A Pilot Study of Brain Metabolites and Their Diagnostic Cutoff Point in Postmenopausal Women With Cognitive Impairment Using Magnetic Resonance Spectroscopy

ABSTRACT

Proton magnetic resonance spectroscopy (¹H‐MRS) was used to detect the levels of N‐acetylaspartate (NAA)/total creatinine (tCr) and choline (Cho)/tCr in bilateral hippocampus and posterior cingulate gyrus of postmenopausal patients with noncognitive impairment and cognitive impairment and to evaluate the cutoff point of related brain metabolites. To enhance the early recognition of postmenopausal cognitive impairment, this study included 62 postmenopausal patients with subjective cognitive decline (SCD), 62 postmenopausal patients with mild cognitive impairment (MCI), and 56 postmenopausal healthy controls (HC) without cognitive impairment. The relationship between brain metabolites and cognitive function was analyzed using Mini‐Mental State Examination (MMSE), Montreal Cognitive Assessment (MOCA), and ¹H‐MRS. The cutoff point for related brain metabolites were predicted by the receiver operating characteristic curve and the Youden index (YDI). The scores of MMSE and MOCA in the SCD group and the MCI group were lower than those in the HC group, and the scores of MMSE and MOCA in the MCI group were lower than those in the SCD group. The differences of NAA/tCr in the bilateral hippocampus, bilateral posterior cingulate gyrus, and Cho/tCr in the right hippocampus among the three groups were statistically significant. In comparison with the HC group, NAA/tCr in the bilateral hippocampus of the SCD group was decreased, and Cho/tCr in the right hippocampus of the MCI group decreased. Compared with the HC group and the SCD group, NAA/tCr in the bilateral hippocampus and bilateral posterior cingulate gyrus of the MCI group significantly decreased. Among them, the correlation coefficient between NAA/tCr in bilateral hippocampus regions and MMSE and MOCA scale scores were the highest. When distinguishing the SCD group from the HC group, the area under the curve (AUC) of NAA/tCr in the left hippocampus was the highest at 0.687 (95%CI: 0.591–0.784, p < 0.001), with a standard error (SE) of 0.049, acceptable sensitivity (71.0%) and specificity (60.7%), a maximum YDI of 0.317, and a corresponding cutoff point of 1.445; when distinguishing the MCI group from the HC group, the AUC of left hippocampal NAA/tCr was the highest at 0.876 (95%CI: 0.815–0.938, p < 0.001), with an SE of 0.031, high sensitivity (77.4%) and specificity (82.1%), a maximum YDI of 0.595, and a corresponding cutoff point of 1.265. Additionally, for distinguishing between the SCD group and the MCI group, the AUC of left hippocampal NAA/tCr was 0.752 (95%CI: 0.665–0.839, p < 0.001), with moderate sensitivity (59.7%) but high specificity (82.3%), a YDI of 0.420, and a cutoff point of 1.155. Postmenopausal SCD patients, postmenopausal MCI patients, and postmenopausal healthy subjects exhibit significant differences in brain metabolite levels. According to the cutoff point for related brain metabolites, early screening for cognitive impairment in postmenopausal middle‐aged and elderly women can serve as a reference for clinical diagnosis.

This study employed ¹H‐MRS to compare brain metabolites among postmenopausal SCD, MCI, and healthy subjects, analyzed metabolite‐cognition score correlations, and determined metabolite cutoffs using ROC curves and YDI. Significant intergroup metabolite differences were found. Left hippocampal NAA/tCr distinguishes these groups well, and the metabolite cutoffs support early clinical screening for cognitive impairment in postmenopausal middle‐aged and elderly women.

Abbreviations

¹H‐MRS

proton magnetic resonance spectroscopy

AD

Alzheimer's disease

AUC

area under the ROC curve

Cho

choline

Cho/tCr

choline/total creatinine

HC

healthy controls

MCI

mild cognitive impairment

NAA

N‐acetylaspartate

NAA/tCr

N‐acetylaspartate/total creatinine

tCr

total creatine

NAA/tCr

N‐acetylaspartate/total creatinine

MMSE

Mini‐Mental State Examination

MOCA

Montreal Cognitive Assessment

ROC

receiver operating characteristic (curve)

SCD

subjective cognitive decline

SE

standard error

YDI

Youden index

1. Introduction

Alzheimer's disease (AD) is the most common type of dementia and becomes irreversible once it begins. With the acceleration of global aging, the socioeconomic burden of AD is expected to keep rising [1]. It is projected that the number of dementia patients worldwide will increase from 5.74 million in 2019 to 152 million by 2050 [2]. Recent studies [3] show that postmenopausal middle‐aged and older women account for about two‐thirds of all AD cases, with a lifetime risk of 1 in 6 for women and 1 in 11 for men to develop AD. A study of Chinese Americans [4] found that the prevalence (77.06%) and incidence (75.20%) of cognitive impairment in women aged 60 and older were higher than those in men, which were 22.94% and 24.80%, respectively. Women face a higher risk of cognitive impairment, including subjective cognitive decline (SCD) and mild cognitive impairment (MCI), which are early stages of AD [5, 6]. Previous retrospective studies have shown [7] that the risk of SCD progressing to MCI is 2.15 times greater than in those without SCD, while the risk of SCD progressing to dementia is 2.17 times higher than in the normal aging population. A prospective study following 211 cognitively normal individuals and 174 SCD patients for 5 years revealed [8] that 12.9% of SCD patients progressed to MCI and 4.6% progressed directly to dementia. It is evident that for postmenopausal middle‐aged and elderly women, a high‐risk group for cognitive impairment, conducting early screening and diagnosis during the intervention “window period” of SCD and MCI, is a critical task for clinically blocking or delaying the progression of AD and also an important clinical issue that urgently requires breakthroughs [9]. Not only will this help alleviate the burden on the public health system and reduce the incidence risk of moderate‐to‐advanced dementia but it will also decrease the consumption of relevant medical resources. Consequently, it confers substantial public health value and social significance in advancing the construction of a “healthy aging” society [10, 11].

