Toward a better understanding of nonoccupational sound exposures and associated health impacts: Methods of the Apple Hearing Study^a)

Noise is a ubiquitous environmental exposure, and the most well-understood health impact of noise—noise-induced hearing loss (NIHL)—is among the most common occupational illnesses (Sataloff and Sataloff, 1996). Individuals with NIHL may experience a wide variety of social, health, and occupational impacts (Goines and Hagler, 2007). These can include communication and relationship difficulties, social withdrawal and isolation, reduced quality of life, depression and mental health issues, reduced fitness-for-duty and loss of employment, and lifelong learning deficits (Olusanya et al., 2014; Hammer et al., 2014). While often considered to be more of a nuisance than a disease, NIHL is, in fact, a debilitating and costly (Neitzel et al., 2017; Carroll et al., 2017)—but preventable—condition that affects tens of millions of people globally (Nelson et al., 2005) and has the highest burden in low-income countries (Zhou et al., 2021).

Historically, NIHL was a disease that primarily affected adult workers with high exposure to occupational noise (Themann and Masterson, 2019). However, in high income countries, industrial and employment trends have resulted in reduced occupational noise exposures in traditionally noisy workplaces (e.g., manufacturing; Themann and Masterson, 2019; Sayler et al., 2019), whereas participation in nonoccupational exposures during recreational activities with high sound levels (e.g., listening to music and other audio content) has increased (Verma et al., 2002; Sliwinska-Kowalska and Davis, 2012; Maassen et al., 2001; Jokitulppo et al., 2006). This has shifted the risk of NIHL from occupational sources of exposure toward nonoccupational activities to the point where the vast majority (approximately 70%–90%) of urban and suburban residents in the United States (U.S.) and other countries exceed the recommended community sound exposure limit of 70 dBA L_EQ(24) by the World Health Organization (WHO) from nonoccupational activities alone, including unremarkable activities such as riding public transit and bicycling (Flamme et al., 2012; Orlando et al., 1994; Zheng et al., 1996; Díaz and Pedrero, 2006; Neitzel et al., 2004; Neitzel et al., 2014; Kraus et al., 2013; Neitzel et al., 2012). Amplified music and music experienced through personal listening devices (e.g., music players and smartphones) are sources of sound exposure that have received great attention in recent years; several studies have indicated that a major source of sound exposure among urban adults was music (Spira-Cohen et al.,2017; Jokitulppo and Björk, 2003; Neitzel et al., 2012). Recent peer-reviewed studies offer conflicting evidence of the risk of NIHL from nonoccupational sound and music (Williams et al., 2015; Neitzel and Fligor, 2019; Degeest et al., 2017), as well as on trends in NIHL risk, with some studies suggesting that NIHL is stable or increasing over time (Rabinowitz et al., 2006; Tsimpida et al., 2020; GBD 2016 Disease and Injury Incidence and Prevalence Collaborators, 2017) and others indicating that the hearing ability of the population has improved over time (Lee et al., 2004; Hoffman et al., 2017; Paulsen et al., 2021; Zhan et al., 2010).

In 1999, the WHO published guidelines for community sound intended to reduce the risk of NIHL (WHO, 1999). In these guidelines, the recommended exposure limit, intended to eliminate the risk of any NIHL from environmental sound in virtually the entire exposed population, was a 24-h equivalent continuous average (L_AEQ) sound pressure level of 70 dBA (i.e., a 70 dBA L_AEQ(24h)). Recently, these guidelines have been supplemented by additional guidance intended to reduce the risk of hearing loss from listening to music. The new WHO music-specific recommendation is that an L_AEQ(24) exposure of 75 dBA, equivalent to an 8-h average exposure (L_EX) of 80 dBA, is sufficiently protective against NIHL for the vast majority of listeners (WHO-ITU, 2019; Neitzel and Fligor, 2019). Note that this level is more protective than the L_EX exposure limit of 85 dBA used to protect workers' hearing in virtually every country in the world with the exception of countries in the European Union, which use a lower exposure action value of 80 dBA L_EX (European Parliament and Council, 2003).

