Multi-dimensional attention framework for personalized Alzheimer's disease progression prediction across sporadic and genetic risk cohorts

Alzheimer's disease manifests through heterogeneous progression patterns across diverse populations, yet current predictive models fail to capture this complexity while maintaining clinical interpretability. Here we present a multi-dimensional attention framework that simultaneously captures temporal dynamics and biomarker importance patterns to predict disease progression across three fundamentally different populations: the general late-onset population using the Alzheimer's Disease Prediction Of Longitudinal Evolution (TADPOLE) dataset (N=1669), cases with Down Syndrome-associated Alzheimer's disease using the Alzheimer's Biomarker Consortium - Down Syndrome (ABC-DS) dataset (N=396) and cases with autosomal dominant Alzheimer's disease using the Dominantly Inherited Alzheimer Network (DIAN) dataset (N=425). Our framework achieved multi-class area under the ROC curve (mAUC) values of 0.793 (TADPOLE), 0.680 (ABC-DS) and 0.902 (DIAN) when trained on each dataset independently to predict individuals' future diagnostic status (cognitively normal, mild cognitive impairment, or Alzheimer's disease) based on their longitudinal biomarker history, outperforming conventional approaches. The model generates individual-specific attention maps revealing distinct biomarker importance over time. Transfer learning from the TADPOLE dataset-which included neuroimaging data- improved prediction performance on the ABC-DS dataset with no neuroimaging data included from 0.680 to 0.771, demonstrating that disease mechanisms transcend both etiological boundaries and data modalities. Ultimately this framework could help enable precision medicine approaches for data-limited cohorts across the Alzheimer's disease spectrum.