Introduction: Acute ischemic stroke (AIS) stands as a leading cause of death and disability globally. This study aimed to investigate the risk factors and relevance linked with AIS in patients undergoing maintenance hemodialysis (MHD) and to create and validate nomogram models. Methods: We examined the medical records of 314 patients with stage 5 chronic kidney disease (CKD 5) undergoing MHD, who sought neurology outpatient department consultation for suspected AIS symptoms between January 2018 and December 2023. These 314 patients were randomly divided into the training cohort (n = 222) and validation cohort (n = 92). The least absolute shrinkage selection operator (LASSO) regression model was employed for optimal feature selection in the AIS risk model. Subsequently, multivariable logistic regression analysis was used to construct a predictive model incorporating the features selected through LASSO. This predictive model’s performance was assessed using the C-index and the area under the receiver operating characteristic curve (AUC). Additionally, calibration and clinical utility were evaluated through calibration plots and decision curve analysis (DCA). The model’s internal validation was conducted using the validation cohort. Results: Predictors integrated into the prediction nomogram encompassed cardiovascular disease (CVD) (odds ratio [OR] 7.95, 95% confidence interval [CI] 2.400–29.979), smoking (OR 5.7, 95% CI: 1.661–21.955), dialysis time (OR: 5.91, 95% CI: 5.866–29.979), low-density lipoprotein (OR: 2.99, 95% CI: 0.751–13.007), and fibrin degradation products (OR: 5.47, 95% CI: 1.563–23.162). The model exhibited robust discrimination, with a C-index of 0.877 and 0.915 in the internal training and validation cohorts, respectively. The AUC for the training set was 0.857, and a similar AUC of 0.905 was achieved in the validation cohort. DCA demonstrated a positive net benefit within a threshold risk range of 2–96%. Conclusion: The proposed nomogram effectively identifies MHD patients at high risk of AIS at an early stage. This model holds the potential to aid clinicians in making preventive recommendations.

Stroke ranks as the second leading cause of disability and death worldwide, imposing a substantial burden on individuals and society. Globally, an estimated 11.6 million cases of acute ischemic stroke (AIS) were reported [1]. Symptoms of stroke encompass sudden-onset numbness or weakness in limbs, facial drooping, speech difficulties, confusion, coordination problems, and vision loss [2]. Notably, chronic kidney disease (CKD) stands as a significant risk factor for cerebrovascular disease, contributing substantially to the morbidity and mortality associated with stroke. Dialysis patients, in particular, face a significantly heightened risk, experiencing a tenfold increase in incidence, with case fatality rates soaring as high as 90% [3]. This increased risk in dialysis patients can be attributed to a multitude of factors, including shared traditional risk factors like hypertension and diabetes, nontraditional risk factors related to uremia such as oxidative stress and abnormal calcium-phosphorus metabolism, as well as dialysis-specific factors like cerebral hypoperfusion and cardiac structural changes. Furthermore, factors like a history of smoking, elevated cholesterol levels, obesity, low serum albumin levels, and hypercoagulation due to dialysis contribute to the heightened risk of stroke. Previous research has also highlighted the association between inflammation biomarkers, particularly serum C-reactive protein (CRP) levels, and the development of AIS [4]. Additionally, risk factors for stroke closely parallel those for cardiac and peripheral vascular diseases.

Stroke represents a significant complication within the dialysis population, with a higher prevalence and incidence in chronic maintenance hemodialysis (MHD) patients compared to the general population. The presence of CKD complicates stroke risk prediction, diagnosis, management, and prevention. Hence, it becomes crucial to identify strategies that can effectively mitigate the risk of stroke in this patient cohort. Early identification and treatment of AIS play a pivotal role in preventing further morbidity and mortality among individuals with end-stage kidney disease.

To address these critical concerns, this study aimed to investigate the incidence and potential predictors of AIS in MHD patients through an extensive review of medical records. Additionally, we intend to develop a nomogram, primarily based on multifactorial logistic regression analysis, which can serve as a valuable tool for predicting the risk of AIS in MHD patients. This nomogram is envisioned to provide a solid foundation and reference point for medical prevention and intervention strategies for AIS in the MHD patient population.

Study Design and Source of Data

This study adopted a retrospective approach for the development and validation of a prediction model. It involved 314 patients receiving MHD who presented with suspected stroke symptoms at our medical facility (outpatient clinic and emergency department) between January 2018 and December 2023. All data used in this study were approved and obtained from the First Affiliated Hospital of Hainan Medical College. The inclusion criteria consisted of (1) patients diagnosed with CKD stage 5 and undergoing MHD. (2) Patients presenting with suspected stroke symptoms, such as numbness or weakness in an arm or leg, facial droop, difficulty speaking or understanding speech, confusion, trouble with balance or coordination, and loss of vision. (3) Confirmation of AIS diagnosis through diffusion-weighted magnetic resonance imaging scans. (4) Availability of complete general and clinical information for the patient. Exclusion criteria encompassed (1) patients diagnosed with CKD stage 5 and undergoing peritoneal dialysis. (2) Patients with suspected stroke symptoms attributed to a previous cerebral infarction. (3) Incomplete or missing crucial clinical data. Informed consent was verbally obtained through telephone interviews with the participants or their legal representatives, including an explanation of data collection, usage, and protection procedures. Following assessment, 62 out of 314 MHD patients were diagnosed with AIS. “We conducted a survey on the medication history of the MHD patients with combined diabetes and hypertension and collected data on their usage of antidiabetic and antihypertensive drugs. Additionally, the aim of this study was to develop a risk model for MHD patients experiencing their first AIS. As these patients had no prior history of cerebrovascular disease, the survey did not include information on the use of antiplatelet or statin medications.”

