Introduction

Cerebral small vessel disease (CSVD) comprises a group of diseases involving cerebral arterioles, venules, and capillaries and is an important cause of cognitive dysfunction in elderly individuals [1,2,3]; lesions can be observed by neuroimaging or pathology [4]. Neuroimaging, represented by brain magnetic resonance imaging (MRI), plays an irreplaceable role in the diagnosis and classification of CSVD. Among the MRI features of CSVD, lacunar infarction (LI), cerebral microbleeds (CMBs), white matter hyperintensities (WMHs), and enlarged perivascular space (EPVS) are the four typical neuroimaging manifestations of CSVD [2]. Clinically, these neuroimaging changes that cause brain injury do not occur alone in the elderly population but mostly appear as two or more superimposed manifestations. Previous studies on the relationship between CSVD imaging markers and cognitive function have focused mainly on one or two imaging markers, which cannot comprehensively predict the risk of pathological changes in the brain during the progression of cognitive dysfunction [5, 6]. Klarenbeek. et al. [7] first proposed the CSVD total burden score in 2013; that is, four findings (LI, WMH, CMBs, and EPVS) on the MRI of the same patient were scored separately, and the total burden score was subsequently calculated to more comprehensively assess CSVD-related brain injury, determine the severity of CSVD, and determine the importance of the severity of cognitive impairment.

Currently, there is no clear explanation for the disease pathogenesis. Many studies have focused on the relationship between the total burden of CSVD and the progression of cognitive decline [1, 6, 8], but it is not yet common to construct a more readable clinical prediction model of cognitive impairment from the perspective of the total burden of CSVD. The nomogram is a new statistical prediction model that transforms risk factors into continuous scoring graphs, which are highly readable [9]. A nomogram can graphically represent the impact of each predictor on the outcome, providing readers with a more specific interpretation of each predictor’s impact on the outcome [10, 11]. The nomogram simplifies the prediction model into a numerical estimate of the probability of death or recurrence, which can guide clinical decision-making and is widely used for cancer prognosis, mortality in acute cerebrovascular disease, and risk of neurological deterioration [12,13,14,15,16,17]. This research aims to further enrich the necessary findings. In this study, risk factors for cognitive dysfunction in CSVD patients were investigated, and a predictive model was constructed for visualizing the nomogram from the perspective of the total burden of CSVD to provide a more efficient and convenient way for clinicians to predict the progression of cognitive dysfunction in patients with CSVD.

Materials and methods

Study design and patient selection

We included patients with cerebral small vessel disease in the Department of Neurology of the Second Affiliated Hospital of Bengbu Medical College. CSVD patients from September 2021 to October 2022 were included as training data. Patients with CSVD from October 2019 to September 2020 were included as test data. Demographic data, clinical history, relevant blood markers, and brain MRI images were recorded for all study subjects. Cognitive function was tested using the MoCA scale. The training data was a cross-sectional observational study based on retrospective analysis, and the data was based on electronic medical record acquisition, and the patient or family member had signed an informed consent form indicating that the medical record information could be used for clinical research. The test data was based on a prospective observational study in the past, and all subjects or family members also signed an informed consent form and agreed to join the clinical study. The inclusion criteria of the two sets of datasets are the same. Individuals with cognitive dysfunction due to metabolic, nutritional, or toxic factors, such as hypothyroidism, vitamin B deficiency, or alcoholism, were excluded from the study. Patients with traumatic brain injury, central nervous system infection, hydrocephalus, or tumors were also excluded. Furthermore, patients with severe psychiatric disease, delirium, hearing loss, visual impairment, or severe psychiatric symptoms that interfere with cognitive assessments were excluded. In this study, patients with WMH caused by demyelinating diseases, radiotherapy, or chemotherapy were excluded. Patients who suffered from large vessel infarctions or severe brain hemorrhages were also excluded. Finally, 227 patients met the criteria for the training data (Fig. 1). The test data included 70 participants, the average age was 67.71 (9.14) years, 16 cases of cognitive impairment appeared, and the incidence of cognitive impairment was 22.86%. Subjects with training data and test data underwent clinical evaluations, neurological examinations, and cognitive assessments conducted at the first visit, followed by an MRI within one week.

