The study’s findings carry significant implications for global health initiatives that rely on mobile technology to bridge gaps in hearing care. Hearing loss affects more than 5 percent of the world’s population, and nearly 80 percent of people with disabling hearing loss live in low- and middle-income countries. Yet the ear and hearing care workforce, including audiologists, remains severely limited across these regions, making accessible screening tools an urgent priority. The new research suggests that simply transplanting apps designed and validated in wealthy nations into different linguistic and cultural contexts may not deliver the promised benefits.

The hearWHO app, launched in 2019, uses a technique called digits-in-noise testing to estimate speech recognition thresholds. Rather than measuring the faintest sound a person can detect, as pure tone audiometry does, the test presents recorded digit triplets against background noise and asks listeners to identify them. This approach was considered promising for real-world settings because it does not require the artificially quiet environment that pure tone audiometry demands. The app employs antiphasic presentation, a method validated to enhance sensitivity and specificity compared to standard diotic delivery, and it applies dynamic cutoff values to account for non-native language effects and age-related performance differences.

Previous reviews of validated hearing assessment apps reported sensitivity values ranging from 0.982 to 1.00, suggesting the technology could reliably detect hearing loss. However, those studies focused almost exclusively on adults in high-income populations. The Tanzanian research team, led by Rachael Collins of Norfolk and Norwich University Hospital and colleagues from Kilimanjaro Christian Medical Centre and other institutions, set out to examine whether the app would perform in the low-resource setting for which it was ostensibly designed.

Over a three-day period, 52 patients self-presented to ear, nose, and throat outreach clinics at Huruma Hospital with ear-related complaints. All underwent clinical history and examination, and 29 patients who met the inclusion criteria, including a clinician-determined indication for audiometry and an age of at least 12 years, completed both pure tone audiometry and the hearWHO test. Pure tone audiometry was performed by a single audiologist using a calibrated portable Interacoustics AD226 audiometer with TDH-39 headphones, recording air conduction thresholds at 500, 1,000, 2,000, and 4,000 Hz. Hearing loss was defined using a four-frequency pure tone average exceeding 25 dB, consistent with WHO criteria for disabling hearing impairment.

The hearWHO testing presented its own logistical challenges. The app was installed on an Apple iPhone XR, which lacks a 3.5 mm audio port, so JBL 520BT Bluetooth headphones were used because no wired audiometric transducers were available. While the app does not specify a required headphone type, WHO guidance indicates it is designed for consumer-grade smartphone use rather than calibrated equipment. Background noise levels in the testing rooms were measured independently at between 51 and 56 dB using a sound level meter, below the app’s 60 dB threshold for proceeding. All patients had Swahili selected as their first language, and each completed three conditioning tests before the full assessment, with a Swahili-speaking nurse translator facilitating communication.

The diagnostic results were unexpectedly poor. Overall sensitivity, the proportion of true hearing loss cases correctly identified, was 0.45, while specificity, the ability to correctly identify normal hearing, was just 0.11. The positive predictive value of 0.53 indicated that a positive result was somewhat more reliable than a negative one, but the negative predictive value of 0.08 meant that a normal app result provided almost no reassurance. Ear-wise analysis revealed a striking pattern: the app correctly identified all ears with pure tone average-defined hearing loss, achieving perfect sensitivity of 1.00 for both ears, but specificity was extremely low at 0.08 for the left ear and 0.05 for the right, reflecting frequent false-positive classifications. Cohen’s kappa, a measure of agreement beyond chance, was 0.08 for the left ear and 0.04 for the right, indicating poor concordance with pure tone audiometry.

When stratified by age, sensitivity declined from 0.67 in participants under 40 to 0.29 in those aged 60 and above, suggesting reduced detection in older groups. This aligns with previous research showing that older age negatively impacts digits-in-noise test scores even after accounting for hearing loss, possibly because older adults are less familiar with smartphone technology. The researchers note that while mobile phones are nearly universal across Sub-Saharan Africa, smartphone ownership remains far less common, potentially compounding the difficulty for older participants navigating the app interface.

The most plausible explanation for the poor performance, according to the authors, is language. All participants were native Swahili speakers, and although Swahili was selected on the app interface, both the test instructions and the digits-in-noise assessment were delivered in South African English. Non-native speakers have been found to perform significantly worse on digits-in-noise testing, and this linguistic mismatch likely depressed scores across the board. The researchers also acknowledge that the assisted administration, with clinicians and translators supporting participants, may have artificially improved results compared to what would occur in real-world self-administered use, meaning the true performance in unassisted conditions could be even worse.

The study’s limitations include a small sample size, which reduces statistical power and generalizability, and a very small number of true negative cases that made specificity estimates unstable. Nevertheless, the authors argue that the findings raise important questions about how mobile health technologies intended for low- and middle-income countries are developed and validated. They contend that apps designed for these settings should be co-designed in partnership with the communities they aim to serve and developed within the contexts in which they will be used. Broader language options or alternative non-language-specific testing methods may be needed to ensure accessibility across diverse populations. The researchers also emphasize the need for pediatric hearing screening tools, noting that early detection of hearing loss in children is critical to prevent delays in speech, language, cognitive, and social development, though current digits-in-noise testing has not effectively differentiated conductive hearing loss from normal hearing in pediatric populations. Future studies, they suggest, should include larger samples from multiple regions with participants whose primary language is not English, ensuring that the promise of smartphone-based hearing screening reaches the populations who need it most.

Subject of Research: Validation of the hearWHO digits-in-noise app against pure tone audiometry in rural Tanzania

Article Title: Caution with app-based hearing self-tests with vulnerable populations: preliminarily findings of digits-in-noise (DIN) testing vs. pure tone audiometry (PTA) in a rural, low-resource outreach clinic in Huruma, Tanzania

Article References: Caution with app-based hearing self-tests with vulnerable populations: preliminarily findings of digits-in-noise (DIN) testing vs. pure tone audiometry (PTA) in a rural, low-resource outreach clinic in Huruma, Tanzania. (n.d.). https://doi.org/10.3389/fauot.2026.1896918

Image Credits: AI Generated

DOI: 10.3389/fauot.2026.1896918

Keywords: hearWHO app, digits-in-noise testing, pure tone audiometry, hearing screening, low-resource settings, Tanzania, mHealth, global health, audiology, smartphone diagnostics, language barriers, LMICs