Data Collection and Assessment

Data were extracted from the electronic medical records system of the First Affiliated Hospital of Hainan Medical University from January 2018 and December 2023. Patients provided information regarding their medical and pharmaceutical histories. Key study parameters were obtained using a standardized data abstraction form, including individuals’ medical and pharmaceutical histories, laboratory test results, and erythropoietin usage. Demographic variables included gender and age at the time of AIS diagnosis. Lifestyle factors covered current smoking and alcohol consumption status. All patients were followed up by regular records of each clinic recheck or phone calls.

Statistical Analysis

All statistical analyses were conducted using R version 3.3.3 (http://www.r-project.org/). The patients in both the training and validation cohorts were allocated in a 7:3 ratio through the “create Data Partition” function available in the R “caret” package, ensuring random distribution of outcome events. To identify optimal predictive features among the risk factors for MHD patients with AIS, the least absolute shrinkage selection operator (LASSO) regression model, suitable for reducing high-dimensional data [5], was employed. A clinical feature score was calculated for each patient through a linear combination of selected features, weighted by their respective coefficients. Subsequently, a predicting model was constructed using multivariable logistic regression analysis, incorporating the features chosen in the LASSO regression model. Features with nonzero coefficients in the LASSO regression model were selected [6], signifying their high correlation with AIS. These features were presented as odds ratios (ORs) along with their 95% confidence intervals (CIs) and p values. Clinical feature with the p value of ≤0.05 were included in the model, the nomogram was developed using the “rms” package in R software, and its performance was evaluated through the Concordance Index (C-index) to predict accuracy. Additionally, the model’s discriminatory power was assessed by area under the receiver operating characteristic curve (AUC) analysis, with an AUC of 0.75 or higher indicating good discrimination. To further gauge the predictive accuracy of the nomogram, calibration curves were employed to observe the agreement between predicted values and actual values. The clinical utility was evaluated using decision curve analysis (DCA). The nomogram’s validation was performed on the validation cohort, and the net benefit was calculated by subtracting the proportion of false positives from the proportion of true positives, while considering the relative harm of forgoing interventions in comparison to the negative consequences of unnecessary interventions.

Patients’ Characteristics

In total, 314 MHD patients diagnosed with AIS were included in this study, with 222 patients in the training cohort and 92 patients in the validation cohort. The specific inclusion and exclusion processes are shown in Figure 1. Table 1 presents the characteristics of the study cohort. Statistical analysis using SPSS Statistics (Version 26.0) revealed no significant differences (p > 0.05) between the training and validation cohorts. In the training cohort, 44 out of 222 patients (19.8%) were diagnosed with AIS, while in the validation cohort, 19 out of 92 participants (20.6%) received an AIS diagnosis. Based on the Trial of Org 10172 in Acute Stroke Treatment (TOAST) classification, among the 63 patients with AIS, 18 cases were classified as large artery type, 44 cases as small vessel occlusion type, and 1 case as cardioembolic type.

Fig. 1.

Flowchart of patient selection from the surveillance, epidemiology and end results (SEER) database.

Fig. 1.

Flowchart of patient selection from the surveillance, epidemiology and end results (SEER) database.

Close modal
Table 1.