Fig. 1
figure 1

Flowchart of patient selection. Note: Abbreviations: MRI, magnetic resonance imaging; CSVD, cerebral small vessel disease; MoCA, Montreal Cognitive Assessment

Full size image

Clinical data collection

The following demographic characteristics and vascular risk factors were recorded: sex, age, years of education, previous or current smoking status (on average, one cigarette per day and a one-year history of smoking), history of alcohol consumption (an average of 50 ml of liquor consumed daily, 1 time a week, and a one-year drinking history), history of hypertension (any of the following conditions: ①rest quietly for 10 min, and blood pressure measurements on the same day are found to exceed the normal blood pressure standard at least 3 times: systolic blood pressure (SBP) ≥ 140 mmHg and/or diastolic blood pressure (DBP) ≥ 90 mmHg; ②currently taking antihypertensive medications; ②Denying a history of hypertension, but being diagnosed with hypertension by a clinician), history of diabetes (this study refers to type 2 diabetes mellitus, which meets any of the following criteria. The diagnostic criteria are as follows: ① fasting blood glucose ≥ 7.0 mmol/L; ② random blood glucose ≥ 11.1mmol/L; ② Previously diagnosed with diabetes, regardless of current hypoglycemic drugs or insulin therapy), coronary heart disease (patients with a previous clinical diagnosis of coronary heart disease (asymptomatic myocardial ischemia, angina, myocardial infarction, ischemic heart failure) or transcoronary angiography showing varying degrees of coronary artery stenosis), previous stroke (those who have had a transient ischemic attack or symptomatic ischemic stroke in the past), triglycerides, total cholesterol, high-density lipoprotein, low-density protein, serum homocysteine (HCY) and uric acid and serum glucose tests.

Neuropsychological assessment

On the first visit, a trained neuropsychologist evaluated cognitive function using the Chinese version of the Montreal Cognitive Assessment (MoCA) scale. An impairment in cognitive function was identified as a MoCA score < 26 (an additional 1 point for education < 12 years) [9, 18]. The MoCA score in the data analysis of this study was defined as MoCA_score.

Magnetic resonance imaging acquisition

All patients underwent an MRI scan within 7 days after the first visit. MRI was performed on a 1.5T scanner, and T1-weighted imaging, T2-weighted imaging, fluid-attenuated inversion recovery (FLAIR), and susceptibility-weighted imaging (SWI) were performed for each individual.

Total MRI CSVD burden assessment

The CSVD imaging features and clinical information were independently evaluated by two neurologists and radiologists using a blinded method. If there were differences in scores, the authors negotiated to reach an agreement. We defined lacunes as small and involved the subcortical area with a diameter of 3 mm to 20 mm. Circular or ovoid lesions remaining after infarction tissue or bleeding absorption showed a CSF-like signal in a FLAIR sequence with a ring of high signal around it and a CSF-like signal in T1-weighted imaging and T2-weighted imaging. Some studies have referred to these lesions as asymptomatic lacunar infarction [1, 19]. White matter hyperintensities (WMHs) can be divided into periventricular WMHs and deep WMHs. MRI reveals a high signal density on T2-weighted imaging and no significantly low signal density on T1-weighted imaging [1, 19], and WMHs are graded using the modified Fazekas scale [20]. Cerebral microbleeds (CMBs) appear as well-defined circular or oval hypo-homogeneous signals with a diameter of 2 mm to 10 mm on magnetic resonance imaging (MRI) [19, 21]. Enlarged perivascular spaces (EPVSs) refer to circular or tubular spaces with a diameter < 2 mm in the subcortical area and no high signal surrounding the lesion on FLAIR sequence and no absorbent bleeding lesion on axial T2-weighted gradient echo sequence [1, 19].

