We included cognitively normal participants with subjective cognitive decline (SCD) from the Subjective Cognitive Impairment Cohort (SCIENCe) at the Alzheimer Center Amsterdam who had two [18F]florbetapir PET scans available (n = 92). Individuals were referred to the memory clinic (n = 85) by their general physician, a neurologist, or a geriatrician and underwent an extensive standardized diagnostic workup that included a neurologic and neuropsychological examination, laboratory testing, and brain MRI. In a consensus meeting, participants were labeled SCD when cognitive performance appeared within normal limits, and criteria were not met for mild cognitive impairment (MCI), dementia, or other neurological or psychiatric diseases that could possibly cause cognitive complaints. Individuals received a diagnosis of MCI when they had cognitive impairment in one or more cognitive domains, while independence in functional abilities was still preserved . Individuals were diagnosed with dementia when they had cognitive impairment in two or more cognitive domains, which interfered with daily activities .
In addition, seven participants were included via the Dutch Brain Research Registry (hersenonderzoek.nl). They experienced cognitive complaints in absence of objective impairment and received the same baseline work-up. At annual follow-up visits, neuropsychological testing was repeated and diagnoses were re-evaluated.
PET and MRI acquisition
Baseline dynamic [18F]florbetapir PET scans were acquired on a Philips Ingenuity TF PET-CT (n = 82) or a Philips Gemini TF PET-CT (n = 10; Philips, Best, the Netherlands) scanner. These scanners were calibrated to each other. The scan protocol started with a low-dose CT for attenuation correction. Dynamic PET scans of 90 min (n = 82) were obtained starting directly after tracer injection of approximately 370 MBq [18F]florbetapir. During the course of the study, we demonstrated that scan duration could be reduced without compromising the reliability of results . Therefore, subsequent scans had a duration of 70 minutes (n = 9). One scan was terminated early after 79 min due to participant related issues. All underwent a follow-up [18F]florbetapir PET with a mean follow-up time of 2.5 ± 0.7 years (n = 17 90-min scan; n = 75 a 70-min scan). All scans were visually assessed as “positive” or “negative” by a trained nuclear physician, blinded to the amyloid status at the other time point.
Baseline dynamic [18F]flortaucipir PET scans were acquired on a Philips Ingenuity TF PET-CT scanner (Philips, Best, the Netherlands, n = 44). Because substantial tau pathology within A− cognitively normal individuals is not expected to be present, we selected more A+ individuals for the [18F]flortaucipir PET in order to have a broader spectrum of amyloid and tau pathology, resulting in an A+ rate of this subset of about 33%. The scan protocol started with a low-dose CT for attenuation correction. Starting simultaneously with tracer injection of approximately 240 MBq [18F]flortaucipir, a 60-min dynamic emission scan was initiated. After a 20-min break and following a second low-dose CT for attenuation correction, an additional dynamic emission scan was performed during the interval 80–130 min post-injection. This dual time point protocol was validated previously . Forty-two individuals underwent a follow-up [18F]flortaucipir PET scan using the same procedure with a mean follow-up time of 2.1 ± 0.3 years.
Baseline structural MRI images were obtained at five different systems (GE Discovery MR750 3T (n = 22), Philips PETMR 3T (n = 51), Signa 1.5T (n = 1), Titan 3T (n = 17), and external scan (n = 1)). The protocol included 3D T1-weighted images, 3D T2-weighted images, and 3D T2-weighted fluid-attenuated inversion-recovery (FLAIR) images . T1-weighted images were used for coregistration to PET images and for determination of the N status. Follow-up MRI was available for 79 individuals with a mean follow-up time of 2.9 ± 0.9 years.
Data were reconstructed while using standard LOR RAMLA reconstruction algorithm with corrections for dead time, decay, attenuation, random coincidences, and scatter. Images were reconstructed with a matrix size of 128 × 128 × 90 and a voxel size of 2 × 2 × 2 mm3. For [18F]flortaucipir, both scan sessions (0–60 and 80–130 min) were co-registered into a single dataset of 29 frames (1 × 15, 3 × 5, 3 × 10, 4 × 60, 2 × 150, 2 × 300, 4 × 600, and 10 × 300 s), in which the last 10 frames belonged to the second PET session. 3D T1-weighted MR images were co-registered to PET images using Vinci software (Max Planck Institute, Cologne, Germany). Next, regions of interest (ROIs) were defined on the co-registered MRI using the probabilistic Hammers brain atlas  in PVElab. Receptor parametric mapping (RPM) was used to generate parametric binding potential (BPND) images with cerebellar grey matter as a reference region using PPET [11, 15,16,17]. For [18F]florbetapir, we calculated (volume weighted) mean cortical BPND in a composite ROI consisting of orbitofrontal, temporal, parietal, anterior cingulate, posterior cingulate, and precuneus regions [6, 18].