Baseline characteristics of MHD patients with AIS

CharacteristicsTraining cohort = 222Validation cohort = 92p value
Age, n (%) 
 <40 years 13 (5.8) 7 (7.6) 0.068 
 40∼60 years 87 (39.1) 42 (56.6) 
 >60 years 122 (54.9) 43 (46.7) 
Gender, n (%) 
 Male 71 (31.9) 26 (28.2) 0.592 
 Female 151 (68.0) 66 (71.7) 
Hypertension, n (%) 
 Yes 201 (90.5) 78 (84.7) 0.168 
 No 21 (9.4) 14 (15.2) 
Hypertension with antihypertensive therapy, n (%) 
 Yes 178 (80.1) 66 (71.7) 0.152 
 No 44 (019.9) 26 (28.2) 
Diabetes, n (%) 
 Yes 125 (56.3) 61 (66.3) 0.592 
 No 97 (43.6) 31 (33.6) 
Diabetes with hypoglycemic therapy, n (%) 
 Yes 118 (53.1) 59 (64.1) 0.177 
 No 104 (46.8) 33 (35.8) 
Coronary disease, n (%) 
 Yes 52 (23.4) 21 (22.8) 0.517 
 No 170 (76.5) 71 (77.1) 
Smoke, n (%) 
 Yes 90 (40.5) 34 (36.9) 0.613 
 No 132 (59.4) 58 (63.0) 
Dialysis time, n (%) 
 <1 80 (36.0) 33 (35.8) 0.795 
 2–5 125 (56.3) 52 (56.5) 
 6–10 15 (6.7) 6 (5.5) 
 >10 2 (1) 1 (1.0) 
Hemoglobin, n (%) 
 >120 70 (31.5) 32 (34.7) 0.108 
 90–120 77 (34.6) 34 (36.9) 
 60∼90 68 (30.6) 23 (25) 
 <60 7 (3) 3 (3.2) 
Atrial fibrillation, n (%) 
 Yes 6 (2) 5 (5.4) 0.310 
 No 216 (97.2) 87 (94.5) 
Parathyroid hormone, n (%) 
 Normal 31 (13.9) 16 (17.3) 0.488 
 Increased 191 (86.0) 76 (82.6) 
Creatinine, n (%) 
 144–177 19 (85.5) 6 (6.5) 0.410 
 178∼442 60 (27.0) 11 (11.9) 
 443∼707 58 (26.1) 23 (25) 
 >707 85 (38.2) 52 (56.5) 
Blood urea nitrogen, n (%) 
 1–20 104 (46.8) 30 (32.6) 0.433 
 21∼30 88 (39.6) 47 (51.0) 
 >30 30 (13.5) 15 (16.3) 
Calcium, n (%) 
 <2.25 31 (13.9) 13 (14.1) 0.004 
 2.25–2.75 176 (79.2) 70 (76.0) 
 >2.75 15 (6.7) 9 (9.7) 
Phosphorus, n (%) 
 <0.74 42 (18.9) 9 (9.7) 0.227 
 0.74–1.39 88 (39.6) 29 (31.5) 
 >1.39 92 (41.4) 54 (58.6) 
Uric acid, n (%) 
 <200 27 (12.1) 7 (7.6) 0.785 
 200∼400 134 (60.3) 56 (60.8) 
 400∼600 49 (22.0) 22 (23.9) 
 >600 12 (5.4) 7 (1.0) 
Serum albumin, n (%) 
 Normal 66 (29.7) 27 (29.3) 0.120 
 35–30 70 (31.5) 37 (40.2) 
 25–30 55 (24.7) 17 (18.4) 
 <25 31 (13.9) 11 (11.9) 
C-reactive protein, n (%) 
 0–5 105 (47.2) 47 (51) 0.143 
 5–10 38 (17.1) 13 (14.1) 
 10–20 26 (11.7) 13 (14.1) 
 >20 53 (23.8) 19 (20.6) 
Total cholesterol, n (%) 
 <5.18 178 (80.1) 76 (82.6) 0.593 
 5.18–6.19 21 (9.4) 4 (4.3) 
 >6.19 23 (10.3) 12 (13) 
Total triglycerides, n (%) 
 <1.7 164 (73.8) 71 (77.1) 0.039 
 1.7–2.25 21 (9.4) 11 (11.9) 
 >2.25 37 (16.6) 10 (10.8) 
LDL, n (%) 
 <3.37 178 (80.1) 73 (79.3) 0.526 
 3.37–4.14 30 (13.5) 13 (14.1) 
 >4.14 14 (6.3) 6 (6.5) 
High-density lipoprotein, n (%) 
 <1.16 129 (58.1) 56 (60.8) 0.134 
 1.16–1.42 70 (31.5) 26 (28.2) 
 >1.42 23 (10.3) 10 (10.8) 
Prothrombin time, n (%) 
 0<11 2 (1) 1 (1) 0.956 
 11∼13 198 (89.1) 80 (86.9) 
 >13 22 (9.9) 11 (11.9) 
PT-INR, n (%) 
 <0.8 4 (1.8) 2 (1) 0.817 
 0.8–1.5 204 (91.8) 83 (90.2) 
 >1.5 14 (6.3) 7 (7.6) 
Activated partial thromboplastin time, n (%) 
 <23 7 (3.1) 4 (4.3) 0.239 
 23∼37 176 (79.2) 67 (72.8) 
 >37 39 (17.5) 21 (22.8) 
FBG, n (%) 
 <1.8 5 (2.2) 0 (0) 0.076 
 1.8–3.5 74 (33.3) 43 (46.7) 
 >3.5 143 (64.4) 49 (53.2) 
AT-III, n (%) 
 <93 43 (19.3) 20 (21.7) 0.407 
 93–103 175 (78.8) 71 (77.1) 
 >103 4 (1.8) 1 (1) 
FDP, n (%) 
 <5 158 (71.1) 63 (68.4) 0.684 
 >5 64 (28.8) 29 (31.5) 
D-dimer, n (%) 
 Normal 11 (4.9) 3 (3.2) 0.765 
 >0.275 211 (95.0) 89 (96.7) 
Erythropoietin used, n (%) 
 Yes 63 (28.3) 25 (27.1) 0.891 
 No 159 (71.6) 67 (72.8) 
Blood pressure difference, n (%) 
 <30 mm Hg 133 (59.9) 57 (61.9) 0.800 