The CSVD total burden refers to the score constructed by the CSVD for the 4 imaging signs on MR images (LI, CMBs, WMH, and EPVS) [22]. Previous studies have quantitatively assessed the severity of CSVD by constructing latent variable models and developed an overall CSVD burden score scale [23]. The scoring criteria [24] for the overall burden of CSVD were as follows: (1) a visual scoring system for the severity of white matter lesions, Fazekas scale score ≥ 2 points for 1 point; (2) one or more lacunar infarction counts as 1 score; (3) one or more deep cerebral microbleeds count as 1 point; and (4) 10 or more basal segment EPVS counts as 1 point. Among them, WMH was graded according to the Fazekas scale: (1) Periventricular WMH score: 0 points for no lesions, 1 point for a periventricular cap or pen-like thin-layer lesions, 2 points for smooth halo-shaped lesions, and 3 points for periventricular signals extending to deep white matter; (2) Deep WMH score: 0 points for no lesions, 1 point for punctate lesions, 2 points for lesion fusion at the beginning, and 3 points for large area fusion lesions. The sum of the two scores is the final score. In this study, the total burden of CSVD was defined as CSVD_score, the lacunar infarct score was defined as Lac_score, the CMBs score was defined as CMBs_score, the WMH score was defined as WMH_score, and the EPVS score was defined as EPVS_score. The overall CSVD burden was 4 points, with higher scores associated with more severe CSVD lesions. However, because the score for brain atrophy has not been widely validated, most studies did not include brain atrophy as an indicator of the overall burden of CSVD.

Statistical analysis

The “tableone” package of R software was used to construct clinical data tables. Continuous variables are represented by the mean (standard deviation) or median (interquartile range), and categorical variables are represented by use cases (percentages). Violin plots and Box plots were drawn with the “ggpubr” package to compare the 4 CSVD burden groups. The logistic regression equation was fitted using the “rms” package. We identified independent risk factors for cognitive impairment in CSVD patients via a nomogram model. We verified the accuracy and utility of the nomogram using bootstrapping, the conformity index (C-index), the calibration curve, and decision curve analysis (DCA). Values of p < 0.05 were considered to indicate statistical significance. All the above statistical analyses were performed using the R software package (version 4.1.1).

Results

Comparison of baseline data

Finally, among the 227 patients (mean age 66.7 ± 6.99 years) with CSVD in the present study, 68.7% were male. One hundred (44.1%) patients experienced cognitive impairment, the others composed the normal cognitive function group, and the differences in demographic data (age), risk factors for cerebrovascular disease (smoke, hypertension, total cholesterol, homocysteine), imaging markers of CSVD (lacuneus, cerebral microbleeds, white matter hyperintensities, total CSVD burden) and MoCA score were statistically significant (p < 0.05) (Table 1).

Table 1 Characteristics of cerebral small vessel disease between cognitive normal and cognitive impairment (n)**
Full size table

According to the total burden of CSVD, the patients were divided into groups according to their CSVD_score1, CSVD_score2, CSVD_score3, or CSVD_score4. CSVD_score1 represents patients with a CSVD total burden score of 1, similarly, CSVD_score2, CSVD_score3, and CSVD_score4 represent patients with CSVD total burden scores of 2, 3, and 4, respectively. The “tableone” package of the R software is invoked to plot the baseline feature table. For the variables with statistically significant (P < 0.05) for the overall comparison, we found that multiple sets of non-parametric tests were needed for the variables of these four groups of data, so the “Bonferroni” method was selected to perform non-parametric tests and intergroup comparisons for these four sets of data. The demographic data and clinical characteristics of the patients in the two groups are shown in Table 2. There were statistically significant differences in age, HCY, MoCA scores, and imaging markers of CSVD (LI, CMBs, WMH, and EPVS) among these groups (p < 0.05). The “ggpubr” package of the R software was invoked, and the “Kruskal-Wallis” method based on “ANOVA” was used to compare the differences between the four groups, and the “Bonferroni” method was used to compare the two groups in pairs, and the results were finally displayed as box plots and violin plots. The results of the intergroup comparisons are presented in Fig. 2. As time progressed, the MoCA score decreased, and the total burden of CSVD increased. There was a significant difference in age between the CSVD_score1 and CSVD_score3 groups (p < 0.05) (Fig. 2a). With the increase in the total burden of CSVD, the HCY level exhibited an increasing trend, and there were significant differences between the CSVD_score1 and CSVD_score2 groups and between the CSVD_score1 and CSVD_score3 groups (p < 0.05) (Fig. 2b). With the increase in the total burden of CSVD, the MoCA score showed a downwards trend, and there were significant differences among the multiple groups (CSVD_score1 and CSVD_score2, CSVD_score1 and CSVD_score3, CSVD_score1 and CSVD_score4, and CSVD_score2 and CSVD_score3) (p < 0.05) (Fig. 2c). The more imaging markers of CSVD were used, the more serious the total burden of CSVD was, and there were statistically significant differences between the different CSVD burden groups. There were statistically significant differences in the biomarkers of CSVD, such as lacunes, microbleeds, white matter hyperintensities, and enlarged perivascular spaces (p < 0.05), between the different CSVD burden groups (Fig. 2d, e, f and g). We further assessed the effect of confounders on cognitive impairment using analysis of covariance. The results showed that HDL, LDL, UA, FBG, and Education were not correlated with the MoCA, and the CSVD burden had a significant effect on the MoCA scores. The results also showed that Age, Hcy, TG, CHOL, Lac_score, CMBs_score, EPVS_score, and WMH_score all affected MoCA scores. After controlling for these confounders, the effect of total CSVD burden on MoCA scores was more accurate (Supplementary Table and Supplementary Figure).

