Prediabetes and structural brain abnormalities: Evidence from observational studies

Summary Type 2 diabetes mellitus has been linked to structural brain abnormalities, but evidence of the association among prediabetes and structural brain abnormalities has not been systematically evaluated. Comprehensive searching strategies and relevant studies were systematically retrieved from PubMed, Embase, Medline and web of science. Twelve articles were included overall. Stratified analyses and regression analyses were performed. A total of 104 468 individuals were included. The risk of infarct was associated with continuous glycosylated haemoglobin (HbA1c) [adjusted odds ratio (OR) 1.19 (95% confidence interval [CI]: 1.05‐1.34)], or prediabetes [adjusted OR 1.13 (95% CI: 1.00‐1.27)]. The corresponding ORs associated with white matter hyperintensities were 1.08 (95%CI: 1.04‐1.13) for prediabetes, and 1.10 (95%CI: 1.08‐1.12) for HbA1c. The association was significant between the decreased risk of brain volume with continuous HbA1c (the combined OR 0.92, 95% CI: 0.87‐0.98). Grey matter volume and white matter volume were inversely associated with prediabetes [weighted mean deviation (WMD), −9.65 (95%CI: −15.25 to −4.04) vs WMD, −9.25 (95%CI: −15.03 to −3.47)]. There were no significant association among cerebral microbleeds, hippocampal volume, continuous total brain volume, and prediabetes. Our findings demonstrated that (a) both prediabetes and continuous HbA1c were significantly associated with increasing risk of infarct or white matter hyperintensities; (b) continuous HbA1c was associated with a decreased risk of brain volume; (c) prediabetes was inversely associated with grey matter volume and white matter volume. To confirm these findings, further studies on early diabetes onset and structural brain abnormalities are needed.

microvascular complications. 10 A dose-response effect of chronic hyperglycemia on microvascular complications exists, which is termed the metabolic memorial effect. 11 Prediabetes is the early stage of diabetes development that includes the impaired fasting glucose (IFG) and impaired glucose tolerance (IGT) conditions, and can be diagnosed using fasting glycaemia and 2 hour glucose load test according to World Health Organization (WHO) definitions. 12 Evidence proves that glycosylated haemoglobin (HbA 1c ) can also be a measure of prediabetes. 13 Indeed, there is strong evidence that suggests prediabetes is associated with an increasing risk of stroke, 14 dementia 15 and cognitive impairment. 16 The relationship between prediabetes and different kinds of structural brain abnormalities is still controversial, 17 although similar micro-and macrovascular dysfunction were shown to be present. 18 In contrast, in an elderly study population, there was no relationship reported between prediabetes and lacunar infarcts (LIs), cerebral microbleeds (CMBs), WMHs, or smaller brain volumes. 19 These inconsistent findings may arise from different definitions and thresholds for prediabetes as well as small sample sizes. Becoming aware of the relationship between prediabetes and structural brain abnormalities would be beneficial to the prevention and treatment of related brain diseases. 20 To our knowledge, this is the first meta-analysis to examine the available evidence of a relationship between prediabetes status and risk for structure brain abnormalities. Considering previous inconsistent results, a meta-analysis to explore the association of prediabetes with the risk of structure brain abnormality may help to clarify this issue. Thus, we searched the related articles and performed a systematic review and comprehensive analysis.

| Study inclusion
The databases PubMed, Embase, Medline and Web of Science were searched for relevant studies published until May 2019. The included terms 'glucose blood level', 'haemoglobin a1c', 'blood fluctuation', 'impaired glucose tolerance', 'blood glucose', 'prediabetes', 'white matter hyperintensities', 'brain atrophy', 'brain haemorrhage', 'lacunar stroke', 'lacunar infarct', 'total cerebral brain volume', 'white hyperintensity volume', 'brain infarction', 'brain infarct', 'cerebral microbleed', 'hippocampal hyperintensity volume', 'magnetic resonance imaging' were used alone and in combination. The search strategy was supplemented by inspecting the references of the included articles. This report was conducted according to the Meta-analysis Of Observational Studies in Epidemiology 21 and the Preferred Reporting Items for Systematic Reviews and Meta-Analysis 22 guidelines.