 >30 mm Hg 89 (40.0) 35 (38.0) 
CharacteristicsTraining cohort = 222Validation cohort = 92p value
Age, n (%) 
 <40 years 13 (5.8) 7 (7.6) 0.068 
 40∼60 years 87 (39.1) 42 (56.6) 
 >60 years 122 (54.9) 43 (46.7) 
Gender, n (%) 
 Male 71 (31.9) 26 (28.2) 0.592 
 Female 151 (68.0) 66 (71.7) 
Hypertension, n (%) 
 Yes 201 (90.5) 78 (84.7) 0.168 
 No 21 (9.4) 14 (15.2) 
Hypertension with antihypertensive therapy, n (%) 
 Yes 178 (80.1) 66 (71.7) 0.152 
 No 44 (019.9) 26 (28.2) 
Diabetes, n (%) 
 Yes 125 (56.3) 61 (66.3) 0.592 
 No 97 (43.6) 31 (33.6) 
Diabetes with hypoglycemic therapy, n (%) 
 Yes 118 (53.1) 59 (64.1) 0.177 
 No 104 (46.8) 33 (35.8) 
Coronary disease, n (%) 
 Yes 52 (23.4) 21 (22.8) 0.517 
 No 170 (76.5) 71 (77.1) 
Smoke, n (%) 
 Yes 90 (40.5) 34 (36.9) 0.613 
 No 132 (59.4) 58 (63.0) 
Dialysis time, n (%) 
 <1 80 (36.0) 33 (35.8) 0.795 
 2–5 125 (56.3) 52 (56.5) 
 6–10 15 (6.7) 6 (5.5) 
 >10 2 (1) 1 (1.0) 
Hemoglobin, n (%) 
 >120 70 (31.5) 32 (34.7) 0.108 
 90–120 77 (34.6) 34 (36.9) 
 60∼90 68 (30.6) 23 (25) 
 <60 7 (3) 3 (3.2) 
Atrial fibrillation, n (%) 
 Yes 6 (2) 5 (5.4) 0.310 
 No 216 (97.2) 87 (94.5) 
Parathyroid hormone, n (%) 
 Normal 31 (13.9) 16 (17.3) 0.488 
 Increased 191 (86.0) 76 (82.6) 
Creatinine, n (%) 
 144–177 19 (85.5) 6 (6.5) 0.410 
 178∼442 60 (27.0) 11 (11.9) 
 443∼707 58 (26.1) 23 (25) 
 >707 85 (38.2) 52 (56.5) 
Blood urea nitrogen, n (%) 
 1–20 104 (46.8) 30 (32.6) 0.433 
 21∼30 88 (39.6) 47 (51.0) 
 >30 30 (13.5) 15 (16.3) 
Calcium, n (%) 
 <2.25 31 (13.9) 13 (14.1) 0.004 
 2.25–2.75 176 (79.2) 70 (76.0) 
 >2.75 15 (6.7) 9 (9.7) 
Phosphorus, n (%) 
 <0.74 42 (18.9) 9 (9.7) 0.227 
 0.74–1.39 88 (39.6) 29 (31.5) 
 >1.39 92 (41.4) 54 (58.6) 
Uric acid, n (%) 
 <200 27 (12.1) 7 (7.6) 0.785 
 200∼400 134 (60.3) 56 (60.8) 
 400∼600 49 (22.0) 22 (23.9) 
 >600 12 (5.4) 7 (1.0) 
Serum albumin, n (%) 
 Normal 66 (29.7) 27 (29.3) 0.120 
 35–30 70 (31.5) 37 (40.2) 
 25–30 55 (24.7) 17 (18.4) 
 <25 31 (13.9) 11 (11.9) 
C-reactive protein, n (%) 
 0–5 105 (47.2) 47 (51) 0.143 
 5–10 38 (17.1) 13 (14.1) 
 10–20 26 (11.7) 13 (14.1) 
 >20 53 (23.8) 19 (20.6) 
Total cholesterol, n (%) 
 <5.18 178 (80.1) 76 (82.6) 0.593 
 5.18–6.19 21 (9.4) 4 (4.3) 
 >6.19 23 (10.3) 12 (13) 
Total triglycerides, n (%) 
 <1.7 164 (73.8) 71 (77.1) 0.039 
 1.7–2.25 21 (9.4) 11 (11.9) 
 >2.25 37 (16.6) 10 (10.8) 
LDL, n (%) 
 <3.37 178 (80.1) 73 (79.3) 0.526 
 3.37–4.14 30 (13.5) 13 (14.1) 
 >4.14 14 (6.3) 6 (6.5) 
High-density lipoprotein, n (%) 
 <1.16 129 (58.1) 56 (60.8) 0.134 
 1.16–1.42 70 (31.5) 26 (28.2) 
 >1.42 23 (10.3) 10 (10.8) 
Prothrombin time, n (%) 
 0<11 2 (1) 1 (1) 0.956 
 11∼13 198 (89.1) 80 (86.9) 
 >13 22 (9.9) 11 (11.9) 
PT-INR, n (%) 
 <0.8 4 (1.8) 2 (1) 0.817 
 0.8–1.5 204 (91.8) 83 (90.2) 
 >1.5 14 (6.3) 7 (7.6) 
Activated partial thromboplastin time, n (%) 
 <23 7 (3.1) 4 (4.3) 0.239 
 23∼37 176 (79.2) 67 (72.8) 
 >37 39 (17.5) 21 (22.8) 
FBG, n (%) 
 <1.8 5 (2.2) 0 (0) 0.076 
 1.8–3.5 74 (33.3) 43 (46.7) 
 >3.5 143 (64.4) 49 (53.2) 
AT-III, n (%) 
 <93 43 (19.3) 20 (21.7) 0.407 
 93–103 175 (78.8) 71 (77.1) 
 >103 4 (1.8) 1 (1) 
FDP, n (%) 
 <5 158 (71.1) 63 (68.4) 0.684 
 >5 64 (28.8) 29 (31.5) 
D-dimer, n (%) 
 Normal 11 (4.9) 3 (3.2) 0.765 
 >0.275 211 (95.0) 89 (96.7) 
Erythropoietin used, n (%) 
 Yes 63 (28.3) 25 (27.1) 0.891 
 No 159 (71.6) 67 (72.8) 
Blood pressure difference, n (%) 
 <30 mm Hg 133 (59.9) 57 (61.9) 0.800 
 >30 mm Hg 89 (40.0) 35 (38.0) 