Table 2 Characteristics of cerebral small vessel disease between groups of total magnetic resonance burden (n)**
Full size table

Construction of predictive models

The MoCA score was used as the dependent variable, and the statistically significant variables shown in Table 1 (age, smoking status, hypertension, total cholesterol, homocysteine level, lacune, cerebral microbleeds, white matter hyperintensities, and CSVD score) were selected as the independent variables. The independent risk factors for cognitive impairment in patients with CSVD were identified by binary logistic regression. Age (OR = 1.070; 95% CI = 1.015 ~ 1.128, p < 0.05), history of hypertension (OR = 2.863; 95% CI = 1.438 ~ 5.699, p < 0.05), HCY (OR = 1.065; 95% CI = 1.005 ~ 1.127, p < 0.05), Lac_score (OR = 2.732; 95% CI = 1.094 ~ 6.825, P < 0.05), and total CSVD burden (CSVD_score) (OR = 3.823; 95% CI = 1.496 ~ 9.768, P < 0.05) were found to be independent risk factors for cognitive impairment in patients with CSVD, and a risk prediction model for cognitive impairment was established, which is displayed in Fig. 3 in the form of a nomogram.

Validation of predictive models

The nomogram was verified by using the “rms” package and bootstrap internal verification method, and the C-index was calculated as 0.839 (Fig. 4). Finally, the model is verified against test data from other periods at our institute, as shown in Fig. 5, with a C index of 0.867. Both the predictive model and the validation model showed good performance. As shown in Fig. 6, according to the DCA of the nomogram of the predictive model, the prediction nomogram has a better net benefit in the threshold probability range of DCA. Similarly, the DCA of the test nomogram also revealed a greater net benefit of the test model for predicting cognitive impairment in patients with CSVD, as shown in Fig. 7.

Fig. 2
figure 2

Comparison of clinical characteristics between CSVD burden groups. Note: Abbreviations: Hcy, homocysteine; Lac, lacune; CMBs, cerebral microbleeds; WMH, white matter hyperintensity; EPVS, enlarged perivascular space; CSVD, cerebral small vessel disease; MoCA, Montreal Cognitive Assessment

Full size image
Fig. 3
figure 3

The nomogram for assessing cognitive impairment. As shown in Fig. 3, a total score (0-100 points) for each independent variable was obtained by the use of a vertical line on the scoring scale at the top of the nomogram. The total score of all variables was added together to obtain the predicted risk value on the prediction line at the bottom of the nomogram. The assignment of categorical variables, without the disease or state, is represented by no and otherwise by yes. Of the two categorical variables, hypertension, and Lac_score, 0 is no, and 1 is yes. Note: Abbreviations: Hcy, homocysteine; Lac, lacune; CSVD, cerebral small vessel disease

Full size image
Fig. 4
figure 4

The calibration curves for the prediction nomogram. As shown in Fig. 4, the x-axis represents the nomogram-predicted probability, and the y-axis represents the actual probability of cognitive impairment. A perfect prediction would correspond to the 45°ideal line. The dotted line represents the entire cohort (n = 227), and the solid line represents the data bias-corrected by bootstrapping (B = 1000 repetitions), indicating the performance of the observed nomogram. Note: C-index = 0.839

Full size image
Fig. 5
figure 5

The calibration curves for the test nomogram. As shown in Fig. 5, the x-axis represents the nomogram-predicted probability, and the y-axis represents the actual probability of cognitive impairment. A perfect prediction would correspond to the 45°ideal line. The dotted line represents the entire cohort (n = 70), and the solid line represents the data bias-corrected by bootstrapping (B = 1000 repetitions), indicating the performance of the observed nomogram. Note: C-index = 0.867