While the impacts of NIHL are extensive and widespread and exposure limits have been recommended, information on music and audio content exposure levels and listening patterns, as well as the associated risk of hearing loss, is sparse. Real-world data have historically been difficult to collect due to the logistical and technical challenges in accurate monitoring of the level, duration, and frequency of exposure to music. This has resulted in a reliance on assessments of short-term average or maximum sound levels experienced during listening and survey-based reports of listening durations and frequencies (Portnuff, 2016; Portnuff et al., 2013). Each of these approaches can introduce substantial error into the estimates of music and audio content exposures. Additionally, exposure assessments have typically focused either on music or other types of exposure (e.g., from occupational and nonoccupational activities) and only rarely integrated exposures across these various activities to estimate the total exposure (Neitzel et al., 2004; Neitzel et al., 2012). All of these issues contribute to the current lack of clarity regarding both exposures to and health outcomes potentially associated with nonoccupational sound and music.

Technologies available in the iPhone and Apple Watch (Apple Inc., Cupertino, CA), along with enhancements to iOS, iPhone operating system, have created the opportunity to evaluate exposures to music, as well as exposures to environmental sound, and easily assess hearing ability using conventional clinical tests in an at-home setting. The Apple Hearing Study is using a dosimetry-based approach to characterize the levels of sound at which participants listen to music and audio content (“headphone audio”). The iPhone is also used to collect app-based audiometry and speech in noise (SIN) test results from participants. The Apple Watch enables the measurement of environmental sound levels and heart rate and activity related measures, which affords the opportunity to explore the impacts of sound on cardiovascular health, a topic on which there is a rapidly growing body of evidence (Münzel et al., 2014; Münzel et al., 2018). By collecting this information from a target sample of 150 000 adult participants, this study will allow for estimates at multiple spatial scales, including the state- and national-levels, of exposure to music and audio content, environmental sound, and the association of these exposures with hearing and heart rate measures. This paper provides an introduction to the methods used in the study and examples of several preliminary results. Subsequent manuscripts will provide detailed results.

The Apple Hearing Study is a prospective cohort, two arm, experimental, and nonsignificant risk study sponsored by Apple Inc. (Cupertino, CA) and conducted by the University of Michigan (UM). The study participants are adults in the U.S. and Puerto Rico who have access to Apple technologies and provide informed consent for participation. The study connects and integrates five topical areas: personal characteristics, sound exposure, hearing ability, cardiovascular health, and general health. The study has four main objectives: (1) to understand typical headphone listening patterns and sound exposures and their relationships to current and future hearing health; (2) for participants who are Apple Watch users, to understand typical environmental sound exposures and their relationship to current and future hearing health; (3) to study how participants interact with data about their own headphone listening patterns and environmental sound exposures; and (4) for participants who are Apple Watch users, to understand the relationship between environmental sound exposures and cardiovascular health.

The study procedures described below have been reviewed and approved by the Institutional Review Board (IRB) for the study, Advarra, Inc. (Pro00037864, Columbia, MD). Participation occurs virtually through the Apple Research app,¹ which can be downloaded at no cost.² The data collection for the study begins after a participant provides informed consent via the research app. The participants have not received any financial incentives for enrollment, participation, or completion of the study. All of the analyses described in this paper and all data that have and will be collected and analyzed as part of the study, are provided by participants who have provided informed consent to participate in the study and share their data.

Demographic data, such as age, race, gender, and home zip code, are encrypted and stored on-device (iPhone, Apple Inc., Cupertino, CA) and shared with the Apple Hearing Study only after a participant provides informed consent. Any identifying data in the research app, such as name, date of birth, phone number, or email address, are stored separately from other study data in a repository with appropriate security controls and generally accessible only to the study principal investigator (PI) and team at UM with further access only permitted based on legal and regulatory obligations and similar limited criteria set forth in the study protocol and informed consent form.

The criteria for participation in the study are (1) possession of an iPhone running an iOS version capable of supporting the research app; (2) age ≥18 years of age (≥19 years of age in Alabama and Nebraska and ≥21 years of age in Puerto Rico per the legal age of consent laws in those locations) at the time of eligibility screening, ascertained via the self-reported date of birth; (3) live in the U.S. and Puerto Rico at the time of the eligibility screening, ascertained via self-report; and (4) proficiency in written and spoken English, defined via self-report. While not a requirement for eligibility, participants using a compatible Apple Watch may contribute activity, heart rate, and environmental sound level data. There are no additional exclusion criteria.

