Study participants were recruited offline, using face-to-face prompting. Eligibility assessment was conducted among subjects who owned an Android-based device with bundled headphones. Eligibility criterion, apart from the willingness to participate in the study, included owning the device, for which model-specific calibration coefficient has been previously determined by means of biological method [1]. All the study participants conducted examinations on their mobile devices.

Measurements involved conducting pure-tone audiometry and the test on a mobile device. The order was arranged in a counterbalanced manner. The measurements for both methods were conducted in a sound booth for the frequencies 250 Hz, 500 Hz, 1 kHz, 2 kHz, 4 kHz, 6 kHz, and 8 kHz.

Pure-tone audiometry was conducted by means of conventional 10 dB down and 5 dB up bracketing method (modified Hughson-Westlake method) in accordance with the recommendations of the British Society of Audiology [35]. The measurements were conducted by an audiologist on a clinical audiometer Interacoustic AD229e using TDH-39 headphones (Interacoustics, Denmark, Audiometer Allé, 5500 Middelfart) previously calibrated according to ISO 389-1:1998.

Hearing examinations on mobile devices were conducted using a free app Hearing Test [36,37] available on Google Play. The examinations involved self-determining of the lowest audible sound generated by the device and were audiologist-assisted. The participant changed intensity of the test sound using the buttons “I can hear” and “I cannot hear” and then confirmed the lowest audible sound using the button “Barely audible.” During the adjustment of the sound intensity, temporary test result was continuously presented on the device screen (Figure 1). The role of the supervising audiologist was to prevent obvious mistakes, such as switching the sides of headphones or omitting the frequency by accidental button pressing. The audiologist also provided assistance in cases of doubts concerning proper execution of the test. Test tone was amplitude modulated with the depth of 100% and the frequency of 2 Hz. Sound intensity was changed in 5 dB steps. When the intensity of the test sound exceeded 40 dB HL, a masking sound was generated contralaterally. The tests on mobile devices were conducted twice: test and retest. Before the retest, the headphones were removed and put on again.

Calibration of mobile devices was described in detail in the previous paper [1]. The objective of calibration was to determine the intensity of the signal generated by the device, which would generate a sound in bundled headphones at the reference level of 0 dB HL. Calibration was performed by means of a biological method, which led to elaboration of a new mobile hearing level (mHL) scale. The level of 0 dB mHL was defined as the sound level generated by the mobile set equivalent to the hearing threshold of normal-hearing persons, reduced by their estimated hearing threshold. The advantage of this approach is the ability to calibrate new device models based on measurements conducted by users themselves, which is essential due to high rotation of mobile models on the market. The calibration is required only for users who are the first to become the owners of new devices, whereas the others can reuse their results.

Calibration measurements involved self-determining the hearing thresholds by means of Bekesy’s method [1,10,11,22,34] by users aged 18-35 years who considered themselves normal-hearing persons. There were no other requirements for calibration. However, if these requirements were not met, a user could ask another person for help. These calibration thresholds were then decreased by the population median to obtain the final reference level. More precisely, the estimator of central tendency of the determined thresholds were approximated by 37th percentile and then decreased by the median of the hearing threshold estimated based on literature data [38] for the corresponding population, that is, aged 18-35 years [1]. Calibration measurements were conducted in uncontrolled conditions, irrespective of this research, before its commencement and when it was in progress, by application users who were not trial participants. To calibrate the model, at least 15 measurements were required to be conducted on various devices belonging to the same model [1]. After each subsequent measurement, the calibration results were automatically updated.

Sample size was determined on the basis of the standard deviation (SD) between the hearing thresholds in pure-tone audiometry and thresholds determined on mobile devices estimated at 8.42 dB [1]. With the statistical significance of 0.05, statistical power of 0.8, and the effect size of 2.0 dB, the sample volume was estimated on 140 ears (70 subjects). Hearing thresholds obtained on mobile devices have been compared with pure-tone audiometry and analyzed in test-retest examination. Differences in hearing thresholds, intraclass correlation coefficients, and Cronbach alpha with respective confidence intervals have been determined. The results were presented in dependency plots as well as in Bland-Altman plots.