Currently, we primarily use neuropsychological scales to assess SCD and MCI, which are prone to missing and misdiagnosing. Other technical methods, such as cerebrospinal fluid biomarkers and positron emission tomography, have also been used [12]. However, cerebrospinal fluid biomarker testing is invasive, and positron emission tomography scans are expensive, making them difficult for most people to accept [13]. Some studies have examined changes in blood biomarkers in patients with cognitive impairment and their relationship with cognitive function, but the results have not been very reliable or specific [14]. Conversely, proton magnetic resonance spectroscopy (¹H‐MRS) is a noninvasive semiquantitative technique for detecting brain metabolites such as N‐acetylaspartate (NAA), choline (Cho), and total creatine (tCr) and is currently the most widely used and effective imaging method in cognitive neuroscience [15]. Metabolic changes in the brains of patients with cognitive impairment occur before any morphological changes. ¹H‐MRS can detect changes in brain tissue based on metabolite levels, which result from neuronal and axonal damage, reflecting early neurodegenerative changes [16]. This technique helps early identification and assessment of cognitive impairment status and severity. The hippocampus and posterior cingulate cortex in the default mode network are brain regions that are most closely associated with memory, perception, and understanding abilities. These regions are commonly impaired in MCI and AD and are among the first areas to show abnormal Aβ deposition [17]. Early detection of metabolite levels in these regions is highly valuable for clinical indications. Furthermore, numerous studies have indicated [18, 19] that due to significant changes in physiological characteristics, postmenopausal middle‐aged and elderly women have a high incidence of cognitive impairment, and there may be potential differences in the critical thresholds of their related brain metabolites compared with other populations. As a gender‐specific risk factor for cognitive impairment in women, the dramatic decline in sex hormone (especially estrogen) levels in postmenopausal middle‐aged and elderly women is one of the key predisposing factors for their development of ad. This physiological change can directly regulate the homeostasis of cognition‐related metabolites in the brain, impair the integrity of neural function, and thereby exacerbate cognitive impairment [20]. However, there remains a lack of characteristic studies on brain metabolites specifically targeting postmenopausal middle‐aged and elderly women, and early screening lacks specific imaging indicators.

This study aims to explore the relationship between brain metabolites and cognitive function in Chinese postmenopausal women of middle to old age and to establish a cutoff point for relevant brain metabolites. This will provide crucial information for the early clinical screening of postmenopausal cognitive impairment. The first large‐scale observation of differential changes in brain metabolites among three groups of postmenopausal subjects (those with SCD, those with MCI and healthy subjects) will supplement characteristic data on brain metabolites in postmenopausal middle‐aged and elderly women, facilitating subsequent in‐depth exploration of brain metabolic regulatory mechanisms and interventional studies. The results of this research can guide clinicians in quickly identifying high‐risk groups and provide a basis for individualized interventions, such as acupuncture and cognitive training. This will effectively reduce the burden on family carers and the consumption of social medical resources.

2. Materials and Methods

2.1. General Information and Sample Size

Based on sample size estimation from previous imaging studies and literature on MCI‐related mechanism research [21], 20 cases per group can basically meet the requirements for statistical analysis. We selected 62 postmenopausal subjects with SCD and 62 postmenopausal subjects with MCI from the outpatient department of the First Affiliated Hospital of Guangxi University of Traditional Chinese Medicine and the Ren'ai Branch from January 2021 to January 2025. We also randomly enrolled 56 postmenopausal healthy subjects without cognitive impairment as a healthy control (HC) group.

This study adheres to the principles of the Declaration of Helsinki. All participants were informed about the study procedures and signed informed consent forms. The study protocol was registered with the Chinese Clinical Trial Center (http://www.chictr.org.cn), with registration number ChiCTR2200060683. The Medical Ethics Committee of the First Affiliated Hospital of Guangxi University of Traditional Chinese Medicine granted ethical approval (approval document number 2022‐001‐02). We used a computer‐generated central randomization scheme that was managed by a designated person. Except for the designer and the central randomization officer, other researchers were unaware of the randomization scheme. Clinicians requested random numbers and group assignment numbers from the central randomization staff based on the participant's enrollment number. They then carried out clinical assessments and treatments in accordance with established protocols for each group. The study used both participant blinding and third‐party evaluation blinding.

2.2. Diagnostic Criteria

The diagnostic criteria for MCI, based on the “Interpretation of the Guidelines for the Diagnosis and Treatment of Dementia in China ‐ Guidelines for the Diagnosis and Treatment of Mild Cognitive Impairment” [22], include the following criteria: (1) The main complaint is memory impairment, confirmed by an informant familiar with the patient's cognitive status; (2) other cognitive functions are relatively intact or only mildly impaired; (3) mot meeting the diagnostic criteria for dementia; (4) activities of daily living are unaffected; (5) exclusion of other systemic diseases that may cause cognitive decline; (6) Global Deterioration Scale score of 2–3, Clinical Dementia Rating score of 0.5, memory test score at least 1.5 standard deviations below the mean of age‐ and education‐matched controls, and a Mini‐Mental State Examination (MMSE) score of at least 24. The diagnostic criteria for SCD, based on the “Diagnosis and Treatment Strategies for Subjective Cognitive Decline in Alzheimer's Disease Clinical Preclinical Stage” [23], are as follows: (1) a persistent self‐perceived decline in cognitive function compared with a previously normal state, unrelated to acute events; (2) standardized cognitive tests adjusted for age, gender, and years of education show normal results and do not meet the diagnostic criteria for MCI. The diagnostic criteria for menopause, based on the “Guidelines for Menopause Management and Hormone Therapy in Menopause 2023 Edition” [24], include women over 40 years old who have been amenorrheic for 12 months, excluding pregnancy and other diseases that may cause amenorrhea.