Feature Selection

In the initial phase, we employed LASSO regression to preliminarily select predictors of AIS. To ensure robustness, we centralized and normalized variables using 10-fold cross-validation. Subsequently, we identified six predictors as independent risk variables for constructing a prediction model through multivariable logistic regression. Among the initial 30 features, we reduced them to five potential predictors based on the data from 244 patients in the cohort (approximately a 6:1 ratio; see Fig. 2a, b). These five predictors were as follows: cardiovascular disease (CVD) (OR: 7.95, 95% CI: 2.400–29.979), smoking (OR: 5.7, 95% CI: 1.661–21.955), dialysis time (OR: 5.91, 95% CI: 5.866–29.979), low-density lipoprotein (LDL) (OR: 2.99, 95% CI: 0.751–13.007), and fibrin degradation product (FDP) (OR: 5.47, 95% CI: 1.563–23.162).

Fig. 2.

Demographic and clinical feature selection using LASSO binary logistic regression model. a Optimal parameter (lambda) selection in the LASSO model was determined through fivefold cross-validation using the minimum criteria. The partial likelihood deviance (binomial deviance) curve was plotted against log(lambda). Dotted vertical lines were drawn at the optimal values based on the minimum criteria and the 1 SE of the minimum criteria (the 1-SE criteria). b LASSO coefficient profiles of the 30 features. A coefficient profile plot was generated against the log(lambda) sequence. A vertical line was drawn at the value selected using fivefold cross-validation, resulting in optimal lambda, which retained five features with nonzero coefficients. LASSO, least absolute shrinkage and selection operator; SE, standard error.

Fig. 2.

Demographic and clinical feature selection using LASSO binary logistic regression model. a Optimal parameter (lambda) selection in the LASSO model was determined through fivefold cross-validation using the minimum criteria. The partial likelihood deviance (binomial deviance) curve was plotted against log(lambda). Dotted vertical lines were drawn at the optimal values based on the minimum criteria and the 1 SE of the minimum criteria (the 1-SE criteria). b LASSO coefficient profiles of the 30 features. A coefficient profile plot was generated against the log(lambda) sequence. A vertical line was drawn at the value selected using fivefold cross-validation, resulting in optimal lambda, which retained five features with nonzero coefficients. LASSO, least absolute shrinkage and selection operator; SE, standard error.

Close modal

Development of an Individualized Prediction Model

The results of the logistic regression analysis for the variables smoke, CVD, dialysis time, LDL, and FDP are summarized in Table 2. The predictive model, incorporating these independent variables, was developed and presented as a nomogram using the training cohort (see Fig. 3). The C-index for the prediction nomogram was calculated as 0.877 (95% CI: 0.820–0.907) for the training cohort. This metric reflects the nomogram’s ability to distinguish between patients with different outcome events. The ROC curves are depicted in Figure 4a, demonstrating that the nomogram displayed relatively good discriminative performance with an AUC value of 0.857.

Table 2.

Prediction factors for ischemic stroke in patients undergoing MHD

Intercept and variablePrediction model
βOR (95% CI)p value
Coronary disease 2.0738 7.95 (2.400–29.979) 0.00113 
Smoke 1.7406 5.70 (1.661–21.955) 0.00743 
Dialysis time 4.0804 5.91 (5.866–16.986) 0.00266 
LDL 1.0977 2.99 (0.751–13.007) 0.00819 
FDP 1.7001 5.47 (1.563–23.162) 0.01217 
Intercept and variablePrediction model
βOR (95% CI)p value
Coronary disease 2.0738 7.95 (2.400–29.979) 0.00113 
Smoke 1.7406 5.70 (1.661–21.955) 0.00743 
Dialysis time 4.0804 5.91 (5.866–16.986) 0.00266 
LDL 1.0977 2.99 (0.751–13.007) 0.00819 
FDP 1.7001 5.47 (1.563–23.162) 0.01217 

Note: β is the regression coefficient.

Fig. 3.

Nomogram for estimating the risk of ischemic stroke. The corresponding score for each indicator can be found by moving vertically down, and the total score for each patient can be obtained on the total points scale.

Fig. 3.

Nomogram for estimating the risk of ischemic stroke. The corresponding score for each indicator can be found by moving vertically down, and the total score for each patient can be obtained on the total points scale.

Close modal
Fig. 4.

