Defining the ATN framework using longitudinal biomarker trajectories reveals an emerging amyloid accumulation group

Background The ATN framework is defined by cross‐sectional biomarkers of β‐amyloid (Aβ), tau and neurodegeneration. Given that prevention trials, e.g., AHEAD 3‐45, are focused on individuals who have lower Aβ than established thresholds, we investigated whether defining the ATN framework using longitudinal biomarker trajectories might better identify an at‐risk sample within this boundary. Here, we applied a data‐driven method to re‐define the ATN with longitudinal biomarker data from the Harvard Aging Brain Study (HABS) and we then replicated this longitudinal framework in ADNI. Method 157 HABS participants were clinically‐normal at baseline and underwent at least two Pittsburgh Compound‐B [PiB]‐PET, Flortaucipir‐PET, and T1‐weighted MRI scans. To define longitudinal ATN, we applied latent class mixture models (LCMM) to each biomarker (global Aβ DVR, entorhinal tau SUVr, ICV‐adjusted hippocampal volume) separately, adjusting for age, and including random intercept and slopes. Akaike information criteria (AIC) determined the best‐fitting models out of two‐group or three‐group solutions with linear or spline‐link functions. We compared longitudinal ATN profiles on demographics and an optimized estimate of cognitive change (derived from longitudinal Preclinical Alzheimer Cognitive Composite (PACC) data). Result Aβ trajectories (Fig.1a) were best categorized by one stable (A→) and two accumulating subgroups, a predominantly amyloid‐negative at baseline subgroup (A‐↑) and an entirely amyloid positive at baseline subgroup (A+↑). Tau (Fig.1b) and neurodegeneration (Fig.1c) were optimally defined by stable (T→/N→) vs accumulating/atrophying (T↑/N↑) groups, respectively. These groups were replicated in ADNI (Fig.2). The entire A‐↑ subgroup were stable on T and N (A‐↑/T→/N→) and were predominantly A‐/T‐/N‐ at baseline (86%; Table 1). By contrast, 38% of A+↑ individuals changed on T, or T&N. A‐↑/T→/N→ demographically most closely resembled the longitudinally‐stable ATN group (A→/T→/N→), but were older, more likely to carry e4+ and exhibited higher baseline Aβ (Table 2). Although demonstrating Aβ accumulation, A‐↑/T→/N→ did not exhibit greater cognitive decline versus the stable group (A→/T→/N→; Fig. 3). Conclusion Our findings suggest that a longitudinal biomarker run‐in of Aβ‐PET may be useful for the identification of early‐risk groups for prevention trials. Future work will establish whether other features (e.g. genetics, neuroinflammatory markers, functional ima