respectively (P = 0.90). Ischemic stroke event, admission NIHSS score, BI score, TMT were used to construct the prediction model. The combined model presented good discriminatory potential in the training and test sets (AUC = 0.929 and 0.930, respectively), which was verified in the external validation cohort (AUC = 0.902). For the combined model, the P values of the Hosmer–Lemeshow test in the training set, the test set, and the external validation were < 0.001 (χ2 = 44.007), 0.472 (χ2 = 7.611), and <0.001 (χ2 = 4919.666), respectively. The combined model showed good calibration and discrimination. The clinical usefulness of the model was confirmed by decision curve analysis.

Conclusion

This study developed a combined model incorporating ischemic stroke event, admission NIHSS score, BI score, TMT and infarct volume to predict 6-month adverse outcomes in AIS patients, providing clinicians with a practical tool for treatment decisions and prognosis assessment.

Keywords

acute ischemic stroke outcome sarcopenia temporal muscle prediction model

Introduction

Stroke is the second leading cause of death worldwide. The burden of disability in stroke survivors is high, particularly in low- and middle-income countries, where the burden is increasing more rapidly compared with high-income countries.¹ In China, 3.94 million new cases of stroke occurred in 2019; among these, 2.87 million were ischemic in origin. The same year, stroke caused 2.19 million deaths nationwide.² Better neurological status 6 months after acute ischemic stroke (AIS) is associated with longer survival.³ Therefore, accurately predicting 6-month functional outcome at the time of stroke would help guide appropriate management and improve healthcare resource allocation.

Sarcopenia is an aging-related loss of skeletal muscle mass and function which has been associated with poor functional outcome in AIS patients.⁴ The diagnosis of sarcopenia is typically difficult in stroke patients with impaired consciousness or cognitive dysfunction because they cannot cooperate adequately to participate in clinical assessments of muscle strength and physical function.⁵ Temporal muscle thickness (TMT) and area (TMA) as measured on magnetic resonance imaging (MRI) have been associated with functional outcome in patients with brain tumors, subdural hematomas, and stroke.^6-9 However, in the real world, functional outcome is influenced by multiple clinical variables—relying on a single one to predict outcome would not be accurate. Optimally, models used to predict patient outcomes would include these multiple variables, yet they are often overlooked.

Several studies have designed and evaluated model for predicting 6-month outcome in AIS patients. Sun et al developed the model, which is based on National Institutes of Health Stroke Scale (NIHSS) score on admission, age, previous diabetes, and creatinine, to predict the probability of poor outcome 6 months after AIS in Chinese patients.¹⁰ Li et al used machine learning to develop another one that incorporates age, NIHSS score on admission, premorbid modified Rankin scale (mRS) score, fasting blood glucose, and creatinine; their model exhibited better predictive performance.¹¹ However, both models require acquisition of blood samples and have not been externally validated. In contrast, MRI examination offers the advantage of being non-invasive, and the TMT is more readily measurable.

Models combining clinical features with radiomics based on diffusion-weighted imaging (DWI) or fluid-attenuated inversion recovery (FLAIR) images have performed well in predicting AIS outcome.^12,13 However, the assessment of AIS lesions on MRI relies on manual measurement or segmentation, which is time-consuming, requires trained personnel, and is prone to interrater inconsistency.¹⁴ Thus, the aim of this study was to develop and validate a combined model combining temporal muscle variables and clinical predictors for individualized 6-month unfavorable outcome prediction in AIS patients.

Methods

Study Participants

A total of 479 patients admitted for AIS to two clinical centers from December 2022 to December 2023 were retrospectively analyzed. Those from the First Affiliated Hospital of Shantou University Medical College (n = 403) were randomly divided into the training (n = 283) and test (n = 120) sets, while those from Meizhou People’s Hospital (n = 76) were conducted as the external validation set. A study flow chart is shown in Figure 1. Institutional ethics review board approval was obtained (no. B-2024-164). The requirement for informed consent was waived owing to the study’s retrospective design.

Figure 1.

Study flow chart

Inclusion and Exclusion Criteria

Patients’ inclusion criteria were an AIS diagnosis and an age > 18 years. The exclusion criteria were: (1) history of intracerebral hemorrhage, brain trauma, or previous neurological disease; (2) transient ischemic attack or subarachnoid hemorrhage; (3) severe artifact on imaging studies; and (4) severe concomitant heart/liver/kidney disease. We also excluded patients who underwent endovascular AIS treatment and those whose clinical, radiological, or follow-up data were unavailable.

Data Collection

The following data was collected for all participants: age, sex, vascular risk factors, previous medical history, Barthel Index (BI) score, and admission NIHSS score. All patients were clinically evaluated 6 months after AIS using the mRS questionnaire. Evaluations of the BI, NIHSS and mRS scores were performed by trained and certified assessors. Favorable outcomes were mRS 0-2, while unfavorable outcomes were mRS 3-6.¹⁵ When the missing value of the observation index data was greater than or equal to 20%, the observation index was removed. For missing values below 20%, the multiple imputation method was employed to fill the gap.

MRI and Image Analysis

All patients underwent MRI of the brain on a 1.5 T GE SIGNA HDXT scanner (GE Healthcare, Milwaukee, WI, USA) with an eight-channel head coil or a Siemens Magnetom Trio 3.0 T scanner (Siemens Healthineers, Erlangen, Germany) with an 18-channel head/neck coil. Detailed acquisition parameters are summarized in Supplementary Table 1. T1-and T2-weighted sequences, DWI, and FLAIR imaging were performed in all patients.

