Comparison of the clinical value of MRI and plasma markers for cognitive impairment in patients aged ≥75 years: a retrospective study

Patient selection

We retrospectively selected all patients who visited the Affiliated Hospital of Xuzhou Medical University and were clinically diagnosed with or without cognitive dysfunction between May 2018 and November 2021.

The inclusion criteria were (1) clinical diagnosis of CI or noncognitive impairment (NCI), (2) aged 75 years and over, (3) available plasma Aβ42 and P-tau181 test results, (4) an MRI examination was performed, and (5) the interval time between the blood test and MRI was <1 week. The exclusion criteria were (1) brain disease, such as massive infarction, hemorrhage, malacosis or tumor, (2) brain injury, (3) toxic or metabolic diseases, or (4) medication history that affects cognitive function or can lead to WMH.

Data collection

The clinical data of the patients were extracted from the patient charts, including age, sex, cognitive status, medical history, hypertension, heart disease and diabetes, systolic blood pressure (SBP), diastolic blood pressure (DBP), the levels of triglycerides (TG), total cholesterol (Tcho), high-density lipoprotein cholesterol (HDL-C), low-density lipoprotein cholesterol (LDL-C), Aβ42 and P-tau181 in plasma. The overall cognitive status was assessed by neurologists with >5 years of working experience using the Mini-Mental State Examination (MMSE) and Montreal Cognitive Assessment (MoCA). CI was indicated by an MMSE score <27 or a MoCA score <26. Otherwise, a patients was considered to be in the NCI group (Nasreddine et al., 2005; Folstein, Folstein & McHugh, 1975).

MRI data were acquired using one of the five scanners used at the hospital during the study period. All scanning protocols included a T1-weighted gradient-echo sequence, fluid-attenuated inversion recovery (FLAIR) sequence, and T2-weighted sequence. The detailed scanning parameters are shown in Table S1.

WMH was defined as the areas of intensity signal on the T2-weighted FLAIR sequence (Wardlaw et al., 2013) divided into periventricular WMH (PVWMH) and deep WMH (DWMH) (Fazekas et al., 2002). PVWMH regions were defined as the regions within 1 cm from the ventricular border in each axial slice. DWMH was defined as the regions more than 1 cm from the edge of the lateral ventricle (Fig. 1A). Semiquantitative PVWMH and DWMH evaluations were performed by one senior radiologist. The evaluation was performed according to the Fazekas scale (score 0-3) (Fazekas et al., 2002) (Fig. S1). The patients were stratified into two different subgroups. The normal/mild subgroup included Fazekas grade 0-1, and the moderate/severe subgroup included Fazekas grade 2-3 (Peng et al., 2020).

Quantitative WMH: The white matter lesions on T2 FLAIR images were segmented using a free and automated method (the lesion prediction algorithm (LPA)), which provided by the Lesion Segmentation Toolbox (LST) (version 2.0.15, www.statistical-modeling.de/lst.html) in SPM (SPM12, MATLAB v.2017b; MathWorks, Natick, MA). Based on this algorithms, a lesion probability score for each voxel was calculated (Weeda et al., 2019). A total of 0.01 cm³ was selected to be minimum extent threshold. The outputs of LPA were the lesion probability maps, which were binarized by using a threshold of 0.5, then lesion masks were generated. However, there were some error in the lesion masks generated by the algorithm, it was necessary to visually confirm and correct the false-negative or false-positive. The probability graph marked with errors was evaluated by two radiologists and imported into ITK-SANP software (Version 3.8.0; http://www.itksnap.org/pmwiki/pmwiki.php) for manual adjustment. WMH volume in cm³ was defined as the voxel size multiplied by the total number of voxels labeled as lesions in the cerebrum (Fig. 1B).

The traditional linear measurement method has been widely used for brain atrophy evaluation in the clinic and research (Jeong et al., 2016; Pontillo et al., 2020; Wu, 1985. The measurement was performed by two neuroradiologists. Cerebral atrophy includes central and cortical atrophy. The lateral ventricle body index (LVBI) was used to evaluate central atrophy. LVBI = b/a, a = minimum distance of lateral wall of bilateral lateral ventricle, and b = brain transverse diameter at the same level. Axial views of the brain T1WI were reviewed to locate and measure a and b (Fig. 2A), and the average value was taken. Central atrophy was assigned scores of 0-3 based on LVBI (Table S2). Cortical atrophy was considered present whenever two or more sulci each had a width >3 mm. Two or more sulci with the widest relative width of each lobe were selected, and the width was measured perpendicular to each sulci on the T1WI. Cortical atrophy was also scored on a scale ranging from 0-3 based on the maximum sulci width. Two observers measured the sulci widths of the frontal, parietal, temporal, and occipital lobes and assigned scores directly. In cases of disagreement, a final result will be determined after another measurement (Fig. 2B, Table S2). Observers scored and measured the neuroimaging markers in a blinded manner.

Statistical analysis

Participantcs were described into the NCI and CI groups. The test kit instructions and defined reference range of Aβ42 and p-tau-181 (Anqun Biological Engineering Co., Shenzhen, China) were used. A cutoff of 110 pg/ml was used to classify the patients as amyloid-positive (Aβ+) or amyloid-negative (Aβ-). A cutoff of 30 pg/ml was used to classify the patients as p-tau181 positive (tau+) or negative (tau-). The NCI and CI groups were further divided into Aβ-, tau- and Aβ-/tau- subgroups.

