🤖 AI Summary
This work proposes a novel method for computing multidimensional patient similarity from unstructured clinical text to support precision medicine. For each patient, all clinical notes are first embedded and aggregated into a matrix, from which a low-rank latent representation is learned to capture shared patterns across dimensions such as clinical history, treatment regimens, and adverse events. By modeling the entirety of clinical notes as a matrix and leveraging low-rank representations for similarity analysis, this approach overcomes the limitations of traditional methods that rely on structured data or keyword matching. Experiments on data from 4,267 breast cancer patients demonstrate that the proposed method effectively supports downstream tasks including personalized treatment recommendation and toxicity risk prediction.
📝 Abstract
Clinical notes hold rich yet unstructured details about diagnoses, treatments, and outcomes that are vital to precision medicine but hard to exploit at scale. We introduce a method that represents each patient as a matrix built from aggregated embeddings of all their notes, enabling robust patient similarity computation based on their latent low-rank representations. Using clinical notes of 4,267 Czech breast-cancer patients and expert similarity labels from Masaryk Memorial Cancer Institute, we evaluate several matrix-based similarity measures and analyze their strengths and limitations across different similarity facets, such as clinical history, treatment, and adverse events. The results demonstrate the usefulness of the presented method for downstream tasks, such as personalized therapy recommendations or toxicity warnings.