Full size image
Fig. 6
figure 6

Assessment of clinical usefulness of the prediction nomogram with a decision analysis curve. A horizontal straight line denotes a net benefit of 0, indicating that all CSVD patients have normal cognitive function and no treatment is needed; the gray line represents that all subjects have developed cognitive deterioration and are being treated; the blue curve represents the intervention of CSVD patients according to the outcomes in the predictive model; and the prediction nomogram has a better net benefit than the other two extreme curves in the threshold probability range of DCA.

Full size image
Fig. 7
figure 7

Assessment of clinical usefulness with a decision analysis curve in the test nomogram. A horizontal straight line denotes a net benefit of 0, indicating that all CSVD patients have normal cognitive function and no treatment is needed; the gray line represents that all subjects have developed cognitive deterioration and are being treated; the blue curve represents the intervention of CSVD patients according to the outcomes in the test model; and the test nomogram has a better net benefit than the other two extreme curves in the threshold probability range of DCA.

Full size image

Discussion

Currently, the etiology of CSVD-related cognitive deterioration has not been fully elucidated, and the underlying risk factors are controversial. The ability of the prediction model to predict CSVD-related cognitive impairment has become a hot topic in clinical research. The purpose of this study was to establish a clinical prediction model of cognitive dysfunction in patients with CSVD and guide clinicians in the individualized prevention and treatment of these patients. Research has shown [25,26,27] that CSVD is an important cause of stroke, cognitive decline, and age-related disability. It is a comprehensive disease that includes brain atrophy, cerebral microbleeds, lacuna, and other pathological changes. Moreover, CSVD-related cognitive impairment is an important subtype of vascular cognitive impairment. According to this research, the probability of cognitive impairment in patients with CSVD has reached 41%, which is consistent with previous research results [28], suggesting that prevention and treatment of cognitive impairment in patients with CSVD have become the main goals of clinicians.

Age is the main risk factor for the development and progression of CSVD [28]. Our research also reached the same conclusion. In addition, this study confirmed that there was a statistically significant difference in age between different groups of participants according to the severity of the total burden of CSVD and that the total burden score of CSVD increased with age. This finding is highly consistent with previous research results [29]. We also showed that the prevalence of neuroimaging markers and cognitive decline in patients with CSVD increased with age and was found in most people older than 60 years. When we conducted a univariate analysis between the cognitive impairment group and the cognitively normal group, smoking was statistically different between the two groups (p < 0.05). However, we included it in logistic regression, and after adjusting for confounders, smoking was screened out, suggesting that smoking was not an independent risk factor for cognitive impairment in CSVD in this study. The same conclusions have been found in the previous literature [30, 31]. According to a previous study, hypertension has been proven to be an important independent risk factor for cognitive impairment in patients with CSVD [7, 13, 28, 32]. Fortunately, our findings were consistent with those of our predecessors. A cross-sectional study showed that age and hypertension were significantly and independently associated with the burden of CSVD [33]. Patients with CSVD and hypertension are prone to cognitive progress, possibly because hypertension can lead to successive oxidative stress and inflammatory responses, leading to changes in nerves and blood vessels; endothelial dysfunction; increased permeability of the blood-brain barrier; and accelerated pathological changes in CSVD, such as WMH, lacunar, EPVS, and CMBS [34]. Therefore, identifying and understanding the early subclinical stage of brain health deterioration and studying hypertension exposure may provide more in-depth insights into the role of hypertension in the pathophysiology of cognitive dysfunction.

In this study, we confirmed that homocysteine (Hcy) concentration was an independent risk factor for cognitive impairment in patients with CSVD. Moreover, the data showed that with increasing Hcy concentration, the total burden of CSVD and cognitive function tended to increase. As previously mentioned, Hcy produces oxygen-free radicals during metabolism and strongly oxidizes vascular endothelial cells [35]. An increase in Hcy levels has been proven to be related to cardiovascular disease, Alzheimer’s disease, and cerebrovascular disease [36], suggesting that endothelial dysfunction is the main mechanism involved. Endothelial dysfunction has also been studied as the main mechanism of cognitive impairment in patients with CSVD. The effect of the serum Hcy concentration on the development of CSVD was dose-dependent [37]. These findings provide clues for us to further study the pathophysiology of CSVD.

