The association between diabetes measures and Alzheimer’s disease biomarkers in CSF – A meta‐analysis

Abstract

AbstractBackgroundDiabetes and blood glucose markers have previously been linked to cognitive decline. However, findings on the relation of diabetes with underlying Alzheimer’s disease (AD) biomarkers are inconclusive. We aimed to explore whether diabetes and blood glucose markers are associated with AD biomarkers in cerebrospinal fluid (CSF) by performing a meta‐analysis of existing studies.MethodA systematic search in PubMed and Embase with abstract and full‐text screening resulted in the inclusion of 13 studies with 5,052 participants in our meta‐analysis. We included studies that provided results on the association between a diabetes measure and an AD biomarker in CSF (Table 1). Diabetes measures included diabetes diagnosis or glucose measures, i.e. fasting blood glucose, HbA1c, or insulin resistance index. AD biomarkers included CSF amyloid‐beta42, p‐tau and/or t‐tau. Where needed, results were converted to partial or biserial correlation coefficients. We performed a random effects meta‐analysis on the correlation coefficients per CSF biomarker, using the DerSimonian and Laird procedure and Knapp‐Hartung method. Meta‐regression was used to test potential moderating effects of age, sex, setting (memory clinic or population), % dementia cases, correlation type (Pearson, partial, or biserial), diabetes measure (diabetes diagnosis or glucose marker), MMSE score, and APOEe4 carriership (yes/no). All analyses were performed using the metafor package in R.ResultOur meta‐analysis showed small but significant associations of diabetes measures with increased (more normal) levels of amyloid‐beta42 (r = 0.08, p = 0.04, Figure 1) and increased (more abnormal) levels of p‐tau (r = 0.09, p = 0.04, Figure 2) and t‐tau (r = 0.10, p = 0.04, graph not shown as comparable to p‐tau). Heterogeneity was high for all outcomes (I2>60%, p<0.05 for Q‐test, wide prediction intervals). Meta‐regression analyses showed that heterogeneity in CSF amyloid could partly be explained by setting, as the association between diabetes measures and amyloid was only present in memory clinic studies as compared to population studies (difference estimate = 0.14, p = 0.03). Similarly, studies with more dementia cases showed a stronger association with amyloid (slope estimate = 0.0029, p = 0.001). No other moderating effects were found.ConclusionOur findings suggest that diabetes measures are associated with tau‐related neurodegeneration that is independent from amyloid. This is valuable for improving diagnostics and treatment of patients with diabetes and AD.