Calibration plots of the ischemic stroke nomogram prediction for the training cohort (a) and validation cohort (b). The x-axis represents the predicted ischemic stroke risk, while the y-axis represents the actual ischemic stroke rate. The diagonal dotted line signifies a perfect prediction by an ideal model, while the solid line represents the performance of the nomogram. A closer alignment of the solid line with the diagonal dotted line indicates a better predictive performance of the nomogram.

Fig. 4.

Calibration plots of the ischemic stroke nomogram prediction for the training cohort (a) and validation cohort (b). The x-axis represents the predicted ischemic stroke risk, while the y-axis represents the actual ischemic stroke rate. The diagonal dotted line signifies a perfect prediction by an ideal model, while the solid line represents the performance of the nomogram. A closer alignment of the solid line with the diagonal dotted line indicates a better predictive performance of the nomogram.

Close modal

Evaluation and Validation of Model Performance

The calibration curve of the AIS risk nomogram, used to predict AIS risk in MHD patients, exhibited strong agreement in both the training and validation cohorts, as shown in Figure 4. Verification was conducted in the validation cohort by comparing the predicted probability generated by the nomogram with the actual probability for each patient. The C-index for the prediction nomogram was calculated as 0.915 (95% CI: 0.856–0.946) for the training cohort. Additionally, the AUC of the predictive nomogram was determined as 0.905 (95% CI: 0.801–1.000), as illustrated in Figure 5. These results signify the model’s robust discrimination capability.

Fig. 5.

ROC curves in the training cohort (a) and validation cohort (b).

Fig. 5.

ROC curves in the training cohort (a) and validation cohort (b).

Close modal

Clinical Use

The DCA curve was developed to assess the clinical benefits and utility of the nomogram. Both the training and validation cohorts demonstrated that the nomogram provided a net benefit in predictive models for threshold probabilities at different time points, as depicted in Figure 6. DCA was employed to evaluate the clinical utility, particularly the ability to enhance decision-making, of the prediction models by quantifying the net benefits at various threshold probabilities. The DCA indicated that when a patient’s threshold probability is 2% and a doctor’s threshold probability is 98%, using this AIS nomogram to predict AIS risk offers more significant benefits compared to other schemes.

Fig. 6.

DCA curves for the nomogram in the training cohort (a) and validation cohort (b).

Fig. 6.

DCA curves for the nomogram in the training cohort (a) and validation cohort (b).

Close modal

Patients with stage 5 CKD face an elevated risk of stroke compared to non-CKD patients. Traditional risk factors and mechanisms encompass hypertension, diabetes, atrial fibrillation, carotid artery disease, heart failure, obesity, and dyslipidemia, all of which are often comorbid with CKD and exacerbated in its presence [7]. The incidence of stroke is notably higher among MHD patients compared to those in earlier CKD stages. In this study, we have developed a straightforward, valid, and clinically valuable model for predicting the likelihood of AIS in MHD patients. We have identified five predictive factors for ischemic stroke: smoking, CVD, dialysis time, LDL, and FDP. Importantly, our nomogram exhibits the capacity to predict patient-specific AIS probabilities with exceptional discrimination and calibration.

Significant temporal relationships exist concerning the timing of dialysis initiation and stroke risk. The period surrounding dialysis initiation, including the 30-day period both before and after, is associated with a threefold increase in the risk of stroke, transient ischemic attack, or recurrent stroke [7]. The day following a prolonged interdialytic gap also carries a higher risk than other days. Regarding dialysis modality, hemodialysis patients appear to have a higher stroke risk compared to peritoneal dialysis patients [8]. Consequently, based on our research findings, we assert that dialysis time can be incorporated as one of the predictors in constructing the AIS Nomogram model (p = 0.00266).

Smoking constitutes a crucial modifiable risk factor for the development of CVD and stroke. Plausible mechanisms include carboxyhemoglobinemia, increased platelet aggregability, elevated fibrinogen levels, and reduced high-density lipoprotein (HDL) cholesterol [9]. Our predictive model has identified that CKD patients with a history of smoking have an elevated probability of experiencing ischemic stroke, thus underscoring smoking’s significance as a risk factor in the model construction (p = 0.00743).

Uremia can lead to protein carbamylation, which exerts proatherosclerotic effects by exacerbating dyslipidemia. In our investigation of dyslipidemia, we explored four factors: total cholesterol, total triglycerides, LDL, and high-density lipoprotein. Our study has revealed that LDL has statistical significance in model construction (p = 0.00819). Elevated LDL levels increase the risk of stroke and vascular events. A previous trial demonstrated that patients achieving an LDL cholesterol level of less than 70 mg per deciliter (1.8 mmol per liter) had a 28% lower relative risk of stroke compared to those attaining a level of 100 mg per deciliter (2.6 mmol per liter) [10].

Numerous risk factors are common to both coronary heart disease and ischemic stroke. Carotid atherosclerosis, in particular, has gained recognition as a risk factor for both conditions [11]. Previous studies have regarded kidney dysfunction as an independent risk factor for carotid atherosclerosis in acute stroke patients [12]. Hence, it is unsurprising that we have identified a significantly higher incidence of ischemic stroke in CKD patients with coronary heart disease, underscoring its importance as a predictive factor in constructing our nomogram (p = 0.00113).

