This was a cross-sectional study of hearing loss prevalence carried out by means of the open-access, mobile-based Hearing Test app [26]. Subject consent to use the results for scientific purposes was required the first time the app was started. The consent to research was issued by the Bioethics Committee at the Medical University in Wroclaw. Evaluation of the app, rationale behind the app, and comparison with other studies concerning mobile and web-based pure-tone hearing screening have been presented in prior articles [24,25,27,28].

The participants were recruited via the Hearing Test app that runs on Android systems. The app is offered free of charge in the Google Play store, where the user can become familiar with the app's features and download it to a phone or tablet. Thus, eligibility criteria for participants only included access to an Android device that is supported by the app. The test could be carried out by the phone owner or with his or her assistance.

The study included tests carried out on mobile devices that supported the Hearing Test app for which calibration coefficients had been determined. All the tests in the study were carried out on bundled headphones that were supplied by the manufacturer in a set with the device. Devices of the same model used common calibration coefficients that were determined by the biological method [27,28] on the basis of at least 16 measurements carried out by subjects with normal hearing on different devices of the model [25]. This method is characterized by the standard error of determining the calibration coefficients below 5 dB and within-model variability at 4 dB [25]. The headphone connection status was monitored by the app, whereas the usage of bundled headphones was confirmed by the user. Tests taken using headphones other than bundled headphones were not taken into account.

The hearing test consisted of self-determining the quietest audible sound by adjusting its intensity using the “I can hear” and “I can't hear” buttons. The intensity of a test tone could be changed many times. The quietest audible sound was confirmed with the “Barely audible” button. A modulated test tone was used, with a 100% modulation depth, 2 Hz modulation frequency, and an intensity changed in steps of 5 dB. When the intensity of the tone exceeded a 40 dB hearing level (HL), a narrowband masking noise was generated contralaterally at an intensity of 40 dB HL. For test tones above 60 dB HL, the intensity of the masking noise was increased to 60 dB HL. By default, the tests were carried out for frequencies from 250 Hz to 8 kHz; however, this range could be changed in the settings. The time to determine the hearing threshold was measured separately for each frequency. The hearing threshold determined in this way has been compared in previous works with pure-tone audiometry. A difference was found at the level of 2.6 dB (SD 8.3) [24], and the absolute difference was found at 8.8 dB [17].

Ambient noise was monitored during the measurement after obtaining the user's consent to access the microphone resources. An equivalent continuous A-weighted sound pressure level LAeq was registered in accordance with the Android specification [29] that sets a reference point at 90 dB SPL.

At the end of the test, the users could optionally enter their age to compare the obtained hearing threshold against the age-based norm. By design, the app was developed for adults. Therefore, the user chose his or her age by selecting a value from 18-90 years. Providing the age was not obligatory for this study, to minimize the amount of potentially incorrect data. All the entered age values were recorded in the database to eliminate tests for which several different values were given. The age was also determined from the note, where it could optionally be given. User gender was not collected.

The user's country was determined using the geoPlugin service [30] on the basis of the IP address.

In the period from November 23, 2016 to November 22, 2019, 733,716 tests were carried out on 236,716 devices in 212 countries. The hearing threshold measurement at the fundamental frequencies 0.25 kHz, 0.5 kHz, 1 kHz, 2 kHz, 4 kHz, 6 kHz, and 8 kHz was performed in 728,674 of 733,716 (99.3%) tests. After examining headphone connection status and rejecting the tests carried out with disconnected headphones, 637,169 of 733,716 (86.8%) tests were obtained (Figure 1).

Time limits were introduced to discard tests with a duration that actually prevented a correct measurement. The grand standard deviation of the hearing threshold in relation to the duration of the measurement at a single frequency is presented in Figure 2. Stabilization of the deviation is observed at longer durations. It has been assumed that tests in which the measurement at any frequency was carried out quicker than 6.7 seconds are subject to significant error. This threshold was set heuristically by fitting the data with a lognormal distribution and assuming its value at a cumulative probability density level of 0.99 (Figure 2). The time requirements were met by 239,752 of 733,716 (32.6%) tests.

