Exploring Individual Brain Heterogeneity in a Failed Disease-Modifying Drug Trial to Treat Alzheimer's Disease

Serena Verdi¹², Neil Oxtoby¹, Jonathan M. Schott², Andre F. Marquand³⁴, James H. Cole¹²

¹ UCL Hawkes Institute, Department of Computer Science, University College London, UK
² Dementia Research Centre, UCL Queen Square Institute of Neurology, London, UK
³ Donders Centre for Cognitive Neuroimaging, Radboud University Medical Centre, Nijmegen, The Netherlands
⁴ Department of Psychiatry, UMC Utrecht, Utrecht, The Netherlands

Background: Alzheimer's disease (AD) is characterised by heterogeneity that is often obscured by conventional group-average analyses. Neuroanatomical normative modelling enables quantification of individual deviations from healthy brain charts, providing personalised measures of neuroanatomical heterogeneity.

Methods: We applied neuroanatomical normative modelling to longitudinal MRI from 1,120 cognitively unimpaired, amyloid-positive participants enrolled in the A4 Study, a phase III trial of solanezumab. Regional cortical thickness and subcortical volumes were compared with a large normative reference cohort (~58k  healthy individuals) to generate regional Z-scores. Regions with Z < −1.96 were classified as outliers and summarised as total outlier count (tOC). Longitudinal changes in tOC, inter-individual Hamming dissimilarity, and cognition (PACC) were analysed using mixed-effects and spline-based generalised least squares models, adjusting for relevant demographic and clinical covariates.

Results: Total outlier count increased significantly in both treatment groups over follow-up and was greater in the solanezumab arm at Week 240 (ΔtOC = 1.20, 95% CI 0.01–2.39, p = 0.049). Neuroanatomical outlier patterns became increasingly dissimilar between individuals over follow-up (β = 0.0127, p < 0.001), with a greater increase in the solanezumab group than placebo (time × treatment interaction: β = 0.0059, p = 0.002). Higher baseline neuroanatomical heterogeneity was associated with steeper cognitive decline over follow-up (time × tOC interaction: χ²(2) = 33.8, p < 0.001), and inclusion of baseline tOC significantly improved the longitudinal PACC model (likelihood ratio χ²(3) = 102.4, p < 0.001). Baseline neuroanatomical heterogeneity was also positively associated with tau PET SUVR (β = 6.30, p = 0.020) and CSF pTau217 (β = 5.27, p = 0.002), while no significant associations were observed with plasma NfL or GFAP.

Conclusions: Neuroanatomical heterogeneity increased throughout the trial, was associated with greater cognitive decline and Alzheimer's disease pathology, and differed between treatment groups. Normative modelling provides biologically meaningful, individual-level measures of disease heterogeneity that may improve participant stratification and provide sensitive outcome measures for future Alzheimer's disease clinical trials.