CKD is linked to a chronic inflammatory state, resulting in elevated levels of fibrinogen, factor VIII, von Willebrand factor, and C-reactive protein [13]. A majority of dialysis patients are in a hypercoagulable state, further exacerbated by the hemodialysis procedure itself [13]. Research has indicated that abnormal plasma levels of D-dimer and FDPs in ischemic stroke patients reflect anomalies in the coagulation-fibrinolysis system. D-dimer, a characteristic product of cross-linked fibrin degraded by plasmin, is elevated in plasma, suggesting vascular thrombosis and secondary fibrinolysis. It serves as an important indicator for diagnosing thrombotic diseases and indicating a hypercoagulable state [14]. FDP plays a pivotal role in thrombus formation and atherosclerosis development, making it a marker for excessive fibrinolysis. Therefore, our study suggests that FDP can be utilized as a predictor in constructing prediction models (p = 0.01217).

Furthermore, we conducted an analysis to understand why certain features included in the study were not suitable as predictors for constructing the nomogram. Hypertension, as a risk factor, did not emerge as a predictor for constructing the nomogram through logistic regression in our study. We believe this discrepancy can be attributed to the following reasons: CKD serves both as a common cause of hypertension and as a complication of uncontrolled hypertension. Consequently, hypertension often coexists as a symptom of CKD, making it a less specific predictor for AIS in our study. A similar situation arose with diabetes. A meta-analysis involving 1,024,977 participants (with nearly 13% having diabetes) from 30 general population and high-risk cardiovascular cohorts, as well as 13 CKD cohorts, revealed that although the absolute risks for all-cause and CVD mortality are elevated in the presence of diabetes, irrespective of whether diabetes is present or not [15]. These findings emphasize the significance of kidney disease as a predictor of critical clinical outcomes, regardless of the underlying cause of kidney disease, such as diabetes or hypertension.

However, hyperphosphatemia was not included as a predictive factor in our nomogram. Among the 314 patients in our study, 267 of them exhibited abnormally elevated levels of parathyroid hormone. The primary initial abnormalities in serum biochemistry involve increases in PTH while maintaining normal serum calcium and phosphate levels. Subsequently, serum 25-hydroxy vitamin D (calcidiol) decreases, and in later CKD stages, hyperphosphatemia develops in the majority of patients [16].

A previous study investigated the role of inflammation in predicting stroke in patients with CKD and concluded that inflammation is a contributing factor to stroke etiology in patients both with and without CKD. However, it is not the predominant driver of the worse outcomes observed in the CKD and stroke population [17]. This result has also been confirmed in our research, where C-reactive protein levels were abnormally elevated in both the ischemic stroke and non-stroke groups. Nevertheless, this finding lacks statistical significance to form a predictive model. It is possible that there are unmeasured confounders or alternative pathophysiological mechanisms that remain undiscovered or unexplored.

A Scottish cohort study reported that dialysis patients were less frequently admitted to an acute stroke unit (64.6% vs. 79.6%; p < 0.001) and were less likely to receive aspirin acutely (75.3% vs. 83.2%; p = 0.01) [18]. Therefore, it is crucial for us to construct a predictive model for the early detection of brain infarction in dialysis patients. By identifying dialysis patients with a high risk of stroke through the model, proactive stroke prevention measures should be initiated, and prompt treatment in a stroke unit should be provided when a stroke occurs.

We have developed a risk prediction model for AIS in patients with MHD. The nomogram model obtained in this study is convenient and shows better predictive performance than previous models. Therefore, it might be a promising tool to assess the individual risk of developing AIS for patients with MHD and thus facilitate preventive measures. This risk model provides valuable insights for identifying AIS risk and implementing preventive measures. The results are inconclusive and that proper validation must be done in large, diverse, and prospective cohorts. Nonetheless, our study also possesses certain limitations. First, being a retrospective study, it is susceptible to selection bias. Therefore, larger sample sizes and multicenter clinical studies are necessary to further validate the model. Second, the data sources for our nomogram were solely based on retrospective analysis of a single-center database. Third, the survey did not include information on the use of antiplatelet or statin medications. To confirm the model’s generalizability to other centers, we need to conduct a prospective study for further reliability validation. Additionally, we did not analyze the relationship between intradialytic hemodynamic instability and AIS. It is necessary to investigate the connection between changes in blood volume and post-dialysis AIS.

The authors would like to thank the Department of Neurology, the First Affiliated Hospital of Hainan Medical University, for its technical support.

The present work complies with the guidelines for human studies and was conducted ethically in accordance with the World Medical Association Declaration of Helsinki. This study was approved by the First Affiliated Hospital of Hainan Medical University Ethics Committee, and the study approval number is Hyfy0307821. Written informed consent to participate in the study has been obtained from all adult participants and all vulnerable participants’ legal guardian.

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

This study was supported by grants from Hainan Province Clinical Medical Center (2021) and the Excellent Talent Team of Hainan Province (No. QRCBT202121).

Jing Yi Tong and Tingting Ji contributed to the research idea and study design; Jing Yi Tong, Tingting Ji, and Qifu Li were involved in data acquisition; Jing Yi Tong, Tingting Ji, Nan Liu, and Yibin Chen were involved in data interpretation; Jing Yi Tong, Xuejuan Lin, Zongjun Li, and Yi Xing contributed to statistical analysis and manuscript drafting; each author contributed important intellectual content during manuscript drafting or revision and accepted accountability for the overall work by ensuring that questions pertaining to the accuracy or integrity of any portion of the work are appropriately investigated and resolved.

