Precise phenotyping boosts genetic prediction of hearing loss

Andie DeFreese sits at the intersection of two worlds that rarely talk to each other. She is a clinical audiologist—someone who spends her days listening to patients describe what they hear and what they don't—and also a PhD candidate in hearing science at Vanderbilt Health. That dual vantage point led her to notice something that most researchers had overlooked: the way medicine officially records hearing loss doesn't match the way hearing loss actually happens.

When a patient walks into an audiology clinic with hearing problems, the audiologist can measure the loss with precision. They can identify which frequencies are affected, how severe the loss is, whether it's in one ear or both. But when that same patient's condition gets entered into the medical system for billing and research purposes, it gets flattened into a diagnostic code—a standardized label that serves the insurance industry but obscures the clinical reality. "Diagnostic codes are very common because that's the data that's available in large genetic biobanks," DeFreese explained. "Using them leads to a kind of gray area in which we're not able to accurately define who has hearing loss and who doesn't."

This gap between clinical precision and research data became the focus of her team's work. DeFreese, working with colleagues from audiology, ear surgery, and otolaryngology, hypothesized that if they could use actual clinical measurements instead of billing codes, they might be able to identify genetic patterns associated with hearing loss far more accurately. To test this, they de-identified detailed audiometric data from Vanderbilt's clinical records and linked it with genetic information from BioVU, Vanderbilt Health's biobank of anonymous genetic samples. The result was a dataset of 16,000 people with both precise hearing measurements and genetic information.

Using this richer dataset, the researchers developed polygenic risk scores—mathematical models that predict disease risk based on multiple genetic variants working together. The crucial finding: risk scores built from precise clinical phenotypes predicted hearing loss significantly better than risk scores built from the old diagnostic codes. The improvement wasn't marginal. It demonstrated that the way you describe a disease matters enormously when you're trying to predict who will develop it.

The implications ripple outward in multiple directions. For biobank research, the work underscores that institutions need to invest in capturing clinical detail, not just diagnostic labels. For hearing loss specifically, it opens a path toward tools that could be embedded in electronic health records or consumer genetic tests—systems that could flag which patients face elevated risk of developing hearing loss, and potentially which frequencies might be affected. Taha Jan, an otolaryngologist and corresponding author on the paper, noted the timeliness of the finding: the FDA has recently approved its first gene therapy for genetic hearing loss, making the ability to identify at-risk patients before symptoms emerge newly consequential.

But the work also points to a broader methodological lesson. Different medical specialties use diagnostic codes differently; some rely on them heavily, others less so. DeFreese emphasized that bridging this gap requires clinicians in the research room from the start. "The variability in how these different methodologies are used between specialties emphasizes the importance of bringing clinicians into the research team," she said. Vanderbilt's structure—with research labs adjacent to clinical care—positioned the institution to ask this kind of question and pursue it rigorously. The next phase of research will aim to build systems that can predict not just whether someone will develop hearing loss, but which specific frequencies will be affected, moving the field closer to truly personalized prediction and intervention.