Potential participants may become aware of the study through multiple venues, including marketing, registration online,³ registration with a UM-based recruitment platform (),⁴ and the study website ().⁵ To assess study eligibility, the research app automatically detects the iPhone iOS version and determines age and country of residence from the user entered data. The informed consent form is reviewed, signed, and completed within the research app. Once informed consent is obtained, a participant is enrolled in the study. A downloadable copy of the completed consent form is accessible within the research app. Participants are also asked to provide explicit and voluntary consent for the passive collection of select health app data.

Following enrollment, the participants are randomly assigned into one of two user groups, which feature common study experiences (“basic”) as well as certain unique experiences (“advanced”). The participants in the basic group have a user interface (UI) in the research app that provides the ability to review exposure level data for the headphone audio levels and environmental sound levels in the health app. The participants in the advanced group have a UI that includes notifications prompting them to review their data in the health app. Specifically, participants in the advanced group will receive a notification if they experience a high weekly headphone audio (via iPhone) or environmental sound (via Apple Watch) exposure level that meets or exceeds a 7-day L_EQ >80 dBA for >40 h or have an acute exposure >97 dBA L_EQ for >30 min from headphone audio or environmental sound. This L_EQ was adopted based on WHO recommendations (WHO-ITU, 2019; Neitzel and Fligor, 2019). The notification will prompt the participant to review this exposure level data in their health app and answer questions related to their exposure. Survey questions will ask the user to confirm that they reviewed their data and then focus on the perceived accuracy of the measurement, whether or not they used any hearing protection (e.g., hearing protection worn at a rock concert), and any intended listening behavior change resulted from their data review. The advanced group participants are also prompted to complete a short survey and an abbreviated hearing test, both triggered by acute exposures to loud sound; these triggered tests are described further in Sec. II E 3.

Prior to the launch of the study, it was anticipated that a minority of the participants would routinely exceed our exposure levels for notifications and triggered hearing tests and, as such, favored randomization into the advanced user interface group (40/60 distribution of basic/advanced UI groups) to increase the likelihood that a sufficient number of participants will be provided with notifications and asked to complete the associated hearing tests. In an effort to reduce participant burden, the exposure notification is programmed to occur no more than approximately twice per month.

Once consent has been provided by a participant and they are enrolled in the study, collection of the data types covered in the informed consent begins. The primary data source for the study is the iPhone, which runs the research and health apps (Fig. 1). The research app is used to review and collect participant responses to surveys questions and tasks such as hearing tests. The Apple Watch is used to measure the ambient sound pressure level, and the iPhone uses these measurements in combination with the duration of exposure in a participant's environment to compute L_EX. Additional data about non-Apple hardware, such as headphones [e.g., wired vs Bluetooth connection (Kirkland, WA) and calibration information], is collected with participant consent and where possible to allow for the control for such variables during analysis.

The health app securely stores data on the device and select health app data, such as activity and heart related data, may be shared with permission to the research app. At enrollment and all times during the study, participants are able to review and select or modify which of the optional health app data types they wish to share with the study.

During participation, the research app collects and shares study data with Apple's secure cloud platform. The data are constantly and asynchronously uploaded, making the data dynamic in nature with large volumes of information continuously flowing in. With delays in connectivity (such as loss of WiFi), the data may be delayed in arriving to the secure cloud platform and backfill upon arrival, making even retrospective data dynamic and further complicating the data analysis.

The screenshots in Fig. 2 show examples of some of the interactions that participants have with the study through the research app. The participants receive notifications (1a) when a study activity (e.g., 1b-1d) is being requested of them. Participants, over the course of their participation, may be provided with updates on study progress and other related information (1e).

The data collected for this study address five topical areas (Table I): personal characteristics, sound exposure, hearing ability, cardiovascular health, and general health, including an assessment of COVID-19 due to emerging evidence that infection with the SARS-CoV-2 virus may impact hearing ability.

The data collection for each of the five topical areas occurs for each participant over the entire duration of the study via iPhone, Apple Watch, or survey and at temporal resolutions ranging from several seconds to annually. The research activities are distributed to minimize the burden on participants at any given point of time in the study, and participants may opt not to complete any tasks. Figure 3 shows the approximate time line of the study activities for the participants for all five topical areas and highlights the additional activities completed by the participants in the advanced arm.