Cerebral white matter hyperintensity (CWMH) was defined according to the Standards for Reporting Vascular Changes on Neuroimaging guidelines as white matter hyperintensity in the periventricular and subcortical areas of the brain without cavitation on FLAIR imaging.¹⁶Brain-stem white matter hyperintensity (BWMH) was defined as white matter hyperintensity in the brain stem.¹⁷ FLAIR vascular hyperintensity (FVH) was defined as a focal, tubular, or serpentine hyperintensity (relative to cerebrospinal fluid) in the subarachnoid space that corresponded to the typical course of a cerebral artery.¹⁸ The definition of a silent DWI lesion is as follows: 1) A parenchymal area that is markedly hyperintense on FLAIR relative to the surrounding brain and distinct from the hyperacute infarct. 2) A parenchymal area that is mild to markedly hyperintense on DWI with high b = 1000 relative to surrounding tissue.¹⁹ The presence of CWMH, FVH, BWMH, and silent DWI lesions were assessed and recorded. Infarct volume (IV) was calculated in milliliters (mL) by multiplying the number of voxels of the segmented ischemic lesions with its voxel size.

TMT was used as temporal muscle markers of skeletal mass. It was manually measured on axial T2-weighted images obtained at the time of admission as shown in Figure 2. Slice thickness was reconstructed to 6 mm in the picture archiving and communication system before segmentation. TMT was measured perpendicular to the temporal muscle’s long axis, spanning from inner to outer margins (excluding fascia). The landmark for craniocaudal orientation was the roof of the orbit; the Sylvian fissure was used as the reference point for frontal–occipital orientation.^20,21 Bilateral averages of TMT was computed for analysis (Figure 2). All measurements were performed by a single board-certified neuroradiologist with 5 years of experience (HW) who was blinded to all data.

Figure 2.

T2-weighted imaging of the brain measuring temporal muscle thickness (TMT). The red line indicates standard TMT, (A) Patient with high TMT values. (B) Patient with low TMT values

Statistical Analysis

Continuous data are summarized as means with standard deviation and were compared using the Student’s t-test or Mann–Whitney U test. Categorical variables are summarized as frequencies with percentage and were compared using the chi-square or Fisher’s exact test. Through LASSO regression analysis, clinical and imaging features associated with poor prognosis in ischemic stroke were screened to construct a clinical model, an imaging model, and a combined model.

The discrimination of these models was evaluated using the receiver operating characteristic (ROC) curve and the area under the curve (AUC). The DeLong test was used to compare the AUC values between the models. Calibration was assessed with the Hosmer-Lemeshow test and the calibration curve. Furthermore, the clinical utility of the model was examined through clinical decision curve analysis (DCA). In addition, the performance of all models was assessed by calculating accuracy, sensitivity, specificity, positive predictive value (PPV), negative predictive value (NPV) and F1 score.

Statistical analyses were performed using SPSS software version 25.0 (IBM Corp., Armonk, NY, USA) and R version 4.3.1 (R Development Core Team, Auckland, New Zealand). All tests were two-sided. A p-value of less than 0.05 was considered significant.

Results

Patient Characteristics

Overall, the median patient age was 64.00 (57.00, 72.00) years and 66.25% were men. Patient characteristics were well-balanced between the training (n = 283) and test (n = 120) sets (Table 1). The training and test cohorts showed similar unfavorable outcome rates (18.02% vs 17.50%, P = 0.901).

Table 1.

Population Characteristics of Patients With AIS in the Training and Test Sets

Variables	Total (n = 403)	Training dataset (n = 283)	Test dataset (n = 120)	P
Age, median (IQR)	64.00 (57.00, 72.00)	64.00 (57.00, 71.00)	64.50 (55.25, 72.00)	0.810^d
Sex^#				0.420^a
Male	267 (66.25)	191 (67.49)	76 (63.33)
Female	136 (33.75)	92 (32.51)	44 (36.67)
Smoking^#				0.388^a
Yes	198 (49.13)	143 (50.53)	55 (45.83)
No	205 (50.87)	140 (49.47)	65 (54.17)
Drink^#				0.368^a
Yes	45 (11.17)	29 (10.25)	16 (13.33)
No	358 (88.83)	254 (89.75)	104 (86.67)
family history of ischemic stroke^#				0.993^a
Yes	57 (14.14)	40 (14.13)	17 (14.17)
No	346 (85.86)	243 (85.87)	103 (85.83)
Admission NIHSS score, median (IQR)	3.00 (1.00, 5.00)	3.00 (1.00, 5.00)	3.00 (1.00, 4.75)	0.694^d
Number of ischemic stroke event^#				0.034^a
<2	331 (82.13)	225 (79.51)	106 (88.33)
≥ 2	72 (17.87)	58 (20.49)	14 (11.67)
Hypertension^#				0.438^a
Yes	309 (76.67)	220 (77.74)	89 (74.17)
No	94 (23.33)	63 (22.26)	31 (25.83)
Diabetes^#				0.468^a
Yes	157 (38.96)	107 (37.81)	50 (41.67)
No	246 (61.04)	176 (62.19)	70 (58.33)
Atrial fibrillation^#				0.457^a
Yes	22 (5.46)	17 (6.01)	5 (4.17)
No	381 (94.54)	266 (93.99)	115 (95.83)
Hyperlipemia^#				0.628^a
Yes	101 (25.06)	69 (24.38)	32 (26.67)
No	302 (74.94)	214 (75.62)	88 (73.33)
Barthel index, median (IQR)	95.00 (60.00, 100.00)	95.00 (60.00, 100.00)	95.00 (56.25, 100.00)	0.882^d
TMT (mm), mean (SD)	5.43 ± 1.46	5.47 ± 1.48	5.35 ± 1.41	0.465^c
AIS side^#				0.978^a
Bilateral	30 (7.44)	21 (7.42)	9 (7.50)
Unilateral	373 (92.56)	262 (92.58)	111 (92.50)
CWMH^#				0.074^b
Yes	389 (96.53)	270 (95.41)	119 (99.17)
No	14 (3.47)	13 (4.59)	1 (0.83)
BWMH^#				0.468^a
Yes	131 (32.51)	89 (31.45)	42 (35.00)
No	272 (67.49)	194 (68.55)	78 (65.00)
FVH^#				0.290^a
Yes	59 (14.64)	38 (13.43)	21 (17.50)
No	344 (85.36)	245 (86.57)	99 (82.50)
encephalanalosis				0.817^a
Yes	292 (72.46)	206 (72.79)	86 (71.67)
No	111 (27.54)	77 (27.21)	34 (28.33)
Infarct volume (mL), median (IQR)	0.84 (0.28, 3.24)	0.83 (0.24, 3.39)	0.95 (0.38, 2.97)	0.416^d
Silent DWI lesion^#				0.763^b
Yes	13 (3.23)	10 (3.53)	3 (2.50)
No	390 (96.77)	273 (96.47)	117 (97.50)
Poor prognosis^#				0.901^a
Yes	72 (17.87)	51 (18.02)	21 (17.50)
No	331 (82.13)	232 (81.98)	99 (82.50)