Baseline clinical characteristics were compared between CI and NCI groups. First, the Shapiro-Wilk test and visual inspection of histograms were applied to check the normality of the continuity variables. If it was a normally distributed continuous variable, presented as mean ± SD and analyzed by the Student’s t test; otherwise, they were presented as median (range) and tested using the Mann‒Whitney U test. For categorical variables, they were described as n (%) and assessed using the chi-square test. To compare differences in plasma and neuroimaging markers between CI and NCI groups, we performed multivariable logistic regression, in which the variables with P-values <0.10 in univariable analysis were included and variables with collinearity were excluded. Diagnostic sensitivity, specificity and accuracy of biomarkers was assessed using the receiver operator characteristic (ROC) analysis and area under the ROC curve (AUROC). For the comparison of the neuroimaging markers between different subgroups, we used the same statistical methods. Two-sided P values < 0.05 were considered statistically significant. All statistical analyses were performed with SPSS 25.0 (IBM Corp., Armonk, NY, USA) and MedCalc 18.2.1 (MedCalc Software bvba, Ostend, Belgium).

Demographic characteristics of all enrolled patients

We prospectively enrolled 184 patients in this study, including 54 (29.4%) in the CI group and 130 (70.7%) in the NCI group (Fig. 3). The mean age (SD) of all patients was 82.7 (±3.4) years. There were no significant differences between the two groups for age, systolic blood pressure, diastolic blood pressure, blood lipid levels and the percentage of sex, history of hypertension, diabetes, and cardiac diseases (Table 1).

Comparison of neuroimaging markers between the NCI and CI groups

The interclass correlation coefficient (ICC) results of linear indicators showed good agreement between the two radiologists, ranging from 0.996 to 1.000 (Table S3).

Univariate logistic regression analysis showed the frequency of moderate and severe PVWMH was significantly higher in the CI group than in the NCI group (P < 0.001). Multivariate analysis demonstrated that moderate and severe PVWMH were related to CI (OR 2.857, [1.365–5.983], P = 0.005). The LVBI was 3.77 and 4.46 in the CI and NCI groups, respectively (P < 0.001). Multivariate analysis showed that LVBI with an OR of 0.413 (95% CI [0.243–0.700], P < 0.001) and cortical atrophy with an OR of 1.304 (95% CI [1.079–1.575], P = 0.006) were two related factors for CI.

The WMH volume, DWMH, TWMH, total atrophy and TWMH and total atrophy showed no significant difference in multivariate analysis. The frequencies of Aβ42+_, P-tau 181+, and Aβ42+/P-tau181+ were not significantly different between the CI and NCI groups (all P > 0.05) (Table 2). The AUROC was 0.782 by using the combined model (PVWMH, LVBI and CA) to distinguish between CI and NCI, with 68.5% sensitivity and 78.5% specificity (Tables 2 and S4, and Fig. 4A).

Comparison of neuroimaging markers between NCI and CI in the A $\mathrm{\text{β}}$- and tau- subgroups

A total of 58.8% (107/182) and 71.82% (130/181) of patients were Aβ- and tau-, respectively. In the Aβ- subgroup (n = 107), the LVBI was 3.86 and 4.45 in the CI and NCI groups, respectively (P = 0.002). Multivariate analysis showed that LVBI was related to CI (OR = 0.484, 95% CI [0.258–0.907], P = 0.024). The proportion of moderate and severe PVWMH had a certain effect on CI, with an OR of 2.632 (95% CI [1.033–6.706], P = 0.043) after multivariate analysis, although the difference was no significant in univariate analysis (P = 0.089). The AUROC was 0.734 by combining LVBI and PVWMH to differentiate CI and NCI in the Aβ- subgroup, the sensitivity and specificity were 71.0% and77.6%, respectively (Tables 3 and S4, and Fig. 4B).

In the tau- subgroup (n = 130), univariate analysis showed that the percentage of moderate and severe PVWMH was significant different in CI and NCI groups (70.3% vs 46.2%, respectively, P = 0.013). Moderate and severe PVWMH was a related factor for CI in multivariate analysis (OR 2.816, [1.180–6.717], P = 0.020). For patients in the CI and NCI groups, the LVBI was 3.90 and 4.49, respectively (P = 0.001). After multivariate analysis, the LVBI was demonstrated to be a relevant factor for CI with an OR of 0.496 (95% CI [0.279–0.881], P = 0.017). To distinguish between CI and NCI in the tau- subgroup by the model combining PVWMH and LVBI, the AUROC was 0.757, the sensitivity and specificity were 83.8% and 65.6%, respectively (Tables 4 and S4, and Fig. 4C).

Comparison of neuroimaging markers between NCI and CI in the A $\text{β}$-/tau- subgroup

Ninety-four (51.09%) patients were both Aβ negative and tau negative. The LVBI was 3.89 ± 0.79 and 4.46 ± 0.90 in CI and NCI, respectively (P = 0.005). Multivariate analysis showed that the LVBI was significantly different, with an OR of 0.442 (95% CI [0.239–0.819], P = 0.009). Using LVBI to discriminate CI from NCI in the Aβ-/tau- subgroup, the AUROC was 0.720, the sensitivity and specificity were 69.0% and 76.9%, respectively (Tables 5 and S4, and Fig. 4D).