Participants
We included memory clinic patients with a diagnosis of subjective cognitive decline (SCD), mild cognitive impairment (MCI), or dementia from the RWTH Aachen Memory Database from the University Hospital (Germany) and the Maastricht BioBank Alzheimer Center Limburg (BB-ACL) (The Netherlands) study. Patients were included if there was baseline data available on CSF Aβ-42, MRI ratings, and neuropsychological measures at baseline and at least one cognitive follow-up visit. Exclusion criteria were baseline age < 45 years and cognitive impairment caused by other than neurodegenerative or cerebrovascular disease (e.g., post-traumatic, post-infectious, chronic-inflammatory). The study was approved by the local ethics committees (EK 018–19 for Aachen, METC 09-3-037, METC 09-3-038, and METC 15-4-100 for Maastricht) and conducted according to The Code of Ethics of the World Medical Association (Declaration of Helsinki).
Clinical Assessment
All patients were assessed through standard protocol procedures at each site, including neurological and psychiatric assessment, medical history, and neuropsychological examination. Information on medical history was provided by the patients or their informal caregivers, and included, amongst others, vascular risk factors such as hypertension, hypercholesterolemia, and diabetes (type 1 and type 2), as well as vascular diseases such as coronary artery stenosis, carotid artery disease and myocardial infarction. Medication usage was also provided.
Neuropsychological examination was performed according to routine protocols at each site. For the current study, we selected measurements of global cognition, as well as measures categorized a priori within four cognitive domains: memory, processing speed, executive functioning, and language. Global cognition was measured either with the Mini-Mental State Examination (MMSE; RWTH Aachen and BB-ACL) (18) or with the Montreal Cognitive Assessment (MoCA; only RWTH Aachen) (19). If MMSE scores were unavailable (10%), MoCA scores were converted to MMSE scores (20). Memory was measured by the CERAD-NAB-Plus verbal learning task (21) in RWTH Aachen and the 15 Words Test (15WT), a Dutch version of the Auditory Verbal Learning Task (AVLT) (22), in BB-ACL. Processing speed was measured by either the CERAD Trail Making Test A (TMT-A) (23) or by the Concept Shifting Task A (CST-A; only BB-ACL) (24). The Trail Making Test B (TMT-B) and Concept Shifting Task B (CST-B) were used to assess executive functioning. To measure language, a semantic verbal fluency test (one minute, animals) was used. For all tests, raw scores were converted to Z-scores using validated standardized demographic-adjusted test norms. Z-scores below − 5 and above 5 were rounded to -5 and 5, respectively, to avoid extreme outliers in cognitive performance.
Baseline clinical diagnoses were established by specialized medical doctors in the memory clinics. Persons with normal performance on neuropsychological tests were classified as SCD. A diagnosis of MCI was made according to the criteria of Petersen (25) and a clinical diagnosis of dementia was based on the Diagnostic and Statistical Manual of Mental Disorders, Fourth Edition (DSM-IV) criteria. Cognitive testing was performed at baseline and up to five years of follow-up. At the BB-ACL site, structural follow-ups were performed at 1, 2, 3, and 5 years, whereas at RWTH Aachen follow-ups were dependent on clinical referral.
Csf Ad Biomarker Assessment
Levels of amyloid-beta42 (Aβ–42), phosphorylated tau 181 (p-tau), and total tau (t-tau) were obtained from the CSF as part of clinical routine, using lumbar puncture. For both centers, peptide and protein levels were measured by Fujirebio Innotest enzyme-linked immunosorbent assays (ELISA; Innogenetics, Ghent, Belgium). Cut-offs for biomarker abnormalities were determined by Gaussian mixture modelling (26) for each center: Aβ-42 < 722 pg/ml and p-tau > 54 pg/ml for RWTH Aachen hospital, and Aβ-42 < 773 pg/ml and p-tau > 65 pg/ml for BB-ACL. Persons were classified as A + when having abnormal Aβ-42 levels. We excluded persons with abnormal p-tau levels in the A- classification.
Mri Assessment
All persons underwent either a 1.5T (RWTH Aachen only) or 3T MRI scan (RWTH Aachen and BB-ACL) including both T1 and T2 weighted, as well as FLAIR images, in accordance with the study protocol at each site. Images were assessed by two independent raters blinded to clinical information, using validated semi-qualitative visual rating scales. Average values of the raters were reported. We used the Scheltens scale for assessment of medial temporal atrophy (MTA) with a score ranging from 0–4 (27), the Koedam scale for assessment of posterior atrophy with a score ranging from 0–3 (28), and the Fazekas scale for rating WMH with a score ranging from 0–3 (29). The total count of cortical and subcortical microbleeds, ischemic strategic and non-strategic infarcts and intracranial macrohemorrhage were also determined. Persons were classified as V + if they had a Fazekas score ≥ 2 (30), or one or more cortical or subcortical microbleeds, strategic or non-strategic infarcts, or intracranial macrohemorrhage. If none of the above was present, persons were classified as V-.
Patient Classifications
In our main analyses, all persons were classified in AVMRI subgroups based on presence of amyloid abnormality (A) and MRI vascular burden (V): (1) no amyloid abnormality and no MRI vascular burden (A-V-); (2) only MRI vascular burden (A-V+); (3) only amyloid abnormality (A + V-); and (4) both amyloid abnormality and MRI vascular burden (A + V+). In secondary analyses, we used two additional vascular definitions: vascular risk (Vrisk) and vascular disease (Vdisease). For vascular risk, Vrisk+ indicated the current presence of either hypertension, hypercholesterolemia, diabetes, or medication use for these conditions. For vascular disease, Vdisease+ indicated the current presence of either coronary artery disease, carotid artery stenosis, past myocardial infarction, past vascular surgery, or use of antithrombotics.
Statistical Analyses
Pearson chi-squared and ANOVA tests were used to compare baseline characteristics between the four AVMRI groups based on amyloid abnormality and MRI vascular burden. To this end, Aβ-42, p-tau, and t-tau levels, and microbleed count were log transformed due to their skewed distributions.
Subsequently, linear mixed models were used to predict slopes of decline in global cognition, memory, processing speed, executive functioning, and language within the four AVMRI groups, by including time x group interactions in the model. In secondary analyses, two other definitions of V were used, i.e., vascular risk factors (Vrisk) and vascular disease (Vdisease).
Additionally, to investigate the role of individual factors of MRI vascular burden, we performed linear mixed models including interactions of WMH and microbleeds (total and cortical microbleeds only) with amyloid β1–42 abnormality in the prediction of cognitive decline in all domains. We did not include the presence of infarcts, macrohemorrhages, and subcortical microbleeds in the individual analyses, as the prevalence in our sample was low (7.5%, 0.9%, and 4.8% respectively).
As a sensitivity analysis, we repeated our analyses only in persons without dementia (SCD and MCI) and investigated differences between diagnostic groups by exploring interactions with Aβ-42 abnormality and the AVMRI groups on cognitive decline. All linear mixed model analyses used the restricted maximum likelihood (REML) method and were adjusted for age, sex, and education years. A model with random intercept and random slopes, and center as random factor was used. All analyses were performed using SPSS Statistics version 26.0 software (IBM, Armonk, NY, USA) and with significance levels set at p < 0.05.