NISSH, The National Institutes of Health Stroke Scale; TMT, temporal muscle thickness; TMA, temporal muscle area; BWMH, brain-stem white matter hyperintensity; CWMH, cerebral white matter hyperintensity; FVH, fluid-attenuated inversion recovery vascular hyperintensity; AIS, acute ischemic stroke; IQR, interquartile range; DWI, diffusion weighted image.

^#Results are measurements with corresponding ratio in parentheses, and the remainder results are mean value with standard deviation or median with interquartile range.

^aP: chi square test,

^bP: Fisher exact test,

^cP: student’s t test,

^dP: Mann–Whitney U test.

Characteristics of the 403 patients in the training and test sets combined grouped according to outcome are shown in Table 2. Patient age [73.00 (63.00, 78.00) vs. 63.00 (56.00, 70.00) years; P <0.0001] and NIHSS score on admission [7.00 (4.00, 10.00) vs. 2.00 (1.00, 4.00); P <0.0001) were significantly higher in the unfavorable outcome group. BI score [30.00 (20.00, 50.00) vs. 100.00 (80.00, 100.00); P <0.0001], TMT (4.34 ± 1.43 vs. 5.67 ± 1.35 mm; P <0.0001) were significantly lower in the unfavorable outcome group. The two groups also significantly differed with respect to sex distribution, smoking status, drinking status, family history of ischemic stroke, number of ischemic stroke event, prevalence of atrial fibrillation, prevalence of BWMH, prevalence of FVH, prevalence of encephalanalosis and infarct volume.

Table 2.

Clinical, Demographic, and Radiological Data of the Study Population Stratified According to 6-Month Favorable or Unfavorable Outcome After Acute Ischemic Stroke

Variables	Favorable prognosis (n = 331)	Poor prognosis (n = 72)	P
Age, median (IQR)	63.00 (56.00, 70.00)	73.00 (63.00, 78.00)	< 0.001^d
Sex^#			< 0.001^a
Male	236 (69.79)	31 (43.06)
Female	95 (28.70)	41 (56.94)
Smoking^#			0.001^a
Yes	176 (53.17)	22 (30.56)
No	155 (46.83)	50 (69.44)
Drink^#			0.037^a
Yes	42 (12.69)	3 (4.17)
No	289 (87.31)	69 (95.83)
family history of ischemic stroke^#			0.235^a
Yes	50 (15.11)	7 (9.72)
No	281 (84.89)	65 (90.28)
Number of ischemic stroke event^#			< 0.001^a
<2	285 (86.10)	46 (63.89)
≥ 2	46 (13.90)	26 (36.11)
Admission NIHSS score, median (IQR)	2.00 (1.00, 4.00)	7.00 (4.00, 10.00)	< 0.001^d
Hypertension^#			0.390^a
Yes	251 (75.83)	58 (80.56)
No	80 (24.17)	14 (19.44)
Diabetes^#			0.989^a
Yes	129 (38.97)	28 (38.89)
No	202 (61.03)	44 (61.11)
Atrial fibrillation^#			0.008^b
Yes	13 (3.93)	9 (12.50)
No	318 (96.07)	63 (87.50)
Hyperlipemia^#			0.070^a
Yes	89 (26.89)	12 (16.67)
No	242 (73.11)	60 (83.33)
Barthel index, median (IQR)	100.00 (80.00, 100.00)	30.00 (20.00, 50.00)	< 0.001^d
TMT (mm), mean (SD)	5.67 ± 1.35	4.34 ± 1.43	< 0.001^c
AIS side^#			0.751^a
Bilateral	24 (7.25)	6 (8.33)
Unilateral	307 (92.75)	66 (91.67)
CWMH^#			0.085^b
Yes	317 (95.77)	72 (100.00)
No	14 (4.23)	0 (0.00)
BWMH^#			0.017^a
Yes	99 (29.91)	32 (44.44)
No	232 (70.09)	40 (55.56)
FVH^#			< 0.001^a
Yes	38 (11.48)	21 (29.17)
No	293 (88.52)	51 (70.83)
encephalanalosis			0.001^a
Yes	228 (68.82)	64 (88.89)
No	103 (31.12)	8 (11.11228)
Infarct volume (mL), median (IQR)	0.61 (0.21, 2.23)	2.65 (1.10, 10.60)	< 0.001^d
Silent DWI lesion^#			0.261^b
Yes	9 (2.72)	4 (5.56)
No	322 (97.28)	68 (94.44)

^#Results are measurements with corresponding ratio in parentheses, and the remainder results are mean value with standard deviation or median with interquartile range.