Many devices were used for more than one test. To eliminate tests repeated by the same person or carried out in bulk (eg, in outpatient clinics or as part of screening programs) only one test from each device was qualified for the analysis. The last test was selected, at first considering the test with an age provided. Finally, 116,733 of 733,716 (15.9%) tests were obtained for the analysis. Age was provided for 8194 of 116,733 (7.0%) tests, and monitoring for ambient noise was present in 30,119 of 116,733 (25.8%) tests, whereas only 2556 of 116,733 (2.2%) tests included both age and noise data.

The subject's age was provided for 8194 of 116,733 (7.0%) tests that qualified for the analysis. The age characteristics of the research group is shown in Table 1. The age group with the highest number of tests was 30-39 years, while the fewest subjects were in the oldest groups (80-89 years and >89 years). The median age was 39 years, and the average age was 40.0 years.

Tests with age provided were compared with those without age provided to estimate the bias resulting from the generalization of the results for the whole group. There were no statistically significant differences between the distributions of the hearing thresholds at the level of P=.05.

Hearing threshold in relation to age was analyzed based on all 7332 of 116,733 (6.3%) tests for which the age provided was within the range of 20 to 89 years. Tests completed by subjects aged 18, 19, or 90 years were omitted from the analysis to obtain equal age ranges (20-29 to 80-89 years). Moreover, the ages of 18 and 90 years were characterized by a higher number of tests in comparison with the adjacent age groups, suggesting the presence of additional data, such as results for subjects younger than 18 years or older than 90 years, or with incorrectly provided data. The median thresholds by age group and frequency are given in Figure 3 and in Multimedia Appendix 1. A comparison with the prior work presented in Figure 3 is provided in the Discussion section.

The prevalence of hearing loss was determined based on 733,716 tests conducted in 212 countries. The required sample size of 90 tests was reached for 74 countries. The prevalence of hearing loss, defined as a mean hearing threshold >25 dB HL in the better ear at 0.5-4 kHz, is presented in Figure 4, Figure 5, and Multimedia Appendix 2. The highest prevalences were obtained for Pakistan, Bangladesh, and India at 37.8% (95% CI 31.4-44.2), 32.2% (95% CI 23.6-40.8), and 28.5% (95% CI 27.1-29.9), respectively, whereas the lowest prevalences were obtained for Taiwan, Finland, and South Korea at 9.6% (95% CI 7.2-12.1), 9.8% (95% CI 5.7-13.9), and 10.2% (95% CI 9.5-10.9), respectively. The global prevalence of hearing loss was calculated at 15.6% (95% CI 15.4-15.8).

The effect of ambient noise was estimated based on 30,119 of 116,733 (25.8%) tests conducted with ambient noise monitoring. The difference in the hearing threshold in relation to the ambient noise level is presented in Figure 6. Hearing thresholds were adjusted for ambient noise by assuming the reference level LAeq=35 dB(A) [32]. The average decrease in the hearing threshold was obtained at a level of 2.53 dB (SD 0.74 dB). Additionally, the distributions of the hearing threshold in the groups with and without ambient noise monitoring turned out to be significantly different at the level of P<.001, with the monitoring group having a hearing threshold that was lower by 1.25 dB (SD 0.12 dB). This produces an overall effect of ambient noise of 3.78 dB (SD 0.75 dB), which corresponds to a 4.16% (SD 1.46%) reduction in the country-specific prevalence.

The highest ambient noise levels were found in Pakistan, Bangladesh, and India, which corresponded to 10.7%, 8.6%, and 6.2% decreases, respectively, in the prevalences of hearing loss. Nevertheless, these countries still had the highest prevalences of hearing loss.

Differences in hearing thresholds between device models were found to be significant at the level of P=.01. These differences result directly from the biological calibration method that is based on determination of the reference sound level in relation to subjects with normal hearing. The standard error of the biological calibration estimated at 4.2 dB for 16 independent measurements [25] justifies the discrepancies. After adjusting for the device model, the mean hearing threshold by country changed by a mere −0.07 dB (SD 0.96 dB).

