A Nomogram Including Total Cerebral Small Vessel Disease Burden Score for Predicting Mild Vascular Cognitive Impairment in Patients with Type 2 Diabetes Mellitus

Introduction

Vascular cognitive impairment (VCI) has emerged as a leading public health challenge due to an increase in population aging and vascular risk factors.¹ VCI is the second most common cause of dementia after Alzheimer’s disease (AD) and accounts for at least 20–40% of dementia.^1,2 VCI refers to cognitive impairment due to vascular risk factors or diseases, including mild VCI and vascular dementia (VaD).^1,3,4 Mild VCI is generally recognized as an early and reversible stage of VaD.^1,3,5

Accumulating evidence indicates that T2DM is associated with an increased risk of VCI.^6–9 Individuals with T2DM exhibit a heightened risk of developing dementia, ranging from 1.5 to 2.8 times more likely than those without T2DM.^10,11 In the T2DM population, approximately 20% of individuals older than 60 years may have dementia.⁸ What’s more, T2DM increases the risk of incident mild cognitive impairment (MCI) and accelerates the conversion from MCI to dementia.^8,12,13 In addition, VCI, especially VaD, may lead to poor self-management of T2DM, which in turn leads to more cerebrovascular events and poorer cognitive function.⁸ Therefore, the early diagnosis of mild VCI in patients with T2DM plays a crucial role in halting or delaying the progression of VCI.

Although the underlying neuropathological mechanisms of T2DM-related cognitive impairment is complicated, multifactorial and incompletely understood, evidence from preclinical and clinical studies support that cerebral microvascular dysfunction may play an important role.^14–16 Cerebral microvascular dysfunction can be reflected noninvasively by assessing the magnetic resonance imaging (MRI) features of cerebral small vessel disease (CSVD).^15–17 Visible imaging markers of CSVD mainly include white matter hyperintensity, lacune, enlarged perivascular space and cerebral microbleed.^18,19 Previous studies mostly focus on the association between a specific imaging marker of CSVD and cognitive impairment in the T2DM population.^20–23 However, CSVD is a dynamic whole-brain disorder and different imaging markers of CSVD can co-exist or be transformed into each other.^24,25 The total CSVD burden score, including above four main imaging markers, has good construct validity and may better represent the severity of CSVD or cerebral microvascular dysfunction.^19,26

Therefore, the total CSVD burden score should be an important predictor of VCI in T2DM patients. However, few predictive nomogram models of VCI in patients with T2DM have included the total CSVD burden score, which reflects underlying cerebral microvascular dysfunction, especially in the early stage of VCI. Therefore, this study aimed to develop an accurate and individualized predictive nomogram, including the total CSVD burden score for mild VCI in patients with T2DM, by comprehensively evaluating the contributions of demographic data, laboratory data, and imaging markers of CSVD.

Materials and Methods

Study Design and Participants

This retrospective observational prediction model development and validation study included participants from the Department of Neurology of Hebei General Hospital between October 2016 and January 2022. Participants were included in this study if they met the following inclusion criteria: (a) aged ≥ 50 years; (b) diagnosed with T2DM based on the criteria of the World Health Organization;²⁷ (c) completed whole MRI sequences to assess the total CSVD burden score; and (d) completed assessment of cognitive function. Participants were excluded if they had dementia or conditions that may affect the results of the cognitive function assessment, such as acute cerebrovascular events, epilepsy, anxiety, depression, brain injuries, schizophrenia, hypothyroidism, malignancy, or metabolic encephalopathy. This study followed the principles of the Declaration of Helsinki and was approved by the Ethics Committee of Hebei General Hospital (No.2023176).

Assessment of Mild Vascular Cognitive Impairment

All participants underwent neuropsychological testing, including the standardized translated version of the Montreal Cognitive Assessment (MoCA) Beijing version (www.mocatest.org), basic and instrumental activities of daily living (ADL), Hamilton Anxiety Scale (HAMA) with 14 items, and Hamilton Depression Scale (HAMD) with 24 items. Objective evidence of cognitive impairment was based on the results of neuropsychological testing as follows: MoCA (≤13 for illiterate, ≤19 for 1–6 years of education, ≤24 for 7 or more years of education), HAMA <6, and HAMD <8.²⁸ The diagnosis of mild VCI was based on the previously established criteria,^3,29 including (a) objective evidence of cognitive impairment, (b) clinical features consistent with a vascular etiology, (c) evidence of cerebrovascular disease that was considered sufficient to account for cognitive impairment, and (d) no dementia (basic ADL was normal and instrumental ADL<10).