^aP: chi square test,

^bP: Fisher exact test,

^cP: student’s t test,

^dP: Mann–Whitney U test.

Screening of Clinical and Imaging Variables and Establishment of the Models

Through the 10-fold cross-validation LASSO regression method, 5 clinical features and 3 radiological features were selected (Figure 3). Given 72 positive events in this study, variables were further selected based on the EPV principle and clinical experience for constructing the final clinical, imaging, and combined models. The clinical model included number of ischemic stroke event, admission NIHSS score and BI score. The radiological model included TMT and infarct volume. The combined model was then constructed using all five of the aforementioned variables. All models were developed using logistic regression analysis.

Figure 3.

The figure illustrates the process of clinical and radiological features selection. (A) and (C) are LASSO coefficient profiles for clinical and radiological features. (B) and (D) are selections of the optimal tuning parameter (λ) in LASSO regression via 10-fold cross-validation

Performance Evaluation and Validation

The AUCs for the clinical model in the training set and test set were 0.920 (95%CI 0.874-0.967) and 0.917 (95%CI 0.830-1.000), respectively, while in the external validation set, the AUC was 0.880 (95%CI 0.789-0.972). The radiological model achieved AUCs of 0.784 (95%CI 0.708-0.859) and 0.790 (95%CI 0.688-0.892) in the training and test sets, respectively, and an AUC of 0.651 (95%CI 0.459-0.843) in the external validation set. For the combined model, the AUC was 0.929 (95%CI 0.876-0.981) in the training set, 0.930 (95%CI 0.856-1.000) in the test set, and 0.902 (95%CI 0.832-0.972) in the external validation set (Table 3). The apparent and optimism-corrected AUCs were nearly identical across all models (differences: 0.002, <0.001, and 0.011 for clinical, radiological, and combined models, respectively), indicating excellent internal stability and minimal overfitting (Supplementary Table 2). The DeLong tests on the training set showed that the clinical model (Z = 3.553, P = 0.004) and combined model (Z = -4.327, P < 0.0001) significantly outperformed the radiological model. The DeLong tests on the test set revealed that the combined model significantly outperformed the radiological model (Z = -2.493, P = 0.013) and the clinical model showed borderline superiority over the radiological model (Z = 1.941, P = 0.052). The DeLong tests on the external validation set showed that the clinical model (Z = 1.993, P = 0.046) and combined model (Z = -2.778, P = 0.006) significantly outperformed the radiological model. However, the clinical and combined models demonstrated comparable discriminative ability, with a non-significant AUC difference in the training (Z = -0.637, P = 0.524), test (Z = -0.836, P = 0.403), or external validation (Z = -0.651, P = 0.515) set. The ROC curves for the three models are shown in Figure 4.

Table 3.

Prediction Model Performance of the Clinical, Radiological and Combined Model for the Training Set, Test Set and External Validation Set

Model	AUC (95%CI)	Accuracy	Sensitivity	Specificity	PPV	NPV	F1 score
Training dataset
Clinical model	0.920 (0.874-0.967)	0.908	0.667	0.9612	0.7907	0.9292	0.7234
Radiological model	0.784 (0.708-0.859)	0.852	0.2549	0.9828	0.7647	0.8571	0.3824
Combined model	0.929 (0.876-0.981)	0.915	0.7059	0.9612	0.8000	0.9370	0.7500
Test dataset
Clinical model	0.917 (0.830-1.000)	0.942	0.714	0.990	0.938	0.942	0.811
Radiological model	0.790 (0.688-0.892)	0.842	0.238	0.970	0.625	0.857	0.345
Combined model	0.930 (0.856-1.000)	0.942	0.810	0.970	0.850	0.960	0.829
External validation dataset
Clinical model	0.880 (0.789-0.972)	0.842	0.500	0.919	0.583	0.891	0.539
Radiological model	0.651 (0.459-0.843)	0.816	0.000	1.000	0.000	0.816	0.000
Combined model	0.902 (0.832-0.972)	0.842	0.357	0.952	0.625	0.868	0.455

AUC, area under the curve; PPV, positive predictive value; NPV, negative predictive value.

Figure 4.

ROC curve, calibration curve, and clinical decision curve. (A) ROC curve of the training set; (B) ROC curve of the test set; (C) ROC curve of the external validation set; (D) the training set calibration curve; (E) test set calibration curve; (F) external validation set calibration curve; (G) training set clinical decision curve; (H) test set clinical decision curve; and (I) external validation clinical decision curve

For the clinical model, the P values of the Hosmer–Lemeshow test in the training set, the test set, and the external validation were 0.855 (χ2 = 4.024), < 0.001 (χ2 = 38.926), and <0.001 (χ2 = 2246.729), respectively. For the radiological model model, the P values of the Hosmer–Lemeshow test in the training set, the test set, and the external validation were 0.103 (χ2 = 13.255), 0.090 (χ2 = 13.699), and <0.001 (χ2 = 1340.090), respectively. For the combined model, the P values of the Hosmer–Lemeshow test in the training set, the test set, and the external validation were < 0.001 (χ2 = 44.007), 0.472 (χ2 = 7.611), and <0.001 (χ2 = 4919.666), respectively. The calibration curves demonstrated good calibration of the combined model in both the training, test and external validation sets (Figure 4).