This paper presents the prevalence of hearing loss worldwide based on the 116,733 hearing tests conducted by Android users on mobile devices. The global prevalence of hearing loss was 15.6% (95% CI 15.4-15.8). Statistically significant differences were found across countries (P<.001). Of the 212 countries, 74 countries exceeded the number of tests required for the assumed sample size. Amongst these countries, the highest prevalences were found in Pakistan, Bangladesh, and India at 37.8% (95% CI 31.4-44.2), 32.2% (95% CI 23.6-40.8), and 28.5% (95% CI 27.1-29.9), respectively, whereas the lowest prevalences were found in Taiwan, Finland, and South Korea at 9.6% (95% CI 7.2-12.1), 9.8% (95% CI 5.7-13.9), and 10.2% (95% CI 9.5-10.9), respectively. The risk of hearing loss in Pakistan was 74.6% (95% CI 38.2-110.7) higher than in Taiwan. The absolute difference was 28.2% (95% CI 14.4-41.8).

Of the 733,716 tests, 116,733 (15.9%) were selected for the analysis. The number of tests rejected due to incompleteness, short duration, lack of headphones, or repetitions on a single device seems reasonable, especially as prior to performing a proper test, users often become familiar with the app by conducting a trial test.

In addition to Finland, relatively low hearing loss values were obtained for the other two Scandinavian countries, Sweden and Norway, at 12.1% (95% CI 8.4-15.9) and 12.7% (95% CI 8.1-17.4), respectively. The prevalences of hearing loss for the other two Asian Tigers (ie, Singapore and Hong Kong) were slightly higher, at 14.6% (95% CI 10.0-19.2) and 15.3% (95% CI 10.3-20.4), respectively. The high prevalence of hearing loss in India, which, along with Pakistan and Bangladesh, is one of the most populated countries in the world, is supported by the findings of Garg et al [33]. Similar hearing loss prevalences were found in most Western and Central European countries, which were slightly higher than Australia (11.4%, 95% CI 9.8-13.0) and Canada (12.2%, 95% CI 10.4-13.9) but lower than that in the United States (14.6%, 95% CI 13.9-15.4). Similar prevalences of hearing loss were obtained among the countries of the former Soviet Union (ie, Russia, Belarus, Ukraine, and Kazakhstan), at levels between 15.4% (95% 12.9-17.9) for Ukraine and 18.9% (95% CI 14.3-23.6) for Kazakhstan.

In Italy, 20.0% (95% CI 19.3-20.7) of the subjects had hearing loss, and this value was definitely higher than in the other Western and Central European countries. The credibility of this result is questionable due to the much higher use of apps in this country, being the highest in the world and reaching 222.3 people per million population (Multimedia Appendix 2). Attention should also be paid to the low Iranian result of 10.3% (95% CI 7.8-12.7), which significantly differs from other countries of North Africa and the Middle East. On the other hand, according to a study by the World Health Organization [5] in the region of North Africa and the Middle East, the prevalence of hearing loss is lower than in developed countries.

The required sample size was not reached in a significant number of African countries. In the poorest countries, the limiting factor is access to mobile devices and the internet. The same situation may also occur in countries with significant social stratification, contributing to the lower prevalences of hearing loss resulting from a lack of data from the poorer part of the population. The lack of data from China is related to the lack of active Google Play service in this country.

The prevalence of hearing loss in countries has been correlated with the infant mortality rate [34], which is often used as an indicator of the health status in a country. A Cronbach alpha coefficient of 0.76 (95% CI 0.52-0.90) has been obtained.

This paper presents global country-specific prevalences of hearing loss based on self-tests carried out by Android users. Unsupervised self-tests require additional quality control based on the test duration and connection status of the headphones. Moreover, ambient noise and calibration method introduce additional bias. Despite this, hearing thresholds measured by means of mobile devices were congruent with data reported in the literature, whereas uniform recruitment criteria facilitate the comparison between countries. Hearing tests on mobile devices may be a valid tool in epidemiological studies carried out on a large scale.