Quantitative audiometric measurements identified more genetic associations with sensorineural hearing loss (SNHL) than diagnostic codes, despite a substantially smaller study population. The audiometry-based approach also yielded higher heritability estimates and stronger associations between polygenic risk scores and self-reported hearing loss.
Researchers conducted 2 genome-wide association studies (GWASs) using whole-genome sequencing data from Vanderbilt University Medical Center's BioVU biobank. The diagnostic code–based analysis included 61,468 participants: 10,164 with SNHL and 51,304 controls, whereas the audiometry-based analysis included 16,057 participants with available hearing measurements. Both cohorts were restricted to participants of European genetic ancestry. The diagnostic approach classified hearing loss using medical record codes, while the audiometry-based approach characterized hearing sensitivity as a continuous trait using 3-frequency pure-tone averages (PTAs) at 500, 1,000, and 2,000 Hz from the better-hearing ear. Data analysis was conducted from August 2025 to April 2026.
The primary analyses compared the identification of genome-wide significant genetic loci and estimates of single-nucleotide variant–based heritability between the 2 phenotyping approaches. Secondary analyses examined genetic correlations with previously reported hearing-related traits and evaluated polygenic risk scores (PRS) in an independent cohort from the National Institutes of Health's All of Us Research Program. GWAS models were adjusted for age, sex, and genetic ancestry.
The PTA-based GWAS identified 3 genome-wide significant loci that were mapped through functional annotation to 4 genes: EML6, SPTBN1, ARHGEF28, and EYA4. In contrast, the diagnostic code–based analysis identified no loci reaching genome-wide significance, despite including nearly 4 times as many participants. The audiometry-based approach also identified 41 loci meeting the threshold for suggestive significance, compared with 21 in the diagnostic code–based analysis.
The 4 genes identified through functional annotation have previously been associated with hearing or sensory and nervous system function. Notably, pathogenic variants in EYA4 can cause autosomal dominant nonsyndromic hearing loss. The researchers suggested that quantitative audiometric phenotyping may improve the detection of genetic signals associated with hearing sensitivity.
Heritability estimates further distinguished the 2 approaches. The PTA-based analysis yielded an estimated single-nucleotide variant–based heritability of 11.8%, compared with 2.9% for the diagnostic code–based analysis. Among 7,479 participants included in both cohorts, diagnostic coding and PTA measurements showed a weak correlation, suggesting that diagnostic classifications captured only a limited proportion of variation in measured hearing thresholds.
The audiometry-based analysis also demonstrated statistically significant genetic correlations with previously reported hearing-related traits, including hearing aid use, clinically diagnosed or self-reported hearing loss, and PTA measurements from an earlier GWAS. The strongest correlation was observed for hearing aid use, with a correlation coefficient of 0.985. No statistically significant genetic correlation was observed with speech recognition in noise, which the researchers considered clinically distinct from hearing sensitivity measured by pure-tone averages.
In the independent All of Us cohort of 109,096 participants, PTA-derived polygenic risk scores showed a statistically significant association with self-reported deafness or serious hearing difficulty, with 9% higher odds of reporting hearing loss for each 1-standard-deviation increase in the score. Diagnostic code–derived scores showed no corresponding association. Although the audiometry-derived scores demonstrated stronger associations, the magnitude was modest, and the authors acknowledged that their usefulness for predicting hearing loss in individual patients remains unproven.
Several limitations warrant consideration. The primary genetic analyses included only participants of European genetic ancestry, limiting generalizability to other populations. The audiometric analysis relied on a single 3-frequency average from the better-hearing ear, potentially overlooking differences in hearing sensitivity across frequencies or between ears. Differences in age definitions between cohorts may have introduced bias. Additionally, the independent validation relied on a single self-reported hearing question focused on deafness or serious hearing difficulty, potentially excluding participants with milder hearing loss. The authors emphasized the need for larger, more diverse cohorts with detailed audiometric measurements to validate the findings and improve risk estimation.
The authors suggested that future studies incorporating individual frequency thresholds and full-audiogram measurements may help identify genetic influences that are not captured by conventional pure-tone averages, but cautioned that the current polygenic risk scores' clinical predictive value remains unproven. "Future studies should aim to investigate whether similar genetic associations can be found in other populations," wrote lead study author Andrea J. DeFreese, AuD, of Vanderbilt University Medical Center, Nashville, Tennessee, and colleagues.
DeFreese, Nayak, and Jan reported grant funding. Jan also reported serving as a site principal investigator for Eli Lilly clinical trials and consulting for Merck Manuals outside the submitted work. No other disclosures were reported.
Source: JAMA Otolaryngology–Head & Neck Surgery
Editor's note: The corresponding author confirmed that the diagnostic code–based cohort comprised 10,164 cases and 51,304 controls (61,468 participants), correcting discrepancies in the published abstract and Key Points.