Additionally, the DCA for the clinical, radiological, and combined models is shown in Figure 4, indicating that both the clinical and combined models provide high net benefit, with the combined model slightly outperforming the clinical model. The accuracy, sensitivity, specificity, PPV, NPV and F1 score of all models were detailed in Table 3.

Discussion

In this study, we created a combined models based on ischemic stroke event, admission NIHSS score, BI score, TMT and infarct volume to predict the probability of unfavorable outcome 6 months after AIS. The combined models exhibited good discriminatory potential in the training and test sets, which was validated in the external validation cohort.

Several previous prognostic scores^22-25 and models^10,26 have been developed to predict clinical outcome after AIS. The Houston Intra-arterial Recanalization Therapy score uses age, NIHSS score, hypertension, diabetes mellitus, and atrial fibrillation as independent predictors of poor 3-month outcomes. This score was validated using a substantial dataset from the Virtual International Stroke Trials Archive, with an AUC of 0.75²³. The Houston Intra-arterial Recanalization Therapy score incorporates three independent predictors of unfavorable 3-month outcome in AIS patients undergoing intra-arterial recanalization therapy: age over 75 years, NIHSS score >18, and baseline glucose concentration >150 mg/dL. The score demonstrated AUC values of 0.69 and 0.73 across two distinct cohorts.²⁴ However, it is somewhat limited because all three variables in score determination are dichotomized, which results in a loss of information. Our combined model uses continuous variables to predict the robability of 6-month unfavorable outcome in individual AIS patients. This methodology facilitates enhanced precision and granularity in the assessment of individual patient risk profiles, as continuous variables preserve the complete informational spectrum of clinical data, thereby avoiding the oversimplification inherent in binary categorization. As exemplified by the NIHSS core, which quantifies the severity of neurological deficits on a continuous scale ranging from 0 to 42, is employed, enabling the model to capture nuanced gradations in stroke severity that could be crucial for accurate outcome prognostication. By maintaining the granular integrity of these pivotal clinical parameters, our combined model endeavors to deliver more individualized and refined prognostic accuracy for 6-month outcomes, surpassing the limitations of scores reliant on dichotomized variables.

Several studies have recently explored the effect of infarct lesions on prognosis and treatment of ischemic stroke using radiomics. Xu et al investigated the value of radiomics based on DWI and FLAIR imaging combined with total cerebral small vessel disease burden score in predicting AIS outcome after thrombolysis in 201 patients. The FLAIR model AUC was 0.922 (95% CI, 0.876–0.968).¹² Wang et al developed a DWI-based clinical radiomics nomogram that demonstrated accuracy for prediction of ischemic stroke outcome in both the training dataset (AUC = 0.80; 95% CI, 0.75–0.86] and an independent validation cohort (AUC = 0.73; 95% CI, 0.63–0.82).²⁷ However, radiomics depends heavily on handcrafted feature engineering, which is vulnerable to human bias and may result in a high superfluity of information.²⁸ In our study, the combined model included only five variables: ischemic stroke event, admission NIHSS score, BI score, TMT and infarct volume. Unlike previous tools, our model avoids the complex and time-consuming process of handcrafted feature engineering in radiomics. By focusing on these five readily available and clinically meaningful variables, it reduces the potential for human bias introduced by feature selection and minimizes information redundancy. This not only simplifies the model construction but also enhances its practicality and reproducibility in real-world clinical settings.

Sarcopenia has gained significant research attention due to its robust predictive value for adverse functional outcomes in brain tumor patient.¹⁴ It has also shown predictive value in patients with other neurosurgical conditions such as traumatic brain injury²⁹ and subarachnoid hemorrhage³⁰ as well as in patients with stroke.^6,31-33 The temporal muscle is one of the few muscles that can be fully delineated on routinely performed MRI of the brain. Moreover, TMT is highly correlated with the cross-sectional area of lumbar skeletal muscles and the skeletal muscle index, serving as a proxy for sarcopenia.^34-36 Our findings are consistent with previous studies, indicating that lower TMT is associated with poor prognosis in patients with ischemic stroke. The underlying mechanisms may involve the following factors. First, sarcopenia may be associated with poor outcomes via increased skeletal muscle secretion of the inflammatory cytokines interleukin-6 and interleukin-10 and increased interleukin-6:interleukin-10 ratio.³⁷ In addition, patients with neurosurgical conditions are usually bedridden for some time owing to impaired consciousness, sedation, or debility related to surgery. Skeletal muscle is further catabolized during the treatment and hospitalization period, making rehabilitation difficult for those with low skeletal muscle mass.

The NIHSS score is the most widely used stroke scale.³⁸ Wouters et al showed that ischemic stroke patients with a high NIHSS score at discharge typically had greater disease severity and infarct size. This correlated with their 90-day clinical outcome.³⁹ The BI is widely used to assess activities of daily living, with a higher score indicating a lower risk of poor outcome. Govan et al demonstrated that higher NIHSS and mRS scores, as well as lower BI scores, were associated with more severe strokes, larger infarct volumes in the brain, and worse outcomes at 90 days.⁴⁰ Macciocchi et al identified age, previous stroke, initial neurological deficit, and lesion location as significant predictors of functional outcomes in ischemic stroke patients.⁴¹ In this study, ischemic stroke event, BI score and NHISS score at admission were significantly associated with poor 6-month outcomes in patients with acute ischemic stroke (AIS), consistent with previous research findings.

