Grey matter changes on brain MRI in subjective cognitive decline: a systematic review

People with subjective cognitive decline (SCD) report cognitive deterioration. However, their performance in neuropsychological evaluation falls within the normal range. The present study aims to analyse whether structural magnetic resonance imaging (MRI) reveals grey matter changes in the SCD population compared with healthy normal controls (HC). Parallel systematic searches in PubMed and Web of Science databases were conducted, following the Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) guidelines. Quality assessment was completed using the Newcastle-Ottawa Scale (NOS). Fifty-one MRI studies were included. Thirty-five studies used a region of interest (ROI) analysis, 15 used a voxel-based morphometry (VBM) analysis and 10 studies used a cortical thickness (CTh) analysis. Ten studies combined both, VBM or CTh analysis with ROI analysis. Medial temporal structures, like the hippocampus or the entorhinal cortex (EC), seemed to present grey matter reduction in SCD compared with HC, but the samples and results are heterogeneous. Larger sample sizes could help to better determine if these grey matter changes are consistent in SCD subjects.


Introduction
Alzheimer's disease (AD) is the most prevalent neurodegenerative disease and the leading cause of dementia, accounting for an estimated 50-70% of cases [1]. AD is an age-related condition and its global worldwide prevalence is expected to be much greater with increasing in the ageing population, reaching 106.8 million people in 2050 [2]. The global annual economic cost of dementia supposes an amount of one billion US dollars and it will increase up to 2 billion in 2030 [3]. It is estimated that a 1-year delay on disease onset would reduce the number of cases in 12 million by 2050, being an early and precise diagnostic, an essential tool for it [2].
Nowadays, we know that the natural history of AD is divided into three phases: the preclinical phase, where the pathogenic mechanisms of the disease have started but no objective cognitive decline can be diagnosed; the prodromal phase, where mild objective cognitive symptoms can be identified, but they are not severe enough to meet dementia criteria; and the dementia phase, where cognitive decline interferes with daily activities [4]. Some subjects in the preclinical phase of AD declare mild cognitive symptoms with no clinical evidence of cognitive impairment as compared with Open Access † Pablo Arrondo and Óscar Elía-Zudaire contributed equally to this work. *Correspondence: glomaran@gmail.com age-, sex-and education-matched subjects. This clinical construct has historically received many names such as subjective cognitive impairment, subjective memory impairment or decline or memory complaints, although it is currently referred to as subjective cognitive decline (SCD) [5,6].
SCD prevalence is noticeably high (25-50%) in the population over 65 years old, albeit not all causes are AD-related. In fact, the aetiology of SCD is heterogeneous and can also be related to normal ageing and psychiatric or non-degenerative neurological disorders such as depression, cerebrovascular diseases or concussions [5]. To decrease this heterogeneity, Jessen et al. proposed to exclude from this concept those subjects whose cognitive complaints could be accounted for by other disorders (psychiatric, neurological or systemic), drugs or their abuse [5].
Research in this field has been focused on tracking biomarkers that could define the preclinical AD phase in this population, characterising risk groups to start potential treatments that could delay disease progression [7,8]. The most frequently used techniques are cerebrospinal fluid analysis and different neuroimaging modalities such as magnetic resonance imaging (MRI), fluorodeoxyglucose positron emission tomography (PET), amyloid PET or Tau PET.
The morphometric analysis of MRI images of the brain has become a widely used approach to investigate changes in brain structure in neurodegenerative disorders. Typically, changes in the grey matter have been assessed using T1-weighted images and the most frequently used methods to analyse them include the volumetric comparison of (manually, semi-automatically or automatically) delineated regions of interest (ROIs), whole-brain voxel-based comparison of grey matter (called voxel-based morphometry or VBM) and cortical surface-based comparison of cortical thickness. These methods of neuroimaging analyses have their own strengths and limitations and frequently show different results even with identical image sets [9,10].
On the other hand, studies that evaluate changes in brain structure in subjects with SCD compared to control participants have shown heterogeneous results, in terms of areas affected and statistical significance, even in the AD-related structures such as the hippocampus [11][12][13]. The aim of this systematic review is to give an overview of studies examining the differences in the grey matter volume of the brain between individuals with a clinical diagnosis of SCD and cognitive unimpaired persons detected by MRI.