Predictors Collection

The following candidate predictors for further analysis were collected: demographic data, including age, gender, years of education, height, and weight; clinical data, including systolic blood pressure, diastolic blood pressure, current smoking or drinking, duration of T2DM, anti-diabetic drugs, history of stroke, hypertension, and coronary heart disease; and laboratory data, including fasting plasma glucose, hemoglobin A1c (HbA1c), lipid profile, uric acid, fibrinogen, serum homocysteine (Hcy), and imaging markers of CSVD. Body mass index (BMI) was calculated as the weight (kg) divided by the square of height (m²). The total CSVD burden score (range, 0–4) was calculated using four different imaging markers of CSVD, which were described in detail in our previous researches.^30,31

Statistical Analysis

Statistical analyses were performed using the R statistical software (version 4.2.3; R Foundation for Statistical Computing, Vienna, Austria). The variables were described as mean (standard deviation), median (interquartile range), and case (percentage), as appropriate. The Mann–Whitney U-test, t-test, or chi-square test was used to analyze the differences in characteristics between the normal cognitive group and the mild VCI group. P <0.05 was considered to be statistically significant. To develop the nomogram of mild VCI in patients with T2DM, multivariable logistic regression analysis with Akaike information criterion was applied to select the significant predictors using a backward selection method that included variables with a P <0.1 in the univariable analysis. Additionally, a dynamic nomogram was built using the “DynNom” package, which is conveniently available to clinicians. Validation of the predictive model was carried out based on the following aspects: 1) discriminative ability was assessed by calculating the area under the receiver operating characteristic curve (AUC); 2) calibrated ability was performed using the calibration plot and Hosmer-Lemeshow test; 3) overall performance was evaluated using the Brier score, which captures aspects of both calibration and discrimination; and 4) clinical usefulness was evaluated by decision curve analysis (DCA). To reduce overfitting bias, the bootstrapping method (resampling=2000) was used for internal validation.

Results

Participants Characteristics

A total of 322 eligible participants with T2DM were included in this study: 189 with mild VCI and 133 with normal cognitive function. The prevalence of mild VCI in T2DM patients was 58.7%. The characteristics of the study population between the two groups are presented in Table 1. Compared to the normal cognitive group, patients in the mild VCI group were older and less educated (P <0.05). The mild VCI group had higher HbA1c, serum Hcy levels, and total CSVD burden scores than the normal cognitive group (P <0.05).

Table 1 Characteristics of the Study Participants with T2DM Between Normal Cognitive Group and Mild VCI Group

Development of the Predictive Model

The results of the univariate and multivariate logistic regression analyses for risk factors associated with mild VCI in patients with T2DM are shown in Table 2. According to the results of univariate analysis, age, education, stroke, HbA1c, Hcy, and total CSVD burden score were potential predictors of mild VCI in patients with T2DM (P <0.1). In multivariable logistic regression analysis, age (odds ratio [OR]: 1.039, 95% confidence interval [CI]: 1.004–1.075, P<0.001), education (OR: 0.930, 95% CI: 0.872–0.991, P=0.026), HbA1c (OR: 1.310, 95% CI: 1.065–1.612, P=0.011), Hcy (OR: 1.053, 95% CI: 1.000–1.110, P=0.052), and total CSVD burden score (0 score: reference; 1 score: OR: 2.628, 95% CI: 1.083–6.373, P=0.033; 2 score: OR: 3.903, 95% CI: 1.642–9.279, P=0.002; 3 score: OR: 11.371, 95% CI: 4.514–28.641, P<0.001; 4 score: OR: 17.653, 95% CI: 6.067–51.360, P<0.001) were detected by Akaike information criterion as predictors of mild VCI in patients with T2DM.