It is worth noting that generally, we regard the infarct volume as being positively correlated with the prognosis of patients with ischemic stroke. Previous studies⁴² have demonstrated that infarct volume (IV) is a strong independent predictor of functional prognosis in patients with ischemic stroke. Yedavalli Vivek et al⁴³ established that the optimal cutoff value for IV in stroke prognosis assessment is 15 mL, which demonstrates high specificity and sensitivity. When the threshold exceeds 40 mL, a significant reduction in the probability of favorable prognosis is observed. The findings of this study are consistent with most previous literature, indicating a poorer prognosis in patients with larger infarct volumes. However, the infarct volumes in this study were generally smaller, therefore the specific cutoff values for infarct volume in predicting prognosis may differ from those reported in previous studies with larger infarct volume ranges. This suggests that the relationship between infarct volume and prognosis might exhibit some variability depending on the study population’s baseline characteristics, such as the severity of initial stroke, the timing of infarct volume measurement, and potential differences in treatment strategies implemented. Future studies with larger sample sizes encompassing a broader spectrum of infarct volumes, especially including more cases with very small and very large infarcts, would help to further clarify the generalizability and potential adjustments needed for infarct volume thresholds in different clinical contexts. Additionally, subtype analyses based on infarct location could provide more nuanced insights, as infarcts in critical brain regions might have a more profound impact on prognosis even with relatively smaller volumes compared to non-eloquent areas.

Our study has some limitations. First, the study was retrospective in design and selection and attrition biases were present. Second, we did not consider stroke subtype, acute treatments and infarct location.⁴⁴ Future prospective studies should assess whether incorporating novel predictors may improve predictive accuracy. Third, our temporal muscle measurements were based on manual segmentation. Interestingly, a deep learning-based quantification of TMA has been reported.¹⁴ We plan to continue this study using deep learning in the future. Finally, The current model only demonstrates predictive value for 6-month outcomes and cannot dynamically assess corresponding prognostic outcomes as the disease progresses. Future work could incorporate intelligent machine learning methods to fully utilize time-series data (e.g., continuous follow-up mRS scores, NIHSS scores, etc.) for dynamic updates of prognostic predictions, thereby enhancing the model’s accuracy and practicality.^45,46

Conclusion

A combined model using ischemic stroke event, admission NIHSS score, BI score, TMT and infarct volume as input variables may be used to predict 6-month outcome in AIS patients.

Supplemental Material

Supplemental Material - Construction and Validation of a Risk Prediction Model Incorporating Temporal Muscle Thickness for Adverse Outcome in Acute Ischemic Stroke Patients

Supplemental Material for Construction and Validation of a Risk Prediction Model Incorporating Temporal Muscle Thickness for Adverse Outcome in Acute Ischemic Stroke Patients by Huanpeng Wang, Yanchun Wu, Xiaojia Wu, Shuyan Su, Ziting Peng, Minping Lin, Xiaoqin Xu, Dongli Chen, Hong Zhang, Ruibin Huang in Journal of Central Nervous System Disease

Supplemental Material

Supplemental Material - Construction and Validation of a Risk Prediction Model Incorporating Temporal Muscle Thickness for Adverse Outcome in Acute Ischemic Stroke Patients

Footnotes

Acknowledgments

We thank Liwen Bianji (Edanz) () for editing the English text of a draft of this manuscript.

ORCID iDs

Huanpeng Wang

Yanchun Wu

Xiaojia Wu

Shuyan Su

Ziting Peng

Minping Lin

Xiaoqin Xu

Dongli Chen

Ruibin Huang

Ethical Considerations

The study protocol was reviewed and approved by the Ethics Committee of the First Affiliated Hospital of Shantou University of Medical College(no. B-2024-164).

Consent to Participate

This study received ethical approval from the Ethics Committee of the First Affiliated Hospital of Shantou University of Medical College. The present study constitutes a retrospective study into health services, all patient information was de-identified and patient consent was not required.

Author Contributions

ZTP, MPL, XQX, DLC, XJW and SYS acquired data, RBH, HPW, SYS performed the statistical analysis and interpreted data. HPW, YCW and RBH prepared the manuscript. RBH and HZ revised the manuscript. All authors approved the protocol.

Funding

The authors disclosed receipt of the following financial support for the research, authorship, and/or publication of this article: This work was supported by Guangdong Nursing Association Research Fund Project (Grant numbers: GDHLYJYB202430) and 2026 Guangdong Provincial Medical Research Fund Approved Projects (Grant numbers: B2026010).

Declaration of Conflicting Interests

The authors declared no potential conflicts of interest with respect to the research, authorship, and/or publication of this article.

Data Availability Statement

The data in this study are available from the corresponding author on a reasonable request.*

Supplemental Material

Supplemental material for this article is available online.

References

Feigin

Owolabi

SOLN

, et al. Pragmatic solutions to reduce the global burden of stroke: A World Stroke Organization-Lancet Neurology Commission. The Lancet. Neurology. 2023;22:1160-1206. doi:10.1016/S1474-4422(23)00277-6.

Wang

, et al. Temporal trend and attributable risk factors of stroke burden in China, 1990-2019: An analysis for the Global Burden of Disease Study 2019. The Lancet. Public health. 2021;6:e897-e906. doi:10.1016/S2468-2667(21)00228-0.

Slot

Berge

Dorman

Lewis

Dennis

Sandercock

. Impact of functional status at six months on long term survival in patients with ischaemic stroke: Prospective cohort studies. BMJ (Clinical research ed.). 2008;336:376-379. 10.1136/bmj.39456.688333.

Hong

Shi

. Premorbid sarcopenia and functional outcome after acute stroke: A meta-analysis. Asia Pac J Clin Nutr. 2023;32:330-338. doi:10.6133/apjcn.202309_32(3).0004.

Sayer

Cruz-Jentoft

. Sarcopenia definition, diagnosis and treatment: Consensus is growing. Age Ageing. 2022;51:c220. doi:10.1093/ageing/afac220.

