Track papers across labs, topics, and trends.
Search papers, labs, and topics across Lattice.
This paper evaluates traditional, hybrid (autoencoder embeddings + K-means), and deep clustering methods on heart failure patient cohorts using EHR data from the All of Us Research Program. They find traditional methods are surprisingly robust due to deep clustering's design for image data rather than tabular EHR data. To improve deep clustering for EHRs, they introduce an ensemble-based deep clustering approach that aggregates cluster assignments from multiple embedding dimensions, outperforming other methods when combined with traditional clustering in an ensemble framework.
Deep clustering, often designed for image data, underperforms traditional methods on tabular EHR data, but ensembling deep and traditional clustering methods yields superior patient stratification.
In electronic health records (EHRs), clustering patients and distinguishing disease subtypes are key tasks to elucidate pathophysiology and aid clinical decision-making. However, clustering in healthcare informatics is still based on traditional methods, especially K-means, and has achieved limited success when applied to embedding representations learned by autoencoders as hybrid methods. This paper investigates the effectiveness of traditional, hybrid, and deep learning methods in heart failure patient cohorts using real EHR data from the All of Us Research Program. Traditional clustering methods perform robustly because deep learning approaches are specifically designed for image clustering, a task that differs substantially from the tabular EHR data setting. To address the shortcomings of deep clustering, we introduce an ensemble-based deep clustering approach that aggregates cluster assignments obtained from multiple embedding dimensions, rather than relying on a single fixed embedding space. When combined with traditional clustering in a novel ensemble framework, the proposed ensemble embedding for deep clustering delivers the best overall performance ranking across 14 diverse clustering methods and multiple patient cohorts. This paper underscores the importance of biological sex-specific clustering of EHR data and the advantages of combining traditional and deep clustering approaches over a single method.