Participants complete surveys at various frequencies, including at baseline and approximately every two months, for the length of the study. Each survey is presented as a task for the participant via the research app, and app-based notifications are employed to notify and remind participants about their tasks. Surveys are used to collect contact information, as well as information about demographics (e.g., date of birth, race/ethnicity, gender identity, sexual orientation, education level, and geographic information), socioeconomic status, typical headphone use, perceived environmental sound and headphone audio exposures, and listening behaviors. Survey durations are approximately 15 min at baseline and twice annually and 5–10 min every two months for a total of approximately 60–75 min annually. The participants have approximately 30 days to complete the scheduled surveys from the date of their initial notification in the research app. The participants can decline to answer any question on any survey.

Headphone audio levels and environmental sound levels are calculated on participants' devices and shared (with permission) with the research app. While the primary sources of data are iOS and the noise app on watchOS (Apple Watch operating system), the headphone audio levels and environmental sound levels can also be shared from third party applications. Headphone audio level measurements are available for all headphone models and interface types (i.e., 3.5 mm analog jack, lightning connector, and Bluetooth) during all iOS and watchOS media playback (music, podcast, games, movies, etc.), excluding telephony (phone calls, video conferencing, etc.) and streaming to hearing aids. Headphone audio levels are computed on the source device (iPhone or Apple Watch) using the methods standardized in EN 50332–3 (CENELEC, 2017) and ITU H.870 (ITU-T, 2018). Levels less than 80 dBA are aggregated to 30 s equivalent continuous average (L_AEQ) levels; levels above 80 dBA are aggregated into 5 s L_AEQ levels. Participants who share their headphone audio levels also share their headphone device model information, along with the category of content (e.g., music, podcasts, games, etc.) for some applications. Headphone audio level data can be viewed by participants as historical time-series charts and statistics with participant-selectable degrees of temporal resolution in the health app. If the L_EX exposure level is above the WHO recommendation of 80 dBA over 40 h or 97 dBA for 30 min, a notification is generated and can be viewed by the participant in the health app.

Environmental sound levels are measured using the noise app, available on Apple Watch Series 4 and later, and can be shared with the study by participants if they choose to do so. Environmental sound levels can be viewed in real-time by participants using the noise app on their Apple Watch; exposure summaries are available in the health app on the iPhone. Sound level measurements are automatically suspended while the Apple Watch speaker is being used, e.g., while taking a phone call or interacting with Siri, and when the water-lock feature on the Apple Watch is manually initiated (e.g., when swimming). Similar to headphone audio levels, the environmental sound level trends and statistics can be viewed in the health app. The noise app includes a notification feature with a default threshold of 90 dBA. When this threshold is exceeded, a notification to the participant is triggered.

The participants' hearing abilities are repeatedly assessed in multiple ways, including pure tone audiometry (PTA), which provides a frequency-specific measure of hearing sensitivity in decibels of hearing level (dB HL); SIN, which provides a measure of functional word recognition in background sound summarized as a speech recognition threshold (SRT); and a tinnitus matching activity that allows for profiling of the nature of the tinnitus experienced by the participants. The app-based notifications are used to notify and remind the participants about these tests. The participants have the ability to decline to complete any given hearing related test.

The PTA threshold and SIN tests are delivered shortly after enrollment and approximately twice annually afterward. The PTA test and SIN test take approximately 10 min each. The thresholds are measured at six frequencies in each ear (0.25, 0.5, 1, 2, 4, and 8 kHz). For calibration reasons, the PTA and tinnitus tasks require the use of supported Apple listening devices such as EarPods, AirPods, AirPods Pro, and AirPods Max (Apple Inc., Cupertino, CA). The SIN task can be completed with any headphones compatible with an iPhone. The tinnitus matching activity is completed by participants only when they are experiencing an episode of tinnitus and requires approximately 10–15 min to complete.