Table 2 Binary Logistic Regression Analysis of the Associated Factors for the Risk of Mild VCI in Participants with T2DM

Based on these five significant predictors, a nomogram for mild VCI in T2DM patients was developed (Figure 1). Additionally, to facilitate clinical applications, a dynamic nomogram was developed and is available online (https://tengzhenjie.shinyapps.io/DynNomapp3/). For example, in a T2DM patient with an age of 68 years, education of 9 years, HbA1c level of 7%, Hcy level of 15 µmol/L, and CSVD burden score of 2, the probability of mild VCI was estimated to be 56.0% (Figure 2).

Figure 1 Nomogram for the prediction of the probability of mild VCI risk in participants with T2DM. The nomogram was developed by incorporating the following five parameters: age, education, HbA1c, Hcy, and total CSVD burden scores.

Figure 2 Dynamic nomogram for predicting the probability of mild VCI risk in participants with T2DM. An online dynamic nomogram accessible at https://tengzhenjie.shinyapps.io/DynNomapp3/, depicting an example for predicting the probability of having a mild VCI risk in participants with T2DM for a 68 years old patient with 9 years, HbA1c of 7%, Hcy of 15 µmol/L, and the CSVD burden score of 2.

Validation of the Predictive Model

The results of the 2000 bootstrap samples estimated the AUC to be 0.801 (95% CI: 0.752–0.849), which indicated good discrimination of the nomogram (Figure 3).

Figure 3 Receiver operating characteristic curve of the nomogram for predicting mild VCI in participants with T2DM using bootstrap resampling (2000 times).

The bootstrap-corrected (bootstrapping with 2000 repetitions) calibration plot for the probability of mild VCI in patients with T2DM demonstrated an optimal agreement between the prediction by the nomogram and the actual observation (Figure 4). In addition, the Hosmer-Lemeshow test suggested that the predictive model was goodness-of-fit (χ² =9.226, P=0.417), and the Brier score of the predictive model was 0.178. In summary, these results indicate that the model has a good predictive accuracy.

Figure 4 Calibration plot of the nomogram for predicting mild VCI in participants with T2DM. The dotted line represents the performance of the nomogram, whereas the solid line indicates nomogram bias. The dashed line represents the reference line in which the ideal nomogram lies. The prediction accuracy of the nomogram was better when the solid line was closer to the dotted one.

As shown in Figure 5, DCA indicated that when the threshold probability ranged between 16% and 98%, the use of the nomogram to predict mild VCI in patients with T2DM provided a greater net benefit, which suggested the clinical usefulness of the nomogram.

Figure 5 DCA curves of the nomogram for predicting mild VCI in participants with T2DM. The y-axis represents the net benefit. The gray line represents the assumption that all the patients had mild VCI. The black line represents the assumption that none of the patients had cognitive impairments. The dotted line represents the risk nomogram.

Discussion

In the present study, we developed and internally validated a nomogram based on age, education, HbA1c level, Hcy level, and total CSVD burden score to predict the probability of mild VCI in T2DM patients. A dynamic nomogram was developed to facilitate clinical application. The predictive nomogram showed good discriminatory ability, calibration, and clinical usefulness, which could facilitate the detection of mild VCI by clinicians in T2DM patients.

MCI is common in T2DM patients. A meta-analysis showed that the pooled estimated prevalence of MCI in T2DM patients was 45.0%.³² However, the prevalence of mild VCI in patients with T2DM is unknown. Our study utilized the MoCA to assess cognitive function and found that the prevalence of mild VCI was higher (58.7%) than that of MCI in T2DM patients.^32,33 One reason for this change might be that the MoCA was more sensitive to the diagnosis of MCI than the mini-mental state examination (MMSE).³⁴

Age is one of the important risk predictors for VCI,¹ and prevalence of T2DM or T2DM-related cognitive impairment increases with age.³⁵ The increased risk of mild VCI with older age has also been found in our study. In addition, this study observed that patients with low education levels were at a higher risk of mild VCI, which is consistent with previous studies.^3,35 This can be explained by the fact that a high education level was associated with a higher cognitive reserve.³⁶