Methods
This systematic review was conducted in accordance with the Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) guidelines [14].

Selection criteria
We included studies that met the following inclusion criteria: (1) studies restricted to the English or Spanish language; (2) studies including a subjective cognitive impairment group, according to Jessen's criteria [5]; (3) studies including healthy controls (HC); and (4) studies measuring grey volume by MRI. We discarded studies according to the following exclusion criteria: (1) singlesex studies, (2) fMRI studies, (3) studies performing any kind of clinical treatment (chemotherapy, drugs, memory training, physical exercise, etc.), (4) studies restricted to APOE carriers in their sample, (5) studies with a history of hypertension or vascular disease as a selection criterion or (6) systematic reviews, meta-analyses and letters.

Study selection
Two reviewers (PA and OEZ) independently performed the search up to November 2020. After the removal of duplicates, the titles and abstracts of the remaining articles were screened for eligibility. Additionally, we also screened the references cited in the relevant articles to include key studies that had not been previously detected following a snowball technique. Then, the full text of the elected articles was screened according to the selection criteria. Disagreements on study selection were resolved by a third independent reviewer (MR).

Data extraction
We performed a systematic extraction of the following variables from all eligible manuscripts: year of publication, journal, MRI field strength (1.5 or 3 Tesla), type of the study (retrospective vs prospective), sample size in each clinical group (SCI vs HC), age in each clinical group (mean and standard deviation), sample recruitment (population-based, mixed or memoryclinic sample), type of analysis, software used for the analysis, ROIs studied, segmentation applied in those studies based on ROI analysis, main results and statistical significance of the findings. All studies in which the SCD sample was not recruited exclusively from memory clinics were included in the category "mixed". For statistical analysis purposes, we dichotomised the variable sample recruitment into memory clinic vs "other" (population-based and mixed) sample. The variables were previously defined and operationalised in an Excel template. A replicate of the data collection sheet can be found in Supplementary material S3.
Lastly, we further studied if any of the variables included were associated with the identification of statistically significant findings in the study. To compare continuous variables between the groups, we performed a one-way ANOVA or Kruskal-Wallis test and two-tailed Student's T or Wilcoxon rank-sum test, when appropriate. To compare categorical variables, chi-squared or Fisher tests were used as appropriate.

Quality assessment
We used the Newcastle-Ottawa Scale (NOS) assessment to determine the quality of the studies selected [15].

Eligible and included studies
Our parallel searches yielded a total of 365 (PubMed) and 463 (WoS) articles. After removing duplicates, a total of 425 articles were screened by title/abstract reading. After further reading and applying inclusion/exclusion criteria, 51 studies were selected for review (Fig. 1).

Study characteristics
Most of the studies were published between 2010 and 2020 (n = 46). Only 5 studies were published during the first decade of the 2000s, being 2015 the year in which most studies were published (n = 8). The different ways of sample recruitment were patients or referrals from Fig. 1 Flowchart summarising the search and selection of sources of evidence following PRISMA guidelines memory clinics (n =31), population-based cohort (n = 12) or a combination of both (n = 8). Three studies were based on pre-existing data banks. Regarding the design of the study, most of the articles included were retrospective (n = 42).
The studies used different techniques to study the volume of grey matter in the brain. Twenty-six studies exclusively performed a ROI analysis, being the most common analysis; 10 studies exclusively performed a VBM analysis and 6 exclusively analysed the cortical thickness. The remaining 9 studies performed a combination of two types of analysis (ROI and VBM analysis n = 5 or ROI and cortical thickness n = 4; results compared in Supplementary Table 1). Hence, 35 studies performed ROI analysis, 15 VBM analysis and 10 studies cortical thickness analysis.