2.3. Inclusion Criteria

The postmenopausal SCD or MCI inclusion criteria are as follows: (1) Aged 50–75 years, female, right‐handed, naturally postmenopausal with an age at menopause of 45–55 years; (2) participants must meet the diagnostic criteria for postmenopausal SCD or MCI, and their main complaint must be memory decline; (3) native Chinese speakers; (4) activities of daily living are intact or only slightly impaired, with a Functional Activities Questionnaire score ≤ 9; (5) no visual, auditory, or intellectual impairments; no language barriers; and able to answer and complete questionnaires smoothly; (6) no hormone treatment or medication affecting cognition has been used in the past month; (7) no significant cerebral lesions were detected by MRI; (8) cooperate voluntarily and sign the informed consent form. Inclusion criteria for the HC group [25] are (1) 50–75 years of age, female, right‐handed, naturally postmenopausal with an age at menopause of 45–55 years; (2) no complaints of memory loss, with overall cognitive function within normal limits; (3) no cerebral lesions detected by MRI; (4) cooperate voluntarily and sign the informed consent form.

2.4. Exclusion Criteria

(1) Participants diagnosed with AD are not included; (2) participants with other neurological diseases that may lead to cognitive decline such as Parkinson's disease and vascular dementia; (3) severe heart, liver, or kidney diseases that may cause cognitive decline, such as thyroid dysfunction, anemia, and hepatic encephalopathy; (4) brain infections, cerebral infarction, cerebral hemorrhage, severe traumatic brain injury, or other central nervous system diseases that may affect neuropsychological assessment; (5) mental illnesses such as depression, anxiety, bipolar disorder, and schizophrenia; (6) history of substance abuse or addiction within the past year.

2.5. Cognitive Function Assessment

The MMSE encompasses five dimensions: orientation, memory, attention, calculation, language ability, and visuospatial construction. The test consists of a total of 11 items with a maximum score of 30, where lower scores indicate more severe cognitive impairment. The Montreal Cognitive Assessment (MOCA) encompasses nine dimensions: attention and concentration, executive function, immediate memory, delayed recall, language, visuospatial and construction skills, abstract thinking, calculation, and orientation. The maximum score is 30 points, with higher scores indicating better cognitive function. For participants with less than 12 years of education, 1 point is added to the total score as a correction factor [26].

2.6. ¹H‐MRS Examination

High‐resolution T1‐weighted imaging and multi‐voxel scanning were performed using a 3.0T Siemens MAGNETOM Verio superconducting MRI system (Siemens Healthineers, Germany) equipped with an eight‐channel SENSE head coil [27, 28]. All participants were fitted with a custom‐made foam headrest during scanning, with elastic cushions placed bilaterally to immobilize the head and minimize lateral/vertical displacement. Prior to scanning, participants received breathing relaxation guidance and were informed of the importance of maintaining immobility. Senior technicians conducted real‐time monitoring throughout the scanning process; if significant motion (e.g., head shaking) was detected, scanning was paused and resumed only after repositioning. High‐resolution T1‐weighted images were used for the localization of the hippocampus and posterior cingulate gyrus, with the following parameters: Repetition Time (TR)/Echo Time (TE) = 1900 ms/2.22 ms, Field of View (FOV) = 250 mm × 250 mm, slice thickness = 1 mm, resolution = 256 × 256 × 168, flip angle = 9°, and number of slices = 176. A point‐resolved spectroscopy‐chemical shift imaging (PRESS‐CSI) sequence was applied for ¹H‐MRS scanning, with parameters set as follows: TR/TE = 1700 ms/135 ms, frequency = 1200 Hz, FOV = 160 mm × 160 mm, matrix = 16 × 16, flip angle = 90°, and voxel size = 10.0 mm × 10.0 mm × 15.0 mm. For the hippocampus, the volume of interest (VOI) was 80 mm × 66 mm × 15 mm; for the posterior cingulate gyrus, the VOI was 50 mm × 50 mm × 15 mm. For partial volume effect correction, voxel localization strictly referenced anatomical landmarks (e.g., the posterior cingulate gyrus was demarcated by the superior margin of the splenium of the corpus callosum and the cingulate sulcus) to ensure that voxels covered ≥ 80% of the target brain region, thereby reducing interference from non‐target tissues. The VOI and full width at half maximum (FWHM) settings for the hippocampus were configured to match the anatomical characteristics of bilateral hippocampi at the skull base, as reported in reference [29], to minimize interferences from air, skull, cerebrospinal fluid, and other confounding factors. An automatic prescan sequence was applied to adjust voxel gain. Three‐dimensional automatic shimming was used for motion artifact suppression and optimization of local magnetic field homogeneity, with the water suppression level maintained above 95%. Given the larger shimming coverage of multi‐voxel MRS, the initial quality control threshold for FWHM was set to less than 25 Hz based on previous literature [30] and our clinical experience to ensure successful scanning. FWHM was measured manually with the single NAA peak as the reference. In the left hippocampus, the FWHM values were 12.99 ± 2.97 (6.80–18.80) Hz for the HC group, 12.46 ± 3.03 (6.90–20.80) Hz for the SCD group, and 12.70 ± 2.71 (7.40–20.70) Hz for the MCI group. In the right hippocampus, the corresponding FWHM values were 13.18 ± 3.00 (7.10–18.40) Hz, 12.11 ± 2.85 (6.80–21.20) Hz, and 12.32 ± 2.82 (7.60–20.90) Hz across the three groups (Tables S1 and S2). These values complied with the reference criteria for FWHM in multi‐voxel MRS of the hippocampus [31], yielding stable spectral baselines and favorable shimming performance [32]. All spectra were evaluated by experienced technologists. Rescanning and parameter readjustment were performed immediately if motion artifacts, peak shift, baseline distortion, or poor shimming were observed. Finally, post‐processing and quantitative analysis of ¹H‐MRS data were performed using syngo.via, a clinical‐grade medical imaging diagnosis and post‐processing workstation software developed by Siemens. The procedures included water reference processing, filtering, zero‐filling, Fourier transformation, frequency correction, baseline correction, phase correction, and curve fitting. The metabolic chemical shifts of NAA, tCr, and Cho were 2.02, 3.03, and 3.22 ppm, respectively. Taking tCr as the internal reference standard, the ratios of NAA/tCr and Cho/tCr were calculated. Strict spectral inclusion criteria were established to ensure data reliability: the Cramer–Rao lower bounds (CRLB) of all metabolites (NAA, Cho, Cr) ≤ 20%; the signal‐to‐noise ratio (SNR), referenced to tCr, ≥ 4 (calculated as the peak‐to‐peak amplitude of tCr divided by the standard deviation of baseline noise); baseline drift ≤ 5%; and no obvious sharp artifact peaks or peak splitting. Figure 1 presents individual cases of ¹H‐MRS from the left hippocampus of participants in the three groups, which serve as representative fitted spectra. This figure includes original spectra, fitted curves, and labeled resonance peaks of major metabolites (NAA, tCr, and Cho), clearly demonstrating the fitting effect, metabolite resolution, and shimming performance.