| Inclusion and exclusion criteria
Studies were considered for inclusion using the following criteria: (a) was an original article recently published in English, (b) defined prediabetes or impaired glucose tolerance (IGT) or impaired fasting glucose (IFG) and any imaging appearances of structural brain abnormalities clearly, (c) investigated either continuous or categorial structural brain abnormalities, (d) measured the structural brain abnormalities using Magnetic Resonance Imaging (MRI), (e) used physical diagnosis of prediabetes, IGT, IFG or continuous HbA 1c , (f) provided quantitative measures of the association between prediabetes, IGT or IFG and any type of structural brain abnormalities, and their 95% confidence intervals (CIs), (g) used cross-sectional, casecontrol or cohort epidemiological study designs. Exclusion criteria were as follows: (a) the publication was a review, case report, animal study or letter to the editor, (b) the publication did not clearly define clinical outcomes, (c) the authors could not provide valid data after being contacted, (d) the publication provided duplicated data.
For this meta-analysis, only cohort studies about the association between continuous HbA 1c with any type of structural brain abnormalities and cross-sectional studies about the relationship between prediabetes and any type of structural brain abnormalities were included.

| Data extraction and quality assessment
Searching and screening were completed independently by two reviewers (X.Y.T. and Y.P.H.) and any discrepancies were resolved by discussion. Two investigators (X.Y.T. and Y.P.H.) independently search the extracted the data from the enrolled studies. Two investigators (X.Y.T. and F.Q.L.) independently utilised the Newcastle-Ottawa Quality Assessment Scale criteria (NOS) 23 to assess the risk of bias. We rated the quality of the studies by awarding stars following the guidelines of the Newcastle-Ottawa Scale. Three factors were considered for the included cohort studies: selection, comparability, and outcome; for cross-sectional studies, selection, comparability and exposure were considered. If there was disagreement, the investigators discussed the study with the other authors to arrive at a consensus.

| Statistical analysis
Heterogeneity between studies was evaluated by I 2 metric, and the variance between studies by Tau 2 . Random-effects models were performed if I 2 > 50% and fixed-effects models were performed if I 2 ≤ 50%. Odds ratio (OR) as measure of association across all studies were pooled in categorial data defined as meeting the diagnostic criteria or not. Mean difference was pooled by using the inverse variance weighting method and presented as weighted mean difference (WMD) with 95% confidence interval in continuous data. If studies had both unadjusted and covariate-adjusted ORs, we chose the latter.
We regarded IGT and IFG as prediabetes.
In categorial data, for studies providing β, we convert this to OR. Potential publication bias was evaluated by Egger's asymmetry test. 24 P-values were two-tailed, and P < .05 was considered statistically significant. The statistical analyses were performed with STATA version 12.0 (Stata Corporation, College Station, TX).

| Quality assessment
Quality assessment results of the studies all received 6 to 8 stars using the Newcastle-Ottawa scale (NOS) evaluation tool. The quality of the included studies was high. (Table S4). ( Figure 3).
T A B L E 1 Characteristics of the studies on the relationship between prediabetes and structural brain abnormalities
According to the different methods of prediabetes' diagnosis, we divided the data into four groups: HbA 1c 5.7%-6.5%, HbA 1c 6.0%-6.5%, IFG or IGT and IFG. Subgroup analysis revealed that nonsignificant association was found in both HbA 1c 5.7%-6.5% and IFG

| Prediabetes vs continuous structural brain abnormalities
There were three references for white matter volume (Q = 1.93,

| Publication bias
According to the Cochrane Handbook version 5.1.0, 34 as a rule of thumb, tests for funnel plot asymmetry should be used only when there are enough studies included in the meta-analysis, because when there are fewer studies the power of the tests is too low to distinguish chance from real asymmetry. In this study, the P value of the Egger test was >.05 (P = .095) for the relationship between prediabetes and infarct, indicating no significant bias among them. The funnel figure of these studies showed a symmetrical inverted distribution, which is consistent with the results of Egger test ( Figure S2).