Voxel-based morphometry analysis
Fifteen studies performed a VBM analysis (Table 1), and 8 found statistical differences between SCD and HC participants [16,18,21,23,24,[26][27][28]. Five of them found grey matter volume reduction in the hippocampus in SCD compared with HC [21,24,[26][27][28]. Saykin et al. [27] found bilateral volume reduction in the whole hippocampus, Liang et al. [24] found bilateral volume reduction in the hippocampal tail and Perrotin et al. [26] found bilateral volume reduction in the CA1. Lastly, 2 studies only found unilateral volume reduction in the right hippocampal in the SCD group [21,28]. Additional temporal areas were also involved in 3 studies. Volume reductions were found in the SCD group compared with HC in the right insula [18], the right amygdala [21] and the inferior temporal gyrus [23].
In the frontal lobe, Saykin et al. [27] found volume reductions in the whole lobe in the SCD group compared with HC. Other studies found differences in some specific frontal areas like the bilateral [16] and left [18] superior frontal, bilateral [22] or left [18] medial frontal, left inferior frontal [23], the bilateral anterior cingulate [21,23] and the left orbitofrontal cortices [23]. In the parietal cortex, Choi et al. (2015) found SCD volume reductions in the left superior and inferior cortex and in the right precuneus. Hafkemeijer et al. [21] also found bilateral precuneus atrophy in the SCD group compared with HC. In the occipital lobe, volume reductions were found in the bilateral cuneus [21], right calcarine and lingual gyrus [23]. Finally, only one study found a higher volume in SCD compared with HC, located in the paracentral lobe [24].
Entorhinal cortex Three studies found a reduced volume in the entorhinal cortex (EC) bilaterally in SCD compared with HC [40,47,51]. On the contrary, one study also analysed this ROI, but did not find any significant difference [44].
Amygdala Five studies found less grey matter volume in the amygdala in SCD compared with HC, 3 of them bilaterally [49,56,57] and 2 in the right hemisphere [51,60]. Three studies did not find differences between groups [21,37,46].
Cingulate cortex One study found grey matter atrophy in the posterior cingulate in SCD compared with HC [57]. Two studies did not find statistical differences between groups [23,28].
Other Scheef et al. [48] found the cholinergic basal forebrain (Ch1/2 and Ch 4p) smaller in SCD compared with HC. Zhao et al. [57] found the temporal lobe, the occipital lobe and the insular cortex smaller in SCD than in HC. Other studies analysed different brain areas like the thalamus, the putamen, the accumbens nucleus, the caudate nucleus, the globus pallidus [21], the corpus callosum [23], the precuneus, the parahippocampus [13], the inferior parietal, the middle temporal lobe or the retrosplenial cortex [13], but did not find any significant difference between SCD and HC.

Factors determining the statistical significance of findings
We observed that the studies with a recruitment sample in a memory clinic tend to identify more frequently statistically significant findings compared with those with a population-based or mixed recruitment (70% vs 50%, p = 0.09). Moreover, articles that identify statistically significant findings were published earlier than those without statistically significant findings (median 2015 vs 2017, p = 0.03). However, we did not find any other variable related to the statistical significance of findings (type of the study, sample size, age, nor MRI strength, p > 0.05) (Supplementary Table 2).

Quality assessment
All 51 studies included in this review received quality assessment (Table 4) following the Newcastle-Ottawa Quality Assessment Scale [15]. Out of a maximum of 9 points, the average was 6.84, indicating good overall quality in the articles selected for review. However, only two studies correctly reported a non-response rate [25,67], being Ivanoiu et al. [67] the only study obtaining the maximum score. The lowest score was 5 out of 9 points (n = 4).

