Audiometric Phenotypes of Noise-Induced Hearing Loss by Data-Driven Cluster Analysis and Their Relevant Characteristics

Abstract

Background: The definition of notched audiogram for noise-induced hearing loss (NIHL) is presently based on clinical experience, but audiometric phenotypes of NIHL are highly heterogeneous. The data-driven clustering of subtypes could provide refined characteristics of NIHL, and help identify individuals with typical NIHL at diagnosis.Methods: This cross-sectional study initially recruited 12,218 occupational noise-exposed employees aged 18–60 years from two factories of a shipyard in Eastern China. Of these, 10,307 subjects with no history of otological injurie or disease, family history of hearing loss, or history of ototoxic drug use were eventually enrolled. All these subjects completed health behavior questionnaires, cumulative noise exposure (CNE) measurement, and pure-tone audiometry. We did data-driven cluster analysis (k-means clustering) in subjects with hearing loss audiograms (n = 6,599) consist of two independent datasets (n = 4,461 and n = 2,138). Multinomial logistic regression was performed to analyze the relevant characteristics of subjects with different audiometric phenotypes compared to those subjects with normal hearing audiograms (n = 3,708).Results: A total of 10,307 subjects (9,165 males [88.9%], mean age 34.5 [8.8] years, mean CNE 91.2 [22.7] dB[A]) were included, 3,708 (36.0%) of them had completely normal hearing, the other 6,599 (64.0%) with hearing loss audiograms were clustered into four audiometric phenotypes, which were replicable in two distinct datasets. We named the four clusters as the 4–6 kHz sharp-notched, 4–6 kHz flat-notched, 3–8 kHz notched, and 1–8 kHz notched audiogram. Among them, except for the 4–6 kHz flat-notched audiogram which was not significantly related to NIHL, the other three phenotypes with different relevant characteristics were strongly associated with noise exposure. In particular, the 4–6 kHz sharp-notched audiogram might be a typical subtype of NIHL.Conclusions: By data-driven cluster analysis of the large-scale noise-exposed population, we identified three audiometric phenotypes associated with distinct NIHL subtypes. Data-driven sub-stratification of audiograms might eventually contribute to the precise diagnosis and treatment of NIHL.

Prevalence of hearing screening failures in low‐risk childhood cancer survivors

“…Formal audiology testing would be necessary to clearly characterize the phenotype. 43,44 Regardless of the etiology, hearing preservation should include strategies to protect from further loss due to noise.…”

“…Noise‐induced hearing loss could certainly account for others; however, noise‐induced hearing loss typically manifests as a “notch” in the thresholds, which was not reflected in the screens. Formal audiology testing would be necessary to clearly characterize the phenotype 43,44 . Regardless of the etiology, hearing preservation should include strategies to protect from further loss due to noise.…”

Over-the-counter hearing aids: implementations and opportunities

“…While most potential OTC hearing aid users have highfrequency or flat hearing loss, some may have relatively steep sloping or notched hearing loss (Cruickshanks et al, 2020;Parthasarathy et al, 2020;Wang et al, 2021). To further address the need for those with sloping or notched hearing loss, additional two-channel amplification can be provided at high frequencies (bottom two panels in Figure 3).…”

Assessing Clustering Replicability in Sensorineural Hearing Loss: Insights from the UK's Largest Audiogram Cohort

“…Replicability is an established means of assessing the stability of a clustering solution. 23 In the absence of access to the external comparator dataset, cluster replicability can be assessed within a single study by comparing the clusters learned by adapting the model or adapting the dataset. One means of introducing these alterations is through bootstrapping to create new datasets with the same sample size as the original dataset but with the introduction of noise from sample generation with replacement.…”