FIGURE 1.

Representative cases of ¹H‐MRS of the left hippocampus in the three groups. (A) shows a 63‐year‐old postmenopausal healthy woman, with the corresponding metabolite ratios in her left hippocampus being (NAA/tCr = 1.64, Cho/tCr = 1.14); (B) shows a 60‐year‐old postmenopausal patient with SCD, with the corresponding metabolite ratios in his left hippocampus being (NAA/tCr = 1.29, Cho/tCr = 0.85); (C) shows a 69‐year‐old postmenopausal patient with MCI, with the corresponding metabolite ratios in her left hippocampus being (NAA/tCr = 1.03, Cho/tCr = 0.97). Cho, choline; Cho/tCr, choline/total creatine; HC, healthy controls; MCI, mild cognitive impairment; NAA, N‐acetylaspartate; NAA/tCr, N‐acetylaspartate/total creatine; SCD, subjective cognitive decline; tCr, total creatine.

2.7. Statistical Methods

Statistical analysis was performed using SPSS 27.0. For normally distributed data, the data are presented as mean ± standard deviation, and a one‐way ANOVA was performed, followed by post hoc comparisons using the LSD test. For non‐normally distributed data, median and interquartile ranges were used, and the nonparametric Kruskal–Wallis test was performed, followed by post hoc comparisons using Dunn's test. Correlation analysis was performed using Spearman's rank correlation coefficient [33], and correlation heat maps were plotted using OriginPro 2024. Receiver operating characteristic (ROC) curves were constructed to evaluate the predictive performance of various indicators for disease occurrence, with the Youden index (YDI) = sensitivity + specificity − 1, and the maximum YDI value used to determine the optimal cutoff point for related brain metabolites [34]. All statistical tests were conducted with a two‐sided, and a p‐value of < 0.05 was considered statistically significant.

3. Results

3.1. General Information Analysis

There were no statistically significant differences in age and years of education between the three groups (p > 0.05; Table 1). This indicates that the groups are comparable.

TABLE 1.

Descriptive data for the general characteristics (n = 180).

Variable	HC group (n = 56)	SCD group (n = 62)	MCI group (n = 62)	H value	p
Age (year)	62 (57–65)	61.5 (58.5–67.25)	62 (58.75–67.25)	0.520	0.771
Education level (year)	11 (10–14)	12 (11–14)	11 (10–14)	5.065	0.079

Abbreviations: HC, healthy controls; MCI, mild cognitive impairment; SCD, subjective cognitive decline.

3.2. Cognitive Function Scores and ¹H‐MRS Results

Comparison of MMSE and MOCA scores among the three groups showed statistically significant differences (p < 0.05; see Table 2). Specifically, both the SCD and MCI groups had lower MMSE and MOCA scores than the HC group (p < 0.05), and the MCI group's scores were lower than those of the SCD group (p < 0.05). Table 2 also shows that the ¹H‐MRS results were significantly different among the three groups in terms of NAA/tCr ratios in the bilateral hippocampus and posterior cingulate gyrus. Additionally, the ratio of Cho/tCr in the right hippocampus differed significantly (p < 0.05). Compared with the HC group, the SCD group had lower bilateral hippocampal NAA/tCr (p < 0.05), while the MCI group had lower right hippocampal Cho/tCr (p < 0.05). Furthermore, compared with both the HC and SCD groups, the MCI group showed a significant decrease in NAA/tCr in the bilateral hippocampus and bilateral posterior cingulate gyrus (p < 0.05). No significant differences were found in the other comparisons (p > 0.05).