Biomarkers: a, T, N
Availability of biomarker status at two time points differed for each of the biomarkers (A: n = 92; T: n = 42; N: n = 79). For 39 individuals, a complete ATN profile over time could be constructed. The time difference with the [18F]florbetapir scan was 0.05 ± 0.15y for [18F]flortaucipir scans and 0.16 ± 0.62y for MRI scans. We used visual assessment of [18F]florbetapir PET scans to define A in the ATN classification. Since quantitative threshold-based methods usually have a high degree of concordance with visual assessment [19, 20], we chose visual assessment to be consistent with methods used in clinical practice. In additional analyses, we used continuous mean cortical BPND in a composite ROI. We used [18F]flortaucipir PET scans as biomarker for T. We pragmatically used Gaussian mixture modeling as an unbiased, data-driven approach, to obtain a threshold. We first averaged values for the anterior part of the lateral temporal lobe for left and right sides. Since the focus of this study was on cognitively unimpaired individuals, we a priori decided to select this region of interest to capture the earliest changes in neocortical areas . We then fit Gaussian Mixture Models with two components using the normalmixEM function in R. A threshold was derived representing the mean of the calculated mu of both components, resulting in a threshold 0.08 BPND. This threshold separated the two clusters with minimal overlap. We used the average medial temporal atrophy rating (MTA) on MRI as biomarker for N as determined by experienced neuroradiologists. Raters were blinded to amyloid status. For individuals < 65 years of age, an average MTA score of ≥ 1 was considered positive; for individuals ≥ 65 years of age, an average MTA score ≥ 1.5 was considered positive . Additionally, white matter hyperintensities were visually assessed using the Fazekas scale (range 0–3) . Microbleeds were assessed on T2-weighted images and defined as small dot-like hypointense lesions. They were counted and dichotomized into absent (0) or present (≥ 1 microbleed).
All participants underwent annual standardized neuropsychological assessments . For the memory domain, we used the Visual Association Test version A (VAT-A) and the total immediate and delayed recall condition of the Dutch version of the Rey auditory verbal learning task (RAVLT). For the language domain, we used category fluency (animals). For the attention domain, we used the Trail Making Test A (TMT-A) and Stroop task I and II (naming and color naming). For the domain of executive functioning, we used the TMT-B and Stroop task III (color-word). For global cognition, we used the Mini Mental State Examination (MMSE). Because the data were right-skewed, the raw test scores for TMT and Stroop were log transformed. Subsequently, values were inverted, so that a lower score implies worse test performance for all tests. We used available test results of visits before as well as after PET scans, in order to accurately estimate the cognitive slope. The neuropsychological tests administered most closely to baseline [18F]florbetapir were defined as baseline test results. In total, we used longitudinal cognitive data covering 4.9 ± 2.8 years. The proportion of missing tests ranged from 2.7% for MMSE to 8.5% for Stroop II and III. In total, 447 neuropsychological investigations of 92 patients were available (92 ≥ 2 visits, median 4).
All analyses are conducted in R version 4.0.3. We first compared demographic and clinical variables between baseline A− and A+ individuals using t-test, chi-square, or Mann-Whitney U test where appropriate.
Next, we described changes in biomarker status over time. We first investigated changes in A, T, and N biomarkers separately and then combined in ATN profiles. Then, we examined changes in amyloid status more closely. We categorized (change in) amyloid status as a four-level variable: negative at baseline and follow-up (negative-negative), negative at baseline and positive at follow-up (negative-positive), positive at baseline and negative at follow-up (positive-negative), and positive at baseline and follow-up (positive-positive). We investigated change in amyloid status in relation to actual BPND values at baseline and follow-up, using a division into low, grey zone, and high BPND with previously described thresholds by our group of 0.19 and 0.29 BPND . Next, we investigated which factors were associated with change from a negative to a positive amyloid status using logistic regression analyses. In model 1, baseline age, sex, education, baseline MMSE score, and APOE ε4 carriership were evaluated as individual predictors, with the group remaining A− at follow-up as reference group. In model 2, all predictors were entered simultaneously. In an additional analysis, we used amyloid accumulation rate as outcome, using linear mixed models (outcome: BPND composite ROI). We again assessed baseline age, sex, education, baseline MMSE score, and APOE ε4 carriership as predictors. In model 1, variables were assessed individually, and each analysis included the variable of interest, time, and the interaction between the variable and time. Model 2 included all predictors simultaneously (including time and all interactions between predictors and time).
Last, we used change in amyloid status as predictor of cognitive test performance over time, using linear mixed models. We used our four-level variable reflecting change in amyloid status, time, and their interaction as predictors (negative-negative (reference), negative-positive, positive-negative, and positive-positive). Baseline age, sex, and education were used as covariates. Outcome were neuropsychological test scores. Models included a random intercept and additionally a random slope when it improved the model (random slope included for VAT-A, TMT-A, Stroop I-III and MMSE). Separate analyses were run with different tests as outcome.