Additional Information

Jingyi Tong and Tingting Ji should be considered as co-first authors.

Data are not available due to ethical reasons. Further inquiries can be directed to the corresponding author.

1.
Saini
V
,
Guada
L
,
Yavagal
DR
.
Global epidemiology of stroke and access to acute ischemic stroke interventions
.
Neurology
.
2021
;
97
(
20 Suppl 2
):
S6
16
.
2.
Walter
K
.
What is acute ischemic stroke
.
JAMA
.
2022
;
327
(
9
):
885
.
3.
Iseki
K
.
Stroke feature and management in dialysis patients
.
Contrib Nephrol
.
2013
;
179
:
100
9
.
4.
Chou
CY
,
Kuo
HL
,
Lin
HH
,
Liu
JS
,
Liu
YL
,
Huang
CC
.
C-reactive protein predicts ischaemic stroke in haemodialysis patients
.
Int J Clin Pract
.
2009
;
63
(
2
):
243
8
.
5.
Sauerbrei
W
,
Royston
P
,
Binder
H
.
Selection of important variables and determination of functional form for continuous predictors in multivariable model building
.
Stat Med
.
2007
;
26
(
30
):
5512
28
.
6.
Kidd
AC
,
McGettrick
M
,
Tsim
S
,
Halligan
DL
,
Bylesjo
M
,
Blyth
KG
.
Survival prediction in mesothelioma using a scalable LASSO regression model: instructions for use and initial performance using clinical predictors
.
BMJ Open Respir Res
.
2018
;
5
(
1
):
e000240
.
7.
Murray
AM
,
Seliger
S
,
Lakshminarayan
K
,
Herzog
CA
,
Solid
CA
.
Incidence of stroke before and after dialysis initiation in older patients
.
J Am Soc Nephrol
.
2013
;
24
(
7
):
1166
73
.
8.
Foley
RN
,
Gilbertson
DT
,
Murray
T
,
Collins
AJ
.
Long interdialytic interval and mortality among patients receiving hemodialysis
.
N Engl J Med
.
2011
;
365
(
12
):
1099
107
.
9.
Shah
RS
,
Cole
JW
.
Smoking and stroke: the more you smoke the more you stroke
.
Expert Rev Cardiovasc Ther
.
2010
;
8
(
7
):
917
32
.
10.
Amarenco
P
,
Kim
JS
,
Labreuche
J
,
Charles
H
,
Abtan
J
,
Béjot
Y
, et al
.
A comparison of two LDL cholesterol targets after ischemic stroke
.
N Engl J Med
.
2020
;
382
(
1
):
9
.
11.
Román
GC
,
Jackson
RE
,
Gadhia
R
,
Román
AN
,
Reis
J
.
Mediterranean diet: the role of long-chain ω-3 fatty acids in fish; polyphenols in fruits, vegetables, cereals, coffee, tea, cacao and wine; probiotics and vitamins in prevention of stroke, age-related cognitive decline, and Alzheimer disease
.
Rev Neurol Paris
.
2019
;
175
(
10
):
724
41
.
12.
Yu
FP
,
Zhao
YC
,
Gu
B
,
Hu
J
,
Yang
YY
.
Chronic kidney disease and carotid atherosclerosis in patients with acute stroke
.
Neurologist
.
2015
;
20
(
2
):
23
6
.
13.
Taher
AT
,
Cappellini
MD
,
Bou-Fakhredin
R
,
Coriu
D
,
Musallam
KM
.
Hypercoagulability and vascular disease
.
Hematol Oncol Clin North Am
.
2018
;
32
(
2
):
237
45
.
14.
Campello
E
,
Spiezia
L
,
Adamo
A
,
Simioni
P
.
Thrombophilia, risk factors and prevention
.
Expert Rev Hematol
.
2019
;
12
(
3
):
147
58
.
15.
Pavkov
ME
,
Collins
AJ
,
Coresh
J
,
Nelson
RG
.
Kidney disease in diabetes
.
National Institute of Diabetes and Digestive and Kidney Diseases (US)
;
2018
. CHAPTER 22.
16.
Drüeke
TB
.
Hyperparathyroidism in chronic kidney disease
. In:
Wilson
DP
, editor.
Endotext
.
South Dartmouth (MA)
:
MDText.com, Inc.
;
2000
.
17.
Tollitt
J
,
Allan
SM
,
Chinnadurai
R
,
Odudu
A
,
Hoadley
M
,
Smith
C
, et al
.
Does previous stroke modify the relationship between inflammatory biomarkers and clinical endpoints in CKD patients
.
BMC Nephrol
.
2022
;
23
(
1
):
38
.
18.
Findlay
MD
,
Dawson
J
,
MacIsaac
R
,
Jardine
AG
,
MacLeod
MJ
,
Metcalfe
W
, et al
.
Inequality in care and differences in outcome following stroke in people with ESRD
.
Kidney Int Rep
.
2018
;
3
(
5
):
1064
76
.