| DISCUSSION
In this study, we analysed the accumulated evidence to explore the relationship between prediabetes or continuous HbA 1c and structural brain abnormalities from observational studies. Our results indicate that (a) both prediabetes and continuous HbA 1c were significantly associated with an increased risk for infarct or white matter hyperintensities; (b) continuous HbA 1c was negatively correlated with brain volume; and (c) prediabetes was inversely associated with grey matter volume and white matter volume.
Although valuable information regarding relationships between prediabetes or continuous HbA 1c and different kinds of structural The association between continuous HbA 1c with brain volume. Where I 2 is the variation in effect estimates attributable to heterogeneity, overall is the pooled fixed effect estimate of all studies. Subtotal is the pooled fixed effects estimate of sub-group analysis studies. Weights are from fixed-effects analysis. Percentage of weight is the weight assigned to each study, based on the inverse of the within-and between-study variance. The size of the grey boxes around the point estimates reflects the weight assigned to each study. The summarized studies were adjusted for age, sex and BMI. Abbreviation: OR, odds ratio The association between prediabetes with infarct, A, and white matter hyperintensities, B. A, The association between prediabetes with infarct. B, The association between prediabetes with white matter hyperintensities. Where I 2 is the variation in effect estimates attributable to heterogeneity, overall is the pooled fixed effect estimate of all studies. Subtotal is the pooled fixed effects estimate of sub-group analysis studies. Weights are from fixed-effects analysis. Percentage of weight is the weight assigned to each study, based on the inverse of the withinand between-study variance. The size of the grey boxes around the point estimates reflects the weight assigned to each study. The summarized studies were adjusted for age, sex and BMI. Abbreviations: OR, odds ratio; WMD, weighted mean deviation F I G U R E 5 The association between prediabetes with continuous white matter volume, A, and continuous grey matter volume, B. A, The association between prediabetes with continuous white matter volume. B, The association between prediabetes with continuous grey matter volume. Where I 2 is the variation in effect estimates attributable to heterogeneity, overall is the pooled fixed effect estimate of all studies. Subtotal is the pooled fixed effects estimate of sub-group analysis studies. Weights are from fixed-effects analysis. Percentage of weight is the weight assigned to each study, based on the inverse of the within-and between-study variance. The size of the grey boxes around the point estimates reflects the weight assigned to each study. The summarized studies were adjusted for age, sex and BMI. Abbreviation: WMD, weighted mean deviation

| THE POSSIBLE MECHANISMS
Prediabetes stages can influence changes in cerebral energy homeostasis, may cause inflammation, and impact the vasculature at the arterial and the capillary level. 36 Elevated blood sugar leads to the endothelial dysfunction of the cerebral or microcirculatory system, 18 thus contributing to cerebral perfusion deficits and developing into chronic ischemia of the brain tissue. 37 Disruption of the blood-brain barrier can occur because of the production of reactive oxygen species and limited antioxidant defences due to hyperglycemia. 38 In earlier stages of prediabetes, perivascular edema can cause cumulative insidious perivascular tissue damage, eventually resulting in the rarefaction and demyelination of white matter in the pathology, which can induce structural abnormalities and be shown as WMH on MRI. 39 Neurodegeneration and brain atrophy can be caused by cerebral insulin resistance, which may impair regional glucose metabolism and disrupt the intracellular release, and extracellular clearance, of bamyloid. 30 There is also evidence that other structural brain abnormalities associated with prediabetes exist. In a population-based cohort study by Agtmaal et al., a significant association was found between prediabetes and cerebrospinal fluid (CSF) (β = 3.9, 95% CI = 0.3-7.6), and a similar association was observed with continuous HbA 1c (β = 0.09, 95% CI = 0.06-0.11). 30

| Prospects
Future studies should pay more attention to the relationship between prediabetes or continuous HbA 1c and structural brain abnormalities according to different age groups. More systematic literature reviews on the association between prediabetes and structural brain abnormalities should be down to enrich the current findings, for it may add to the evidence that prediabetes is not a benign state. 40 Additionally, this would indicate that the window of opportunity for the prevention of brain disease in diabetes could be provided by prediabetes.
In conclusion, the available evidence indicates direct association of prediabetes or continuous HbA 1c and structural brain abnormalities risk. Thus, prevention and screening of structural brain abnormalities should begin in prediabetes stages. High-quality, longitudinal and agerelated studies are needed to improve our understanding and cognition of this association.