The association between HbA1c level and T2DM-related cognitive impairment remains controversial. A prospective population-based study found the linear correlation between HbA1c levels and global cognitive decline over a longterm 10 year follow-up period.³⁷ In contrast, another prospective study found a J-shaped rather than linear association between HbA1c levels and incident dementia.¹² However, a randomised open-label study indicated that using intensive therapy to reduce the HbA1c levels did not improve cognitive function.³⁸ We found that the level of HbA1c was an important predictor and was related to the increased risk of mild VCI in patients with T2DM. A possible explanation is that a higher HbA1c level is associated with more aggravated T2DM-related microvascular complications.³⁷ Tian et al³⁹ reported that increased serum Hcy levels were associated with T2DM-related MCI, especially executive dysfunction. However, another study found no association between serum Hcy levels and T2DM-related MCI after further adjustment for potential confounders,⁴⁰ which is consistent with the findings of our previous study.⁴¹ To the best of our knowledge, few studies have explored the relationship between serum Hcy levels and mild VCI in T2DM patients. In this study, we found that the level of serum Hcy was an important predictor of mild VCI in patients with T2DM, which suggests that increased serum Hcy levels were more strongly correlated with MCI due to vascular risk factors.

One of the key underlying mechanisms of T2DM-related VCI may be cerebral microvascular dysfunction,^14,15 which is common in T2DM patients.⁴² Detrimental changes in cerebral microcirculation in T2DM include thickening of the basement membrane, increased angiogenesis, increased permeability of the blood-brain barrier, and altered regulation of microvascular blood flow.¹⁵ These changes can further lead to a range of reactions, such as hypoperfusion, hypoxia, oxidative stress, and inflammation,¹⁵ resulting in brain dysfunction and T2DM-related VCI. It is commonly recognized that features of CSVD, as measured by MRI, might reflect cerebral microvascular dysfunction in the T2DM population.^15,24 Previous studies have suggested that the severity of white matter hyperintensity or lacunes is independently associated with cognitive impairment in T2DM patients.^21,23 In addition, the enlarged perivascular space or cerebral microbleed may be a potential biomarker of cognitive impairment in patients with T2DM.^20,22

However, these studies did not further distinguish between the degree (MCI or dementia) and subtype (VCI or AD) of cognitive impairment. Moreover, as a whole-brain disorder,²⁴ different imaging markers of CSVD are interrelated, and it may be one-sided to consider only one specific imaging marker of CSVD. Therefore, the total CSVD burden score, which may better represent the severity of CSVD or cerebral microvascular dysfunction,^19,26 should be considered when developing a predictive model for mild VCI in patients with T2DM, which is rarely performed. In our study, we found that the total CSVD burden score was an important predictor of mild VCI in patients with T2DM and that severe CSVD burden (total CSVD burden score ≥3 points) had a more significant predictive ability. These findings indicate that the total CSVD burden score may be a potential biomarker for mild VCI in T2DM patients. Furthermore, our results may provide evidence for the important role of cerebral microvascular dysfunction in T2DM-related VCI and provide clues for future studies to investigate early diagnostic imaging markers of VCI in T2DM patients.

Taken together, we constructed and internally validated a nomogram to predict the probability of mild VCI in T2DM patients. The predictive nomogram exhibited a good discriminatory ability, calibration, and clinical usefulness. Compared with previous studies, this study has several advantages. First, to the best of our knowledge, this is the first predictive nomogram model for mild VCI in T2DM patients. Second, total CSVD burden score, an important predictor of T2DM-related cognitive impairment, was taken into account in our study, whereas others do not.^33,43,44 Third, the online dynamic nomogram that we constructed is more convenient and intuitive for clinical applications.

Limitations

Our study had some limitations. First, this was a single-center retrospective study with a relatively small sample size. Second, the nomogram developed in this study was internally validated using bootstrapping without external datasets. Third, the patients were recruited from a single center in our study, which may have some restrictions when generalizing to other populations. Future multicenter prospective studies with larger sample sizes and external datasets are warranted to address these issues.

Conclusion

In conclusion, the nomogram based on age, education, stroke, HbA1c, Hcy, and total CSVD burden scores may predict the risk of mild VCI in patients with T2DM. The predictive nomogram showed good discriminatory ability, calibration, and clinical usefulness, which could facilitate the detection of mild VCI by clinicians in T2DM patients. Future prospective studies with larger populations and external validation are warranted to enhance the stability of this model.