As previously mentioned, the total burden of CSVD was evaluated by neuroimaging to evaluate more complete and comprehensive brain injury in patients with CSVD, which represents the severity of brain injury and directly contributes to cognitive impairment [2, 38,39,40]. We confirmed that the greater the burden of CSVD was, the lower the MoCA score, and the greater the degree of brain damage; these results reaffirm the conclusions of the previous literature. Based on studies of neuroimaging features of CSVD, cognitive decline in patients with CVSD may be caused by pathological changes in the brain (LI, CMBS, WMH) that disrupt cortico-cortical, cortical-subcortical fibrous connections, especially when white matter hyperintensities predominate [32, 41]. During the one-year follow-up, the increase in enlarged perivascular spaces (EPVSs), especially in stroke patients, was closely related to other indicators of small vessel disease and cognitive impairment [42]. Enlarged perivascular spaces (EPVSs) are key to evaluating CSVD, but related research is insufficient. Although EPVS is associated with worse cognition, depression, and neurodegenerative diseases, its prognostic significance is unclear.

However, relying on pathological brain impairment does not directly predict the chance of cognitive decline. Therefore, it is necessary to construct a cognitive risk decline prediction system. Although a large number of studies have focused on the risk factors for and pathogenesis of cognitive impairment in CSVD patients [40, 43], they still cannot guide doctors to individualize and forecast the probability of deterioration of cognitive function in patients with CSVD. Currently, there are few clinical predictive models of cognitive deterioration risk in patients with CSVD. The strength of the present study compared to previous studies is that we established and verified a more readable nomogram that can easily display the risk of cognitive impairment in each individual with CSVD, confirming that the predictive model has good clinical application value and extrapolation. In this study, a nomogram for predicting risk factors for cognitive impairment in patients with CSVD was established, and the accuracy and usefulness of the model were verified by calibration curves and decision analysis curves. The results showed that the predictive model had good performance, as indicated by its calibration, discrimination, and clinical utility. Another highlight of the study is that compared with that of previous studies, the reliability of the model was further verified by using external data. The validation model also showed good performance. In conclusion, the results of double verification showed that the nomogram model has good stability and good generalizability. Therefore, we can conclude that except for a small subset of patients with low preferences, treatment strategies that intervene in patients with cognitive impairment based on predictive models have greater benefits than strategies that intervene in all patients and do not intervene in all patients. For patients with CSVD, the conclusion is that using this model to determine whether cognitive function interventions should be performed on patients will improve clinical outcomes. The results of the present study have dual clinical significance. First, the findings of this study could lead to more convenient predictions of the incidence of cognitive impairment in individuals with CSVD, and this topic is worth promoting for clinical application. The second important finding of this study is that additional attention and targeted efforts are needed to better understand the pathogenesis of vascular injury to the brain caused by small vessel disease and to thoroughly define the clinical consequences of these diseases.

Regarding these shortcomings, the study has certain limitations. First, the study included single-center data, and there was a certain bias in the selection of participants. Second, this was a retrospective cross-sectional study, and there was no further follow-up of the study cohort or further observation of the impact of time on the total burden of MRI and cognitive function. Third, the study was limited to assessing global cognitive function and did not evaluate specific cognitive domains; therefore, the impact of different brain pathological changes on the cognitive domain could not be assessed. In addition, the distribution of CSVD burden in this sample was skewed, with a low frequency of severe CSVD, which is also consistent with clinical reality. This may have led to bias in the strength of the associations that we observed for the total CSVD load with the MoCA scores. Therefore, we emphasize the need to replicate our findings in different cohort studies. Because of the above limitations, we will improve upon these findings in subsequent studies, and in the future, prospective cohort studies will be performed to further evaluate the damage caused by the total magnetic resonance burden to specific cognitive domains and further verify the reliability of the CSVD-related cognitive impairment model.

Conclusion

Overall, our research provides further evidence that there is a significant association between the total MRI burden of CSVD patients and cognitive impairment. The nomogram model based on brain MR images and clinical data helps in individualizing the probability of cognitive decline progression in patients with CSVD.