TABLE 2.

Cognitive scale scores and MRS test results (n = 180).

Variable	HC group (n = 56)	SCD group (n = 62)	MCI group (n = 62)	H value/F value	p
MMSE	30 (29–30)	29 (28–30)*	26 (25–27)**##	134.895	< 0.001
MoCA	28 (26.25–29)	27 (26–27)*	22 (21–24)**##	128.232	< 0.001
HIP.L NAA/tCr	1.50 (1.29–1.69)	1.33 (1.18–1.49)**	1.13 (1.03–1.26)**##	56.598	< 0.001
HIP.L Cho/tCr	1.02 (0.87–1.14)	1.01 (0.85–1.09)	0.93 (0.84–1.08)	2.579	0.275
HIP.R NAA/tCr	1.46 ± 0.27	1.33 ± 0.19**	1.14 ± 0.18**##	32.757	< 0.001
HIP.R Cho/tCr	0.98 (0.88–1.12)	0.95 (0.81–1.10)	0.88 (0.79–0.99)**	11.288	0.004
PCG.L NAA/tCr	2.06 (1.97–2.27)	2.01 (1.93–2.14)	1.89 (1.74–2.01)**##	30.602	0.006
PCG.L Cho/tCr	1.02 (0.91–1.16)	0.99 (0.79–1.13)	0.92 (0.84–1.05)	4.434	0.109
PCG.R NAA/tCr	2.05 ± 0.23	1.98 ± 0.18	1.85 ± 0.24**##	13.236	< 0.001
PCG.R Cho/tCr	1.04 ± 0.24	0.94 ± 0.23	0.98 ± 0.22	2.642	0.074

Note: Compared with the HC group, * means p < 0.05, and ** means p < 0.01; compared with the SCD group, ^## means p < 0.01.

Abbreviations: Cho/tCr, choline/total creatinine; HC, healthy controls; HIP, hippocampus; L, left; MCI, mild cognitive impairment; MMSE, Mini‐Mental State Examination; MOCA, Montreal Cognitive Assessment; MRS, magnetic resonance spectroscopy; NAA/tCr, N‐acetylaspartate/total creatinine; PCG, posterior cingulate gyrus; R, right; SCD, subjective cognitive decline.

3.3. Correlation Between Cognitive Function Scores and ¹H‐MRS

As shown in Figure 2, MMSE and MOCA scores are positively correlated with NAA/tCr in both hippocampi, NAA/tCr in the bilateral posterior cingulate, and Cho/tCr in the right hippocampus (p < 0.05), with correlation coefficients ranging from 0.19 to 0.53. Specifically, the correlation coefficients for MMSE and MOCA scores with NAA/tCr in both hippocampi are over 0.40, indicating a moderate association [35]. Further analysis using scatter plots revealed a positive correlation between MMSE and MOCA scores and bilateral hippocampal NAA/tCr levels, as shown in Figure 3.

FIGURE 2.

Heatmap of the correlation between cognitive function scores and ¹H‐MRS. The color of the ellipse changes from blue to green to red, representing the variation in the magnitude of the significant p value. * indicates p < 0.05, and ** indicates p < 0.01. The numbers represent the values of the corresponding correlation coefficients, with the color changing from blue to green to red to represent the variation in the magnitude of the values. Correlation coefficients without significant meaning are not displayed. Cho/tCr, choline/total creatinine; HIP, hippocampus; L, left; MMSE, Mini‐Mental State Examination; MOCA, Montreal Cognitive Assessment; NAA/tCr, N‐acetylaspartate/total creatinine; PCG, posterior cingulate gyrus; R, right.

FIGURE 3.

Analysis of the correlation between cognitive function scores and metabolite levels measured by ¹H‐MRS. (A) For the orange scatter points (MOCA), the fitted linear equation is y = 0.043x + 0.221, R ² = 0.2258; for the blue scatter points (MMSE), the fitted equation is y = 0.074x − 0.767, R ² = 0.2357. For both, p < 0.001, and the NAA/tCr in the left hippocampus is positively correlated with the scores of the MMSE and MOCA scales. (B)For the red scatter points (MOCA), the fitted equation is y = 0.042x + 0.216, R ² = 0.2072; for the gray scatter points (MMSE), the fitted equation is y = 0.058x − 0.535, R ² = 0.2407. For both, p < 0.001, and the NAA/tCr in the right hippocampus is positively correlated with the scores of the MMSE and MOCA scales. Cho/tCr, choline/total creatinine; HIP, hippocampus; L, left; MMSE, Mini‐Mental State Examination; MOCA, Montreal Cognitive Assessment; NAA/tCr, N‐acetylaspartate/total creatinine; PCG, posterior cingulate gyrus; R, right.

