Data-driven segmentation of audiometric phenotypes across a large clinical cohort

Abstract

Pure tone audiograms are used to assess the degree and underlying source of hearing loss. Audiograms are typically categorized into a few canonical types, each thought to reflect distinct pathologies of the ear. Here, we analyzed 116,400 patient records from our clinic collected over a 24-year period and found that standard categorization left 46% of patient records unclassified. To better account for the full spectrum of hearing loss profiles, we used a Gaussian Mixture Model (GMM) to segment audiograms without any assumptions about frequency relationships, interaural symmetry or etiology. The GMM converged on ten types, featuring varying degrees of high-frequency hearing loss, flat loss, mixed loss, and notched profiles, with predictable relationships to patient age and sex. A separate GMM clustering of 15,380 audiograms from the National Health and Nutrition Examination Survey (NHANES) identified six similar types, that only lacked the more extreme hearing loss configurations observed in our patient cohort. Whereas traditional approaches distill hearing loss configurations down to a few canonical types by disregarding much of the underlying variability, an objective probabilistic model that accounted for all of the data identified an organized, but more heterogenous set of audiogram types that was consistent across two large clinical databases.The pure tone audiogram is the current gold standard clinical hearing assessment. Clinical management of hearing loss, from diagnosis to intervention, largely depends upon quantifying pure tone thresholds at octave intervals between 250 and 8000 Hz 1 . Multiple factors including age, genetic background and noise exposure history can influence the level and configuration of pure-tone thresholds within the standard audiometric frequency range 2-5 .A long-standing goal has been to infer the underlying cause of hearing loss using the pure-tone audiogram, medical diagnoses and knowledge from post-mortem human temporal bone and animal studies 6-10 . These approaches have identified four essential types -(i) a normal audiogram, where thresholds across test frequencies are equal or better than 20 dB HL, (ii) a flat hearing loss, where audiogram thresholds are elevated outside of the normal range but are approximately flat across test frequencies, presumably due to possible metabolic losses in the stria vascularis; (iii) a sloping, high frequency hearing loss (HFHL) due to presumed sensorineural damage; and (iv) a mixed phenotype reflecting modest threshold shift at low frequencies and a steeply sloping loss at high frequencies, which has been interpreted as a mixture of metabolic and sensorineural contributors 7,11-15 .These categories are useful as a first approximation, but studies with larger data samples either note the presence of mixed phenotypes or phenotypes with indeterminate causes 6 . Studies using algorithmic clustering methods note the presence of a gradual continuum of shapes rather than distinct categories 16 . Additionally, a recent study that reimaged historical...

Age-Related Changes in the Cochlea and Vestibule: Shared Patterns and Processes

“…Various studies have used functional assessments to characterize ARHL and ARVL (Figure 3). These studies have shown that agerelated hearing loss assessed by pure-tone audiograms begins at 40 years of age and progresses bilaterally from high to low frequencies (summarized in Parthasarathy et al, 2020). The rate of hearing decline with age is highly variable (Gates and Cooper, 1991).…”

Web-based Psychoacoustics: Hearing Screening, Infrastructure, and Validation

“…(2) The test would be applicable even to individuals with audiometric hearing loss. This is because most individuals with hearing loss have a sloping audiogram with highfrequency loss (Parthasarathy et al, 2020), and the BMLD effect is known to be largely preserved when the low-frequency loss is not substantial (Jerger et al, 1984). Participants had to correctly respond to five out of six trials in each task to pass the headphone screening and proceed to the other tasks in the study.…”

Robust data-driven auditory profiling towards precision audiology

“…There have been some attempts to stratify HI listeners based on the shapes of their audiograms. Several classification schemes have been proposed in earlier studies, some of which were based on data-driven approaches (Bisgaard, Vlaming and Dahlquist 2010; Chang, Yoon, Kim, Baek, Cho, Hong, Kim and Moon 2019; Parthasarathy, Romero Pinto, Lewis, Goedicke and Polley 2020), where computational methods for data analysis were used for identifying the most common audiometric profiles. Based on results from human temporal bone studies, Schuknecht and Gacek (1993) proposed four different types of age-related hearing loss: sensory presbycusis, neural presbycusis, metabolic presbycusis and mechanical presbycusis.…”