Nozoe

Kubo

Kanai

, et al. Reliability and validity of measuring temporal muscle thickness as the evaluation of sarcopenia risk and the relationship with functional outcome in older patients with acute stroke. Clin Neurol Neurosurg. 2021;201:106444. doi:10.1016/j.clineuro.2020.106444.

Ohyama

Watanabe

Nosaki

Hara

Iwai

Mokuno

. Correlation between skeletal muscle mass deficit and poor functional outcome in patients with acute ischemic stroke. Journal of stroke and cerebrovascular diseases : the official journal of National Stroke Association. 2020;29:104623. doi:10.1016/j.jstrokecerebrovasdis.2019.104623.

Nozoe

Kanai

Kubo

Yamamoto

Shimada

Mase

. Prestroke sarcopenia and functional outcomes in elderly patients who have had an acute stroke: A prospective cohort study. Nutrition (Burbank, Los Angeles County, Calif.). 2019;66:44-47. doi:10.1016/j.nut.2019.04.011.

Yoshimura

Wakabayashi

Bise

Tanoue

. Prevalence of sarcopenia and its association with activities of daily living and dysphagia in convalescent rehabilitation ward inpatients. Clinical nutrition (Edinburgh, Scotland). 2018;37:2022-2028. doi:10.1016/j.clnu.2017.09.009.

10.

Sun

Song

, et al. A NADE nomogram to predict the probability of 6-month unfavorable outcome in Chinese patients with ischemic stroke. BMC Neurol. 2019;19:274. doi:10.1186/s12883-019-1464-6.

11.

Pan

Jiang

, et al. Predicting 6-Month unfavorable outcome of acute ischemic stroke using machine learning. Front Neurol. 2020;11:539509. doi:10.3389/fneur.2020.539509.

12.

Zhu

Zhang

, et al. Clinical features and FLAIR radiomics nomogram for predicting functional outcomes after thrombolysis in ischaemic stroke. Front Neurosci. 2023;17:1063391. doi:10.3389/fnins.2023.1063391.

13.

Zhou

Yan

, et al. Feasibility of a Clinical-Radiomics model to predict the outcomes of acute ischemic stroke. Korean J Radiol. 2022;23:811-820. doi:10.3348/kjr.2022.0160.

14.

Mauricaite

Pakzad-Shahabi

Chen

Williams

. Deep learning-based quantification of temporalis muscle has prognostic value in patients with glioblastoma. Br J Cancer. 2022;126:196-203. doi:10.1038/s41416-021-01590-9.

15.

Cooray

Mazya

Bottai

, et al. External validation of the ASTRAL and DRAGON scores for prediction of functional outcome in stroke. Stroke. 2016;47:1493-1499. doi:10.1161/STROKEAHA.116.012802.

16.

Wardlaw

Smith

Biessels

, et al. Neuroimaging standards for research into small vessel disease and its contribution to ageing and neurodegeneration. The Lancet. Neurology. 2013;12:822-838. doi:10.1016/S1474-4422(13)70124-8.

17.

Ding

Nieto

Beauchamp

, et al. A prospective analysis of risk factors for white matter disease in the brain stem: The Cardiovascular Health Study. Neuroepidemiology. 2003;22:275-282. doi:10.1159/000071190.

18.

Legrand

Tisserand

Turc

, et al. Fluid-Attenuated inversion recovery vascular Hyperintensities-Diffusion-Weighted imaging mismatch identifies acute stroke patients most likely to benefit from recanalization. Stroke. 2016;47:424-427. doi:10.1161/STROKEAHA.115.010999.

19.

Gaillard

Schmidt

Costalat

, et al. Hemorrhagic risk of recent silent cerebral infarct on prethrombolysis MR imaging in acute stroke. AJNR. American journal of neuroradiology. 2012;33:227-231. doi:10.3174/ajnr.A2768.

20.

Namgung

Hong

Choi

. Association of temporalis muscle mass with early cognitive impairment in older patients with acute ischemic stroke. J Clin Med. 2023;12:4071. doi:10.3390/jcm12124071.

21.

Dubinski

Won

Mattes

, et al. Frailty in cerebellar ischemic stroke-The significance of temporal muscle thickness. Front Neurol. 2023;14:1193685. doi:10.3389/fneur.2023.1193685.

22.

Ntaios

Faouzi

Ferrari

Lang

Vemmos

Michel

. An integer-based score to predict functional outcome in acute ischemic stroke: The ASTRAL score. Neurology. 2012;78:1916-1922. doi:10.1212/WNL.0b013e318259e221.

23.

Flint

Cullen

Faigeles

Rao

. Predicting long-term outcome after endovascular stroke treatment: The totaled health risks in vascular events score. AJNR. American journal of neuroradiology. 2010;31:1192-1196. doi:10.3174/ajnr.A2050.

24.

Hallevi

Barreto

Liebeskind

, et al. Identifying patients at high risk for poor outcome after intra-arterial therapy for acute ischemic stroke. Stroke. 2009;40:1780-1785. doi:10.1161/STROKEAHA.108.535146.

25.

Weimar

König

Kraywinkel

Ziegler

Diener

. Age and National Institutes of Health Stroke Scale Score within 6 hours after onset are accurate predictors of outcome after cerebral ischemia: Development and external validation of prognostic models. Stroke. 2004;35:158-162. doi:10.1161/01.STR.0000106761.94985.8B.

26.

Flint

Faigeles

Cullen

, et al. THRIVE score predicts ischemic stroke outcomes and thrombolytic hemorrhage risk in VISTA. Stroke. 2013;44:3365-3369. doi:10.1161/STROKEAHA.113.002794.

27.