The participants in the advanced group are additionally asked to complete a brief survey and optional abbreviated PTA task (1 and 4 kHz in each ear) within 24 h after an acute loud exposure (L_EQ > 97 dBA for >30 min) from headphone audio or environmental sound. This L_EQ level is energetically equivalent to the 85 dBA 8-h L_EX used for protection against NIHL in workers in nearly every country in the world (Suter, 2003) and has been shown to be sufficiently high to produce a reliable, measurable temporary threshold shift (TTS) in human subjects, which completely resolves after time in a quiet environment (Strasser et al., 2008; Strasser et al., 1999; Irle et al., 1998). The participants in the advanced group are also prompted to review their environmental sound level data when weekly environmental sound or weekly headphone audio exposures exceed an L_EQ equivalent of >80 dBA for >40 h. The survey questions focus on the activity associated with the high exposure, perception of the exposure, and whether or not the participant was wearing hearing protection. If the participant's responses confirm that they experienced a loud exposure and if the participant was not wearing hearing protection, the participant is prompted to complete an abbreviated PTA task. The duration of this combined abbreviated PTA task and survey is approximately 5 min. To minimize the burden for participants, triggered hearing test events are not requested more than, roughly, twice per month.

The Apple Watch measures the heart rate and heart rate related measures, as well as activity information derived from the heart rate, such as heart rate variability, which can be used to assess the balance between the sympathetic and parasympathetic tones of the autonomic nervous system (Bent et al., 2020). As such, it may be used as a marker for cardiac status and a signal related to the physical and psychologic stresses on an individual (Singh et al., 2018; Acharya et al., 2006), and previous studies have suggested a link to noise exposure (Kraus et al., 2013; Burns et al., 2016).

The health data relevant to the study are collected through surveys (baseline and then approximately bimonthly afterward) and, optionally, through information entered into the health app and explicitly approved by participants to be shared with the study. At baseline and approximately annually, the participants are asked about their health status, whether they have been diagnosed with a hearing loss by a medical provider, whether they use assistive devices such as hearing aids or cochlear implants, and their experience with tinnitus.

Approximately every two months, the participants receive two short surveys to collect data about changes in medical history related to their hearing health and perceived stress level using the validated Perceived Stress Scale-4 (PSS4; Warttig et al., 2013). Each survey is about 5 min long. The hearing health survey contains questions around infection by SARS-CoV-2, which may impact hearing (Mustafa, 2020; Karimi-Galougahi et al., 2020; Kilic et al., 2020; Degen et al., 2020; Almufarrij and Munro, 2021).

Our data analyses are structured around the following four study objectives:

1. to understand typical headphone listening patterns and sound exposures and their relationships to current and future hearing health;

2. for participants who are Apple Watch users, to understand typical environmental sound exposures and their relationship to current and future hearing health;

3. to study how the participants interact with data about their own headphone listening patterns and environmental sound exposures; and

4. for participants who are Apple Watch users, to understand the relationship between environmental sound exposures and cardiovascular health.

Descriptive statistics and exploratory analyses will be conducted on all variables considered for analysis. Given the complex, multimodal, and longitudinal nature of our data, different objectives will require different statistical analysis techniques, and where multiple techniques are possible, we will explore and compare the results from those techniques.

For example, when examining hearing and cardiovascular health outcomes over time mixed-effects, multivariable regression models may be used. These models can test for between-subject and within-subject effects. The potential within-subject variables include changes in an individual's environmental sound exposure, changes in an individual's headphone sound exposure, occurrence of loud exposure notification alerts, and before/after a potential positive COVID test result. Potential between-subject variables include age, gender, ethnicity, self-reported hearing difficulty, self-reported hearing loss diagnosed, hearing loss determined from the baseline PTA, and study arm (basic or advanced). The interactions will be investigated, along with the random slopes for the within-subject effects.

In addition to the conventional epidemiological modeling approaches, we will also explore the machine learning techniques such as clustering users based on the level, duration, and time of headphone and/or environment noise exposure. These clusters can help distinguish the types of sound exposure patterns that people experience in their everyday lives. Furthermore, the clusters can be used to explore how different sound exposure groups affect the prevalence of hearing loss. Advanced machine learning classification techniques will be investigated to see if we can accurately predict the users who may experience bad hearing or health outcomes. We will also use time-series models to assess the changes in noise exposure over time by geographic location, something we have seen with the COVID pandemic. Survival models will be considered for the time to event analysis such as exploring how sound exposure levels associate with time until hearing loss onset.

