Speaker
Description
The aetiology of hearing impairment is multifactorial, involving numerous structures within the auditory system. Although pure-tone audiometry remains the clinical gold standard for hearing device fitting, it lacks sensitivity to certain pathologies, including hidden hearing damage. Previous approaches to phenotype classification either relied solely on audiogram data or incorporated additional tests but were limited by small sample sizes. Here, we analyse a large clinical dataset from the Rigshospitalet University Hospital (Denmark) spanning 1995 to 2022, comprising ~300,000 audiograms from over 85,000 unique adult patients, including audiometric thresholds and speech audiometry results.We investigated Principal Component Analysis (PCA) and Uniform Manifold Approximation and Projection (UMAP) for two-dimensional visualization. While PCA captured major audiogram variance, it proved insufficient for multimodal audiological data. UMAP demonstrated greater utility by exploiting both global and local data structures, and was adopted as the basis for our Hearing Loss Map.Comparison with existing phenotype classifications revealed that up to 40% of the data remains unclassified, underscoring the need for a more comprehensive framework. We applied multiple clustering approaches to a sensorineural hearing loss (SNHL) subset — Gaussian Mixture Models and Leiden community detection — identifying novel phenotypes not captured by conventional schemes. A trajectory analysis further represents patients transitioning between phenotypic states over time, enabling longitudinal characterisation of hearing loss progression.We propose this UMAP-based Hearing Loss Map as a powerful tool for navigating the SNHL patient space, with applications in clinical decision-making, patient counselling, evaluation of novel treatments, and development of next-generation hearing technologies.