3.4. ROC Curve and Cutoff Point Analysis

The area under the ROC curve (AUC) is an important measure of diagnostic or predictive model accuracy, with values ranging from 0 to 1. The closer the value is to 1, the better the model can distinguish between classes. The standard error (SE) shows the degree of dispersion of the AUC estimate, with smaller SE values indicating more precise AUC estimates [36]. Further analysis of correlation coefficients greater than 0.40 for the bilateral hippocampal NAA/tCr, as shown in Table 3 and Figure 4, indicates that the curve of NAA/tCr in the left hippocampus has stronger discriminatory ability and higher specificity overall than that of NAA/tCr in the right hippocampus. Compared with the HC group, when distinguishing the SCD group, the highest AUC for left hippocampal NAA/tCr is 0.687 (95% CI: 0.591–0.784, p < 0.001), with an SE of 0.049, suggesting that the discriminative ability measured by AUC is significantly better than chance (random guessing), with a sensitivity of 71.0% and specificity of 60.7%. The maximum YDI is 0.317, which corresponds to a cutoff point of 1.445. This means that if left hippocampal NAA/tCr is less than 1.445, it may indicate postmenopausal SCD. Compared with the HC group, when distinguishing the MCI group, the highest AUC for left hippocampal NAA/tCr is 0.876 (95% CI: 0.815–0.938, p < 0.001), with an SE of 0.031, demonstrating that the model's discriminative ability as evaluated by the AUC is quite good, with a sensitivity of 77.4% and specificity of 82.1%. The maximum YDI is 0.595, which corresponds to a cutoff point of 1.265; if left hippocampal NAA/tCr is less than 1.265, it may indicate postmenopausal MCI. Additionally, for distinguishing between the SCD group and the MCI group, the area under the curve (AUC) of left hippocampal NAA/tCr was 0.752 (95%CI: 0.665–0.839, p < 0.001), with moderate sensitivity (59.7%) but high specificity (82.3%), a YDI of 0.420, and a cutoff point of 1.155. Taken together, based on the above indicators, left hippocampal NAA/tCr may serve as a potential effective indicator for distinguishing between the groups.

TABLE 3.

Analysis of the area under the ROC curve and its corresponding cutoff point among the three groups.

Group	Variable	AUC	Std. error	p	95% CI	Sensitivity	Specificity	YDI	Cutoff value
HC group versus SCD group	HIP.L NAA/tCr	0.687	0.049	< 0.001	0.591–0.784	71.0%	60.7%	0.317	1.445
	HIP.R NAA/tCr	0.629	0.052	0.016	0.526–0.731	82.3%	46.4%	0.287	1.480
HC group versus MCI group	HIP.L NAA/tCr	0.876	0.031	< 0.001	0.815–0.938	77.4%	82.1%	0.595	1.265
	HIP.R NAA/tCr	0.834	0.037	< 0.001	0.761–0.907	96.8%	55.4%	0.522	1.395
SCD group versus MCI group	HIP.L NAA/tCr	0.752	0.044	< 0.001	0.665–0.839	59.7%	82.3%	0.420	1.155
	HIP.R NAA/tCr	0.777	0.041	< 0.001	0.695–0.858	82.3%	61.3%	0.436	1.295

Abbreviations: AUC, area under the ROC curve; Cho/tCr, choline/total creatinine; HC, healthy controls; MCI, mild cognitive impairment; NAA/tCr, N‐acetylaspartate/total creatinine; SCD, subjective cognitive decline; YDI, Youden index.

FIGURE 4.

Area under the ROC curve and 95% CI. (A) HIP. L NAA/tCr: 0.687 (95%CI: 0.591–0.784), p < 0.001; HIP. R NAA//tCr: 0.629 (95%CI: 0.526–0.731, p = 0.016. (B) HIP. L NAA/tCr: 0.876 (95%CI: 0.815–0.938), p < 0.001; HIP. R NAA//tCr: 0.834 (95%CI: 0.761–0.907), p = 0.016. (C) HIP. L NAA/tCr: 0.752 (95%CI: 0.665–0.839), p < 0.001; HIP. R NAA//tCr: 0.777 (95%CI: 0.695–0.858), < 0.001. Cho/tCr, choline/total creatinine; HC, healthy controls; HIP, hippocampus; L, left; MCI, mild cognitive impairment; NAA/tCr, N‐acetylaspartate/total creatinine; PCG, posterior cingulate gyrus; R, right; SCD, subjective cognitive decline.

4. Discussion

In this study, we evaluated the cognitive function of postmenopausal patients with SCD and MCI and postmenopausal healthy subjects. We utilized ¹H‐MRS to assess differences in brain metabolites among the three groups, examined the correlations between brain metabolites and cognitive function scores, and ultimately established the cutoff point of brain metabolites using ROC curves and YDI. This work affords a theoretical basis for clarifying the pathophysiological mechanisms underlying postmenopausal cognitive impairment and facilitates early clinical prevention and intervention, potentially preventing or delaying progression to dementia, such as AD.

MMSE and MOCA scales [37] are two neuropsychological screening tools that have undergone rigorous validation for reliability and validity. Their core value lies in standardized task design that enables multidimensional assessment of cognitive function, allowing precise identification, subtype classification, and severity stratification of cognitive impairment. The combined use of both scales is included in international Multicenter Consensus Guidelines [38] and the “Guidelines for the Diagnosis and Treatment of Alzheimer's Disease Dementia in China” [12]. This combined application has become the global “gold standard” for early detection of cognitive impairment. In this study, the MMSE and MOCA scores of the SCD and MCI groups were both significantly lower than those of the healthy control group (p < 0.05). Furthermore, the MCI group had lower MMSE and MOCA scores than the SCD group (p < 0.05), indicating that postmenopausal middle‐aged women exhibit varying degrees of cognitive impairment.

