2021/11/11 by Oliver Carr, Avelino Javer, Carr, Oliver +7
Computer Science · Health Professions · Medicine · #Machine Learning in Healthcare #Artificial Intelligence in Healthcare #Diabetes Management and Education
paper · pdf · doi:10.48550/arxiv.2111.06152
The increase in availability of longitudinal electronic health record (EHR)\ndata is leading to improved understanding of diseases and discovery of novel\nphenotypes. The majority of clustering algorithms focus only on patient\ntrajectories, yet patients with similar trajectories may have different\noutcomes. Finding subgroups of patients with different trajectories and\noutcomes can guide future drug development and improve recruitment to clinical\ntrials. We develop a recurrent neural network autoencoder to cluster EHR data\nusing reconstruction, outcome, and clustering losses which can be weighted to\nfind different types of patient clusters. We show our model is able to discover\nknown clusters from both data biases and outcome differences, outperforming\nbaseline models. We demonstrate the model performance on 29,229 diabetes\npatients, showing it finds clusters of patients with both different\ntrajectories and different outcomes which can be utilized to aid clinical\ndecision making.\n