Wang

Sun

, et al. A Clinical-Radiomics nomogram for functional outcome predictions in ischemic stroke. Neurol Ther. 2021;10:819-832. doi:10.1007/s40120-021-00263-2.

28.

Berenguer

Pastor-Juan

MDR

Canales-Vázquez

, et al. Radiomics of CT features may be nonreproducible and redundant: Influence of CT acquisition parameters. Radiology. 2018;288:407-415. doi:10.1148/radiol.2018172361.

29.

Uhlich

. Sarcopenia diagnosed using masseter muscle area predictive of early mortality following severe traumatic brain injury. Neural Regen Res. 2018;13:2089-2090. doi:10.4103/1673-5374.241451.

30.

Katsuki

Kakizawa

Nishikawa

Yamamoto

Uchiyama

. Temporal muscle thickness and area are an independent prognostic factors in patients aged 75 or younger with aneurysmal subarachnoid hemorrhage treated by clipping. Surgical neurology international. 2021;12:151. doi:10.25259/SNI_814_2020.

31.

Hou

Liu

. Long-term prognostic significance of sarcopenia in acute ischemic stroke. Medicine (Baltimore). 2022;101:e30031. doi:10.1097/MD.0000000000030031.

32.

Sakai

Katayama

Nakajima

, et al. Temporal muscle thickness is associated with the severity of dysphagia in patients with acute stroke. Arch Gerontol Geriatr. 2021;96:104439. doi:10.1016/j.archger.2021.104439.

33.

Nozoe

Kanai

Kubo

Yamamoto

Shimada

Mase

. Prestroke sarcopenia and stroke severity in elderly patients with acute stroke. Journal of stroke and cerebrovascular diseases : the official journal of National Stroke Association. 2019;28:2228-2231. doi:10.1016/j.jstrokecerebrovasdis.2019.05.001.

34.

Ten Cate

Huijs

SMH

Willemsen

ACH

, et al. Correlation of reduced temporal muscle thickness and systemic muscle loss in newly diagnosed glioblastoma patients. J Neurooncol. 2022;160:611-618. doi:10.1007/s11060-022-04180-8.

35.

Leone

Sferruzza

Calimeri

, et al. Quantitative muscle mass biomarkers are independent prognosis factors in primary central nervous system lymphoma: The role of L3-skeletal muscle index and temporal muscle thickness. Eur J Radiol. 2021;143:109945. doi:10.1016/j.ejrad.2021.109945.

36.

Leitner

Pelster

Schöpf

, et al. High correlation of temporal muscle thickness with lumbar skeletal muscle cross-sectional area in patients with brain metastases. PLoS One. 2018;13:e207849. doi:10.1371/journal.pone.0207849.

37.

Rong

Bian

Zhou

. Study on relationship between elderly sarcopenia and inflammatory cytokine IL-6, Anti-Inflammatory cytokine IL-10. BMC Geriatr. 2018;18:308. doi:10.1186/s12877-018-1007-9.

38.

Saber

Saver

. Distributional validity and prognostic power of the national institutes of health stroke scale in US administrative claims data. JAMA Neurol. 2020;77:606-612. doi:10.1001/jamaneurol.2019.5061.

39.

Wouters

Nysten

Thijs

Lemmens

. Prediction of Outcome in Patients with Acute Ischemic Stroke Based on Initial Severity and Improvement in the First 24 h. Front Neurol. 2018;9:308. doi:10.3389/fneur.2018.00308.

40.

Govan

Langhorne

Weir

. Categorizing stroke prognosis using different stroke scales. Stroke. 2009;40:3396-3399. doi:10.1161/STROKEAHA.109.557645.

41.

Macciocchi

Diamond

Alves

Mertz

. Ischemic stroke: Relation of age, lesion location, and initial neurologic deficit to functional outcome. Arch Phys Med Rehabil. 1998;79:1255-1257. doi:10.1016/s0003-9993(98)90271-4.

42.

Boers

AMM

Jansen

IGH

Beenen

LFM

, et al. Association of Follow-Up infarct volume with functional outcome in acute ischemic stroke: A pooled analysis of seven randomized trials. J Neurointerv Surg. 2018;10:1137-1142. doi:10.1136/neurintsurg-2017-013724.

43.

Yedavalli

Salim

Musmar

, et al. Predictive value of follow-up infarct volume on functional outcomes in middle cerebral artery M2 segment vessel occlusion stroke treated with mechanical thrombectomy. Eur Stroke J. 2025;10:431-441. doi:10.1177/23969873241275531.

44.

Cappellari

Turcato

Forlivesi

, et al. Introduction of direct oral anticoagulant within 7 days of stroke onset: A nomogram to predict the probability of 3-month modified Rankin Scale score > 2. J Thromb Thrombolysis. 2018;46:292-298. doi:10.1007/s11239-018-1700-8.

45.

Athanasiou

Clark

. A Bayesian network model for the diagnosis of the caring procedure for wheelchair users with spinal injury. Comput Methods Programs Biomed. 2009;95:S44-S54. doi:10.1016/j.cmpb.2009.02.018.

46.

Zhuang

Zhang

Liang

Liu

Gao

. Adaptive sequential bayesian iterative learning for myocardial motion estimation on cardiac image sequences. IEEE Trans Med Imaging. 2026;45:406-420. doi:10.1109/TMI.2025.3599487.

Supplementary Material

Please find the following supplemental material available below.

For Open Access articles published under a Creative Commons License, all supplemental material carries the same license as the article it is associated with.

For non-Open Access articles published, all supplemental material carries a non-exclusive license, and permission requests for re-use of supplemental material or any part of supplemental material shall be sent directly to the copyright owner as specified in the copyright notice associated with the article.

0.04 MB

0.17 MB