Here, we present a limited set of example results from our ongoing study. Preliminary results of our environmental sound and headphone audio level measurements from a total of 121 010 participants are listed in Table II. To date, we have measured almost 280 × 10⁶ h of environmental sound and slightly more than 20 × 10⁶ h of headphone audio exposures for a total of approximately 300 × 10⁶ h of sound exposure data. The participants have contributed approximately 16 × 10⁶ participant-days of environmental sound data and roughly 12 × 10⁶ participant-days of headphone audio data. The environmental sound levels have been higher, on average, than headphone audio (mean 68.0 vs 62.7 dBA, respectively) and shown less variability (standard deviations of 6.1 vs 13.8 dBA, respectively). Roughly one-third of the participant-days have exceeded the WHO recommended daily environmental sound exposure limit of 70 dBA L_EQ(24), whereas only approximately one-tenth of the participant-days were above the WHO recommended L_EX limit of 80 dBA for headphone audio exposures.

The number of different hearing assessments conducted to date, as well as the prevalence of self-reported hearing difficulties, is shown in Table III. Over 90 000 PTA and nearly 100 000 SIN tests have been completed. The number of tinnitus matching tests, while substantial, is much smaller owing to the fact that the test is only administered to participants who report that they are experiencing tinnitus at the time they are prompted to complete the test, and the tinnitus matching task was added to the study as an amendment after the study began. Likewise, the number of completed triggered PTA tests is small because of the triggering criteria, which are exceeded by only a small fraction of the participants. Roughly 10% of the participants report that they have been diagnosed with a hearing loss by a medical professional, whereas nearly 20% report that they experience some hearing difficulty (defined as a “fair” or “poor” self-reported hearing ability on the baseline survey).

A single day's data on environmental sound, headphone audio, heart rate, and heart rate variability from three example participants is presented in Fig. 4 and several things are apparent. First, different participants (e.g., participants 1 and 3) can have very different temporal patterns and levels of exposure for environmental sound and headphone audio, and within an individual (i.e., participant 2), the patterns can also be quite different. Second, there are gaps in the data available from all of the participants; these gaps result from, for example, periods where Apple Watch is not worn overnight, e.g., participants 1 and 3); periods where headphones are not used (e.g., participants 1, 2, and 3); and periods where heart rate measurements are not collected (e.g., participants 1 and 3). The heart rate variability data are collected at sparse but regular intervals across all of the participants. The data shown are illustrative of the many challenges involved in analyzing the individual-level data collected as part of the study; additional challenges are introduced when the data are then aggregated across individuals.

Figure 5 shows maps representing the average environmental sound over time across participants in California. For this area, four maps are shown, depicting four different time periods: (a) pre-COVID-19 period 1 (before March 11, 2020); (b) post-COVID-19 period 2 (March 11, 2020–Oct 1, 2020); (c) post-COVID-19 period 3 (Oct 1, 2020–April 1, 2021); and (d) post-COVID-19 period 4 (April 1, 2021–September 1, 2021). Although these maps represent only a snapshot of the geographical data available in the study, they highlight several factors. First, temporal trends associated with the COVID-19 pandemic are evident, a finding further supported by our published analysis of a larger dataset (Smith et al., 2020), and consideration and incorporation of these trends into our analyses are critical. Second, exposure levels and temporal trends can vary widely even across relatively small geographical areas. Third, the richness of the geographical data increase over time as additional participants provide consent to share their data and are enrolled into the study.

Researchers have previously measured personal exposures to environmental sound or measured or estimated headphone audio on samples ranging from hundreds to thousands of participants. However, the Apple Hearing Study may be the first study to measure both types of sound exposure simultaneously and longitudinally over a multiyear period on a large sample of participants and combine those measures with multiple types of repeated hearing assessments. This study demonstrates that this type of research is technically feasible, and it can be leveraged to explore unexpected events that occur during the study period, e.g., the COVID-19 pandemic (Smith et al., 2020). The study has compiled what may be the largest-ever datasets on human exposures to sound and research hearing tests. Our preliminary and interim results suggest that within the study participants, a sizeable fraction of measured days—about 33%—involve exposure to environmental sound levels that exceed the WHO recommended limit, and a smaller but still not trivial fraction—roughly 10%—involve exposure to headphone audio levels that are above the WHO recommendation. Consistent with previous research (Goman et al., 2020; Choi et al., 2019; Newman et al., 1997), these data indicate that only a fraction of the participants who perceive themselves to have hearing difficulty have been diagnosed with hearing loss through audiometric testing. These findings highlight the need for additional intervention efforts to reduce noise exposures and increase access to hearing healthcare services. We expect to make recommendations based on study findings in subsequent manuscripts published closer to the end of the study.

