Improving Patient Subtyping on Longitudinal Data using Representations from Mamba-based Architecture

A research team has proposed a self-supervised machine learning model based on the Mamba architecture to improve how patients are grouped using electronic health record data, according to a preprint posted to arXiv on 26 Jun 2026 [1]. The model, described in a paper submitted by Md Mozaharul Mottalib, is designed to address the complexity and irregularity inherent in temporal EHR datasets [1][2]. Effective patient subtyping, or clustering, can inform precision medicine by identifying groups of patients with similar clinical trajectories [2]. The study evaluates the proposed Mamba-based model on both public and private real-world EHR datasets, demonstrating that its design choices lead to better predictive performance compared to competitive baseline models [1][2]. The authors also assessed several clustering techniques applied to the learned representations, reporting that the findings offer valuable insights for subtyping patients based on temporal records [2]. The implementation code has been made publicly available on GitHub [2]. The preprint was posted to arXiv, an open-access repository for electronic preprints that is not peer-reviewed but is moderated before posting [6]. As of November 2024, the repository was receiving about 24,000 new articles per month [6]. The paper appears under the Machine Learning category and was submitted as a single version on 26 Jun 2026, with a file size of 385 KB [1]. The work falls within a broader landscape of community-driven innovation on the platform, where third-party collaborators develop tools such as the Bibliographic Explorer and CORE Recommender to enhance discovery of relevant research [5].