¹H‐MRS measures changes in brain metabolite concentrations by detecting differences in resonance frequencies, which reflects neuronal integrity and glial activity, enabling early detection of cognitive function abnormalities [39]. Among them, NAA is a marker of neuronal density and mitochondrial function, with reductions often indicating neuronal injury or degeneration, while tCr serves as a reference for energy metabolism homeostasis, with absolute changes reflecting energy metabolism disorders [40]. In diseases associated with cognitive impairment, the hippocampus and posterior cingulate cortex are the first and most susceptible regions to pathological changes [41, 42]. Early studies [43] found that the NAA/tCr ratio in the posterior cingulate cortex was higher in healthy subjects compared with AD and MCI patients, suggesting that MCI patients have neuronal injury or mitochondrial dysfunction, which is more severe in AD patients. A recent study [44] found that the NAA/tCr ratios in the hippocampus and posterior cingulate cortex of MCI patients were lower than those of healthy subjects. Similarly, our previous studies [27, 28] found that the NAA/tCr ratio in the bilateral hippocampus and bilateral posterior cingulate cortex of MCI patients was significantly lower than that of healthy controls. However, due to significant changes in the physiological characteristics of postmenopausal women of middle‐aged and elderly populations, these groups have a high incidence of cognitive impairment. There may be potential differences in the critical thresholds of their related brain metabolites compared with other populations. Currently, there are no reports on the differences in brain metabolites at different stages of early cognitive impairment in postmenopausal middle‐aged women and the extent of cognitive impairment remains unknown. Further in‐depth analysis of this population in our study revealed that the differences in NAA/tCr ratios in the bilateral hippocampus and bilateral posterior cingulate cortex among the three groups are statistically significant. Compared with the HC group, the SCD group showed reduced NAA/tCr in the bilateral hippocampus; when compared with the HC and SCD groups, the MCI group showed significantly lower NAA/tCr in both the bilateral hippocampus and bilateral posterior cingulate cortex. Furthermore, there is a positive correlation between MMSE and MOCA scores and the NAA/tCr ratios in the bilateral hippocampus and bilateral posterior cingulate cortex (p < 0.05), indicating that lower NAA/tCr is associated with more severe cognitive impairment. The levels of NAA/tCr in the hippocampus and posterior cingulate cortex show a progressive pattern of damage from localized brain regions to multiple brain regions in early cognitive impairment among postmenopausal middle‐aged women, indicating their potential as noninvasive imaging indicators for assessing the progression of early cognitive impairment in this population. Additionally, a decrease in Cho/tCr in brain tissue suggests a reduction in the content of choline compounds, which is associated with neurodegenerative diseases, cell necrosis, or metabolic synthesis disorders. A study [45] found that in AD patients, the Cho/tCr in the hippocampus was lower than that of MCI patients and healthy subjects; the Cho/tCr in the hippocampus of MCI patients was lower than that of healthy subjects, showing a gradually decreasing trend. However, some studies have reported inconsistent levels of Cho/tCr; Wang et al. [43] analyzed 32 MCI patients and 56 healthy subjects and found that the Cho/tCr levels in MCI patients were similar to those in healthy subjects, with no statistically significant difference. Similarly, Sun Yong'an et al. and his team analyzed 40 MCI patients and 40 healthy subjects and found no significant difference in Cho/tCr between the two groups (p > 0.05). However, our previous study had similar results [27, 28]; compared with the HC group, only the MCI group showed a decrease in right hippocampal Cho/tCr (p < 0.05), and there was a positive correlation between MMSE and MOCA scores and right hippocampal Cho/tCr, possibly due to more severe cognitive impairment in the MCI group. Differences in study results may be affected by sample size, individual differences in the study subjects' conditions, and detection methods, which can reduce the accuracy of the results. This study strictly controlled confounding variables (e.g., age, years of education) among the three groups in accordance with the inclusion criteria, thereby minimizing confounding effects on the results induced by the introduction of multivariate factors (e.g., duration of menopause, medication history, comorbidities) to a large extent. Future studies should expand the sample size to clarify the differences in Cho/tCr levels.

Previous studies based on ROC curves [43] suggested that a NAA/tCr ratio of ≤ 1.50 in the posterior cingulate cortex is the optimal cutoff point for distinguishing amnestic MCI patients from healthy subjects but did not subdivide the differences in results between the bilateral posterior cingulate cortices. In this study, correlation analysis was performed between cognitive function scores and brain metabolites in the bilateral hippocampi and bilateral posterior cingulate cortices. Further analysis was conducted on the bilateral hippocampal NAA/tCr with a correlation coefficient > 0.40 (indicating a closer correlation). The results showed that compared with the HC group, the left hippocampal NAA/tCr of the SCD group showed the highest AUC of 0.687 (95% CI: 0.591–0.784, p < 0.001), with a SE of 0.049, a sensitivity of 71%, and a specificity of 60.7%. The maximum YDI was 0.317, which is equivalent to a cutoff point of 1.445, demonstrating good discrimination between HC and SCD. To distinguish the MCI group from the HC group, the left hippocampal NAA/tCr showed the highest AUC of 0.876 (95% CI: 0.815–0.938, p < 0.001), with a SE of 0.031, indicating strong classification ability, with a sensitivity of 77.4%, specificity of 82.1%, and a maximum YDI of 0.595, corresponding to a cutoff point of 1.265. These findings suggest differences in the distribution and concentration trends of brain metabolites at various stages of early cognitive impairment in postmenopausal middle‐aged women. Additionally, for distinguishing between the SCD group and the MCI group, the AUC of left hippocampal NAA/tCr was 0.752 (95%CI: 0.665–0.839, p < 0.001), with moderate sensitivity (59.7%) but high specificity (82.3%), a YDI of 0.420, and a cutoff point of 1.155. These findings indicate that left hippocampal NAA/tCr has distinct advantages as a potential discriminative indicator for distinguishing among the groups.