Discussion
The main goal of this systematic review was to investigate whether individuals with SCD present volumetric or grey matter changes when compared with cognitively normal subjects. The main finding is that, among the reviewed studies, there is not a homogeneous and consistent structural change found in SCD compared with HC. The studies that found significant differences (disregarding the analysis method used) did so in the medial temporal lobe, a region implicated in AD [68,69]. However, the results we have observed are heterogeneous in the different imaging analysis methods included in this review. Analysing VBM results from 15 studies, half of them (n = 8) found significant results between groups, and the other half (n = 7) did not. In those studies that found structural atrophy in SCD compared with HC, the hippocampus was the most affected area. Interestingly, the    [41] right hippocampus was found to be smaller more often than the left one, being the right hippocampus affected every time that hippocampal volume was decreased in SCD. This hippocampal asymmetry was analysed for mild cognitive impairment (MCI) and AD groups in a meta-analysis [70]. In contrast with our study, this metaanalysis found a left-less-than-right atrophy pattern and a poorer performance in episodic memory tests in subjects with less left than right hippocampal volume. Fewer studies found affected areas that are also part of the temporal lobe (the amygdala, the insula and the temporal gyrus). Interestingly, up to 5 studies observe decreased volume in different regions of the frontal lobe in participants with SCD. This structure is not typically affected in the early stages of AD and possibly represents the heterogeneous aetiology of this clinical syndrome. Evaluating the results from 35 neuroimaging studies using ROI analysis, half of the sample found significant results (n = 18), and the other half did not (n = 17). The temporal lobe was also the most studied area of the brain in ROI studies. Specifically, the hippocampus, the amygdala, the entorhinal cortex and the posterior cingulate cortex were the most studied regions of interest. Although these areas are usually affected in mild and advanced stages of AD [71][72][73], there is no clear constant evidence of significant differences between SCD and HC individuals in these areas. One of the main limitations of ROI analysis may be the predetermination of the areas to be studied, especially when the underlying causes for SCD are not always AD-related. This selection bias can be avoided using other kinds of analyses such as voxel-based analysis. ROI segmentation is another possible source of bias, especially when manual segmentation is used. The distinction between manual versus automatic segmentation of ROIs could then be expected to be a determinant factor in the finding of significant differences. Manual segmentation was the gold standard for hippocampal volumetry [74,75] but heterogeneity in anatomic definitions and tracing guidelines have hampered comparisons among different studies using hippocampal volumetry for diagnosis. Semiautomatic segmentation tries to solve this limitation and to reduce the inter-and intraobserver variability but fails to do so to the full extent [76]. Automatic segmentation is more consistent and time-efficient The "*" means a star or point allotted for each category Heterogeneity was also observed in studies measuring cortical thickness. Six studies showed statistically significant differences in the cortical thickness of participants with SCD compared to controls, while 4 did not. Although the entorhinal cortex is one of the most affected structures when significant cortical thinning was found, more studies are needed to consider it a reliable biomarker of preclinical AD.

Limitations
The main limitations of our study are the different research settings and operationalisation of SCD used in the studies, the heterogeneity inherent to this clinical syndrome and the small sample of the studies measuring structural changes. Methodologically, an explanation for these heterogeneous results could be that the term SCD was recently established by Jessen in 2014, unifying the diverse diagnostic criteria and terminology used until then to refer to this potential early state of AD. Additionally, SCD may include vastly diverse samples, since it may include patients who underly AD pathology, other kinds of neurodegenerative disorders or cases in which memory complaints are simply associated with normal ageing. More consistent results may be expected by selecting participants with SCD and specific features which increase the likelihood of the presence of preclinical AD (referred as SCD plus [5]). Moreover, participants are studied in different research environments such as clinical settings and population-based cohorts. Rodríguez-Gómez et al. [77] found that SCD patients who have been referred to a memory clinic had an increased risk of developing cognitive impairment than patients from the general population. In this line, we found a trend pointing towards the recruitment from memory clinics as a predictive factor of statistical differences. On the other hand, the use of larger samples and multimodal analysis techniques might help to establish regions associated to SCD and its progression to then develop an early and accurate diagnosis of preclinical AD. Finally, another possible limitation of our study might be the publication bias, given that studies that do not find significant differences are less likely to be published, thus reducing our sample.

Conclusion
As a conclusion, we have found that studies assessing volumetric or grey matter changes in subjects with SCD when compared with cognitively normal subjects showed heterogeneous results. Almost half of the studies do not find any significant difference between both groups, and when differences are observed, diverse structures are involved. However, the medial temporal lobe is the structure more frequently observed.