In this paper, hearing thresholds determined by pure-tone audiometry were compared with hearing thresholds determined on Android-based mobile devices previously calibrated biologically in uncontrolled conditions. At the level of statistical significance P=.05, no differences in relation to the test order were observed. The mean hearing threshold difference of 2.6 dB (SD 8.3 dB) confirms the reliability of the method. Additionally, the mean test-retest difference of −0.1 dB (SD 4.4) indicates its high repeatability.

The results are consistent with the research presented in previous works. The SD of the difference at 8.3 dB (95% CI 7.9-8.7) corresponds to the value estimated for this calibration method of 8.42 dB [1] and is close to the value of 7.8 dB obtained in another study conducted on the same app Hearing Test [7]. The mean difference of 2.6 dB (95% CI 2.0-3.1) and the mean absolute difference of 6.5 dB (95% CI 6.2-6.9) are also comparable with those presented in [7], that is, 0.7 dB and 7.8 dB, respectively. Contrary to study [7], in which all the tests were carried out on one device, each test in this trial was performed on a different device, thus validating this calibration method.

SD of the difference for the proposed method was 8.3 dB (95% CI 7.9-8.7) with the number of differences no higher than 10 dB at 89% (95% CI 88-91). These values are comparable with the results determined on a set calibrated by a normal-hearing person with verified hearing threshold (SD 6.9 and 8.29 dB [10] and SD 7.88 dB [22]) and are significantly smaller when based on a single biological calibration conducted in uncontrolled conditions (SD 10.66 dB [22]). The highest calibration accuracy is obtained in laboratory conditions, which, however, is not scalable for the Android-based devices. In laboratory conditions, SD varies from 6.4 dB to 9.9 dB [5,6,25], whereas the number of differences not higher than 10 dB changes from 81% up to 96% [4,5,6,19], depending on other settings.

SD of test-retest difference was obtained at 4.4 dB (95% CI 4.2-4.6) with the number of differences no greater than 10 dB at 99% (95% CI 99-100). The results are consistent with other works (SD 4.97 [34], number of differences 97% [6]) using the method of determining the lowest audible sound through self-adjustment of the intensity of the test signal. Moreover, these values do not differ much from the test-retest examinations obtained for automated 10 dB down and 5 dB up bracketing method (SD 6.05 dB and 5.00 dB [34], the number of differences 98% [13]). This confirms the comparable accuracy of both methods [9], at least in the cases of mobile device owners who are familiar with the use of touch screens. Additionally, the results of test-retest examinations confirm the efficiency of the method, especially in self-monitoring of hearing.

Tests were audiologist-assisted and conducted in a sound booth; therefore, the results are devoid of user mistakes such as omitting the frequency by accidental button pressing or switching the sides of headphones as well as performing the examination in noise. When tested in uncontrolled conditions, care should be taken to minimize the risk of such mistakes, for example, by monitoring the test duration at single frequency and tracking the background noise using a built-in microphone.

The factors limiting the accessibility of the test in relation to the number of Android-based mobile devices are bundled headphones and calibration coefficients. Not all the devices are offered with bundled headphones. Moreover, calibration coefficients for less popular devices may not be determined if the required number of calibrations is not achieved.

During tests on mobile devices, temporary result was continuously displayed on the screen. Therefore, during the retest, subjects could be biased by previous results, consequently reducing the test-retest difference.

Mean difference in the hearing threshold determined through pure-tone audiometry and on mobile devices differed significantly from 0 dB and reached 7.0 dB (95% CI 5.5-8.5) at 8 kHz. The reason for such discrepancies may be related to the literature-based median values of the hearing threshold of normal-hearing subjects, which were adopted to determine 0 dB HL. The mean hearing threshold estimated on the basis of literature data for people aged between 18 and 35 years displays considerable differences (Figure 5) [38,40-44]. It may be related to differing definitions of the normal-hearing person, standards of the measurement method used, or other factors such as genetic composition or occupational distribution of the population. Therefore, the determination of the values for adoption is complex. This research applies values presented in study [38]. In the future, however, it seems rational to correct them by taking into consideration the identified discrepancies (Table 5).

The method of hearing self-test carried out on mobile devices with bundled headphones calibrated by model-specific coefficients determined in relation to the hearing thresholds of normal-hearing persons demonstrates high compatibility with pure-tone audiometry, which confirms its potential application in hearing monitoring, screening tests, or epidemiological examinations on a large scale. However, one must acknowledge its limitations resulting mainly from the calibration method. Further evaluation of the efficiency of the method, especially in particular applications, is justified and required.