4.1. Study Limitations and Future Research Directions

(1) This study is a single‐center preliminary study. Due to the unstable sex hormone levels and numerous confounding factors in premenopausal women, the sample size is relatively limited. Therefore, no premenopausal women control group was established, and only homogeneous postmenopausal populations were selected to avoid multiple confoundings. Moreover, selection bias introduced by differences in regional demographic characteristics and medical accessibility will further limit the external validity of the study results. In the future, large‐sample, multicenter prospective cohort studies need to be conducted, including premenopausal women for long‐term follow‐up, to dynamically explore the association between sex hormone fluctuations during menopause, cognitive impairment, and brain metabolites, thereby enhancing the statistical robustness of the results and their generalizability to postmenopausal populations in different regions. (2) Declining estradiol (E₂) levels in postmenopausal women represent a well‐recognized pathological basis for cognitive impairment: (1) E₂ maintains neuronal integrity and mitochondrial function via classical nuclear receptor (ERα/ERβ), MAPK/ERK, and PI3K/Akt signaling pathways [46], thereby positively regulating NAA synthesis. Postmenopausal E₂ deficiency induces neuronal damage and reduces NAA levels [47]. (2) E₂ participates in the homeostasis of cell membrane phospholipid metabolism [48], and insufficient E₂ after menopause causes abnormal elevation of choline (Cho) transport and metabolism [49]. (3) E₂ regulates cerebral energy metabolism and maintains creatine (Cr) homeostasis, while E₂ decline disrupts energy balance [49]. (4) E₂ modulates myoinositol (mI) levels through antioxidant effects and inhibition of glial activation [47], and postmenopausal E₂ deficiency exacerbates glial metabolic abnormalities [46]. As this is a preliminary exploratory study focusing on brain metabolites in postmenopausal women with cognitive impairment, and all participants were naturally postmenopausal with stable sex hormone levels, we did not perform invasive blood tests for hormone measurement. In future studies, we will quantitatively measure sex hormones to further verify the underlying mechanisms linking hormone levels to brain metabolites. (3) A long TE time was adopted in ¹H‐MRS. Although this improved the spectral clarity of NAA, Cho, and tCr, the short T2 relaxation time may lead to signal loss of other metabolites [50], limiting the detection of other neurochemical substances with biological significance, and the study results may have slight deviations. Future studies need to further expand the content of brain metabolite detection to reveal the neurochemical network related to postmenopausal cognitive impairment as comprehensively as possible. (4) This study lacks sample size calculation and did not perform a priori power analysis. The sample size was mainly estimated with reference to previous imaging studies and literature on MCI‐related mechanism research, and the number of enrolled subjects was reported based on the actual number recruited in clinical practice. Although the enrolled number can meet the requirements of statistical analysis, a priori power analysis should be added in future studies to enhance persuasiveness. (5) This study adopted the clinically prevalent syngo.via post‐processing software, which cannot automatically export or permanently store spectral quality metrics including FWHM. Accordingly, all FWHM values were manually measured retrospectively from raw spectra, which may introduce operator‐dependent bias. Although all measurements were performed by trained personnel following standard protocols, the lack of automated quality control remains a methodological limitation of the present study. Future investigations may utilize research‐dedicated MRS post‐processing platforms that enable real‐time and objective recording of spectral linewidth parameters for further validation.

5. Conclusion

The present study supplements the characteristic data of brain metabolites in postmenopausal middle‐aged and elderly women by conducting a preliminary analysis of brain metabolites in three study populations: postmenopausal patients with SCD, MCI, and postmenopausal healthy subjects. In clinical practice, screening postmenopausal women of middle age and older who exhibit high prevalence and incidence of cognitive impairment based on these cutoff points of brain metabolites holds significant implications for reducing the rates of missed diagnosis and misdiagnosis. Future research will address and mitigate the aforementioned limitations while closely monitoring changes in statistically significant indicators to improve clinical diagnosis accuracy. Furthermore, we will integrate advanced neuroimaging technologies such as functional magnetic resonance imaging (fMRI) and diffusion tensor imaging (DTI) to synthesize brain metabolic, functional, and structural data, thereby comprehensively investigating the neural mechanisms underlying cognitive decline in postmenopausal women.

Author Contributions

L.Z. and W.M. designed the study. X.C., L.T., Y.L., Y.Q., T.L., and J.M. conducted the experiments. Z.Z. and Y.H. analyzed and interpreted the data. X.C. and L.Z. wrote the manuscript. All authors read and approved the final version and made substantial contributions to its content.

Funding

This research was funded by the National Natural Science Foundation of China (No. 82160933), the Natural Science Foundation of Guangxi Zhuang Autonomous Region (Nos. 2023GXNSFAA026094, 2022GXNSFAA035577), the Joint Project on Regional High‐Incidence Diseases Research of Guangxi Natural Science Foundation under Grant (No. 2024GXNSFBA010165), and the Project of the Universal Support Policy for the First Batch of Young Talent “Qingmiao” Programme in the Department of Human Resources and Social Security of Guangxi Zhuang Autonomous Region (No. 2024‐2).

Conflicts of Interest

The authors declare no conflicts of interest.

Supporting information

TABLE S1: Comparison of FWHM of bilateral hippocampus among three groups (n = 180) (Hz).

TABLE S2: Specific FWHM values of bilateral hippocampus in three groups (n = 180).

Data Availability Statement

The data that support the findings of this study are available from the corresponding author upon reasonable request.