MRI‐guided Clustering of Alzheimer’s Disease patients: A post‐hoc analysis of Phase 3 Trial of Solanezumab for Mild Dementia Due to Alzheimer’s Disease

Abstract

AbstractBackgroundAlzheimer’s disease (AD) is a phenotypically and pathologically heterogenous disorder. We used MRI in amyloid positive patients enrolled in the placebo arm of the Phase 3 Trial of solanezumab for mild dementia due to AD (EXPEDITION3) to identify more homogeneous subgroups based on variation in volumetrics.MethodsWe implemented the self‐organizing map algorithm on baseline volumetric MRI measures from nine brain regions of interest (ROIs) to cluster 1041 individuals enrolled in the placebo arm of the EXPEDITION3 trial. Volumetric MRI differences were compared among clusters. Demographics as well as baseline and longitudinal cognitive performance metrics were used to evaluate cluster characteristics.ResultsThree distinct clusters, with an overall silhouette coefficient of 0.383, were identified based on MRI volumetrics. Cluster 1 (N=400) had the largest baseline volumetric measures across all ROIs and the best cognitive performance at baseline (Table 1). Cluster 2 (N=269) had larger hippocampal and medial temporal lobe volumes, but smaller parietal lobe volumes in comparison with the third cluster (N=372) (Figure 1). Spider plots show the normalized test performance at baseline and follow‐up with larger values corresponding to worse test performance; Cluster 1 shows relatively little cognitive change (Figure 2). Spider plots of regional MRI volumes show smaller volumes for Clusters 2 and 3 relative to Cluster 1 and asymmetries that reflect regional differences in atrophy (Figure 2). The mean change from baseline to week 80 in the ADAS‐cog14 score for Clusters 1, 2, and 3 were 4.28 (SE=0.35), 10.14 (SE=0.51), and 7.90 (SE=0.43), respectively. Significant between‐group mean differences were observed between Clusters 1 and 2 (difference, 2.38; 95% CI, 1.85 to 2.91; P<0.001), Clusters 1 and 3 (difference, 1.93; 95% CI, 1.41 to 2.44; P<0.001), but not between Clusters 2 and 3 (difference, 0.45; 95% CI, ‐0.11 to 1.02; P=0.146) in ADAS‐cog14 (Figure 3).ConclusionsVolumetric MRI can be used to identify homogenous clusters of amyloid positive individuals with mild dementia. The groups identified differ in baseline and longitudinal characteristics. Cluster 1 shows little ADAS‐cog14 change over the first 40 weeks of study on placebo treatment and may be unsuitable for identifying early benefits of treatment.