Searching auditory phenotypes beyond audiometry from a large clinical dataset

Authors: Gerard Encina-Llamas1,2, Erik Kjærbøl3, Abigail Kressner3

1University of Vic
2Central University of Catalonia
3Copenhagen Hearing and Balance Center. Rigshospitalet University Hospital

The aetiology of hearing impairment is complex due to the likely involvement of several interconnected structures and cell types. Pure-tone audiometric threshold is the gold standard and main test to assess hearing in the clinics, and is the fundamental information used to fit hearing devices meant to compensate for the losses. Previous work has attempted to classify patients into different subtypes, known as auditory phenotypes, using two strategies: 1) classifying audiograms based on known audiometric shapes from pathological studies in animal models, and 2) classifying audiometric measurements using unsupervised statistical tools such as cluster analysis. Either strategy has its strengths and weaknesses, but both have a common pitfall since the audiogram is the only source of information. For a long time, researchers are aware that the audiogram is not sensitive to all pathologies in the peripheral auditory system, but this has gained traction in recent years with the discovery of hidden hearing loss and cochlear synaptopathy. Here we show a preliminary analysis on a very large clinical dataset collected at Rigshospitalet University Hospital in Copenhagen (Denmark) from 1995 to 2022 containing 84280 unique patients and 288295 air-conduction audiometric thresholds. Data was analysed using principal component analysis (PCA) and uniform manifold approximation and projection for dimension reduction (UMAP) for data visualization. Results show that most of the audiometric threshold did no fall into the previously proposed auditory phenotypes categories. The inclusion of additional measurements in the same patients beyond audiometric thresholds, such as speech audiometry results and acoustic reflex thresholds, is shown and discussed. A clustering algorithm based on Gaussian mixture models (GMM) is used to provide new auditory phenotypes based on the full data including other test results beyond audiometry.

This work was supported by the GN Foundation.