Previous studies of personal exposures to environmental and leisure sound have typically used conventional noise dosimeters (Neitzel et al., 2004; Neitzel et al., 2014; Flamme et al., 2012; Orlando et al., 1994; Kraus et al., 2015; Kraus et al., 2013; Díaz and Pedrero 2006; Zheng et al., 1996) or, in some cases, sound level meters (Orlando et al., 1994) or smart devices (Roberts et al., 2016). These approaches are broadly similar to the approach used in this study, in which the participants' Apple Watches act as dosimeters, albeit worn in a different manner than traditional dosimeters (e.g., on the wrist instead of on the shoulder near the ear as dosimeters are typically worn; ANSI, 2006). A number of studies (Flamme et al., 2012; Neitzel et al., 2012; Neitzel et al., 2004; Orlando et al., 1994; Zheng et al., 1996) have identified higher L_EQ(24) exposures—on the order of 6–12 dBA higher—compared to our mean of 68.0  $±$  $6.1$ dBA. One likely explanation for this difference is that the majority of previous studies have collected measurements from participants within a geographically constrained area, often a large urban metro area. By comparison, the participant population in this study is drawn from a national catchment area and includes participants in rural areas, which may have lower exposures than urban areas (Casey et al., 2017; Foraster et al., 2017), and inclusion of these participants likely lowers our mean exposure level. Alternatively, this study is limited to users of Apple products, which may not be representative of the entire U.S. population. However, other studies have estimated personal exposures using a combination of self-reported activities and durations and representative measurements, e.g., Neitzel et al. (2012) and Neitzel et al. (2004). The exposures objectively measured in study participants are likely more accurate and less biased than exposures estimated in that way because the measurements used in this study eliminate recall bias and potential social desirability bias (Delgado-Rodríguez and Llorca, 2004).

Studies of personal exposures to headphone audio have used substantially different methods from those employed here. Most of the previous studies have either estimated the maximum output levels from personal music players in the absence of any information about listening duration, measured short-term listening levels from personal music players in combination with self-reported listening durations, or estimated exposures using a combination of self-reported listening durations and representative listening levels (Portnuff 2016; Portnuff et al., 2013; Fligor and Cox, 2004; Levey et al., 2011; Fligor, 2009a; Fligor, 2009b). Studies that measure the maximum output levels provide no information on actual exposures and, instead, report the upper limit of possible exposure, which can easily exceed 100 or even 120 dBA (Fligor and Cox, 2004; Keppler et al., 2010; Kim, 2013; Shim et al., 2018; Breinbauer et al., 2012). These studies overestimate the exposure levels for the vast majority of listeners and are, therefore, useful only as a bounding exercise for the purposes of the health risk assessment. Studies that rely on the use of self-reported listening durations have the potential for recall bias and social desirability bias and, as a result, may either under- or overestimate the actual exposures. Only a small number of studies have attempted to use a dosimetry based approach to measure the listening levels and listening durations (Jiang et al., 2016) and eliminate the potential for recall and social desirability bias, thus, providing what is likely to be a more accurate exposure assessment. The results of this study—a mean headphone audio L_EX of 62.7  $±$ 4.7 dBA—are, not surprisingly, broadly consistent with this latter category of study; Jiang et al. (2016) found seven studies that estimated L_EX exposures from personal music players with a range of 61.6–87.2 dBA and a mean across these studies of 75 dBA. However, the estimates from the Apple Hearing Study represent a much longer monitoring period than any other study (years rather than days or weeks) and expand the understanding of long-term exposures to headphone audio in a way not previously possible.

Use of the research app platform has enabled the research team to conduct a large-scale longitudinal study that is readily accessible to adult iPhone users within the U.S. and Puerto Rico. The framework supports in-app and over-the-air study updates, which further enhance the capability of researchers to evaluate exposures to sound and the associated heath impacts. The framework used for the Apple Hearing Study serves as an example of a data collection mechanism that could potentially be adapted to improve clinical surveillance for hearing health. Additionally, and of critical importance, this framework can be used to provide objective long-term surveillance of environmental exposures such as noise—data that are very rarely available to clinicians. The collection and analysis of environmental exposures in conjunction with observed changes in physical and mental health could substantially enhance the understanding of the impacts of environmental exposures on human health. Although such data collection mechanisms have tremendous potential, prior to their adoption and implementation, there is a need for the measurements made with devices from different manufacturers to be validated using consistent methodologies to ensure that the devices yield comparable results. Prior to developing clinical surveillance programs, it will be critical to address relevant legal and privacy issues.

As with all studies, the Apple Hearing Study has a number of limitations. The first and primary limitation is that the exposures and outcomes observed in our sample of participants may not be representative of the entire U.S. population because voluntary participation is restricted to Apple users, who may differ in key demographic indicators (e.g., gender, race/ethnicity, income, etc.) from the U.S. population. However, nonrepresentative samples can still yield accurate estimates of population level outcomes with proper statistical adjustment. One approach is to use regression and poststratification. That is, estimate outcomes at the various demographic levels using regression and then aggregate these estimates in accordance with the U.S. population demographic composition.

The second limitation stems from the inclusion of non-Apple headphones for the purposes of headphone audio dosimetry; this decision, which expands the number of participants who can contribute data to the study, has implications for the accuracy of estimated headphone audio exposures for users of the non-Apple listening devices. If a calibration value (as described in EN50332–2; CENELEC, 2017) is available for the connected headphone device (such as all Apple, Beats, and select third party devices), this value is applied during computation of the headphone audio level. Uncalibrated headphones can result in less accurate results; in the U.S., where the EN50332–3 standard has been adopted, in most cases, the default sensitivity described in the standard would result in an overestimate of headphone audio levels. We will assess the potential impacts of this limitation through the use of sensitivity analyses comparing dosimetry results from users of Apple and non-Apple headsets to evaluate the possible presence of systematic differences between the two groups.

The third limitation is the fact that notifications provided to participants by the research app and any other app present on iPhones can be silenced by users. In the case of participants in the advanced group, silencing notifications—a decision which cannot be detected by the study team—will limit our ability to evaluate the impact of notifications on behaviors related to sound and hearing by biasing our results toward the null (Delgado-Rodríguez and Llorca, 2004).

The fourth limitation relates to the use of hearing protection by participants. Although some fraction of the participants will likely use hearing protection during exposures to environmental sound, we cannot account directly for the possible attenuation of sound that they receive during these exposures. This is due to the fact that we currently do not have a mechanism for measuring the specific times that hearing protectors are used, and there is no way to obtain information about the level of attenuation achieved by the individual participants. Instead, we will account for the use of hearing protection by controlling for self-reported frequency of hearing protection use with the participants' survey responses describing typical hearing protection behavior.

The fifth and final limitation relates to participant engagement. Unlike a traditional epidemiological study in which participants regularly interact with research staff and are likely to receive compensation for participation, this study is conducted entirely digitally, and participants do not receive any compensation. Additionally, resulting from the novel nature of the hearing tests conducted within the app, at the present time, results from the hearing tests are not being shared with the participants. This has the potential to reduce participation and may result in increased attrition over time. We are employing several approaches to address this limitation, including the use of regular study updates to participants to discuss study progress and findings, as well as the use of mixed-effects regression models that permit us to use all of the data from participants who may not complete all of the study activities.

Despite these limitations, we believe that the national scope of this study and our assessment of the different data types at the scale collected have never before been possible and are quite important. We also continue to explore potential additional measures that participants may choose to share with us, which could further enhance our understanding of the impacts of sound on health. Such measures could include data relevant to diabetes—a risk factor for hearing loss (Kakarlapudi et al., 2003; Teng et al., 2017), general health factors, and additional measures of activity, such as step count, stand time, and calories burned daily through exercise activities, and via basal metabolism. Significant benefits of app based studies, such as this one, are the ease and speed with which additional data collection measures can be shared provided that appropriate IRB approvals and participant consent are obtained.

Given the potentially substantial public health burden associated with sound exposure (Basner et al., 2014; Passchier-Vermeer and Passchier, 2000), research is necessary to improve the understanding of the impacts of sound on human health to develop interventions to address and reduce these impacts. The Apple Hearing Study is the first national-scale study to evaluate exposures to sound from headphone audio and environmental sound simultaneously and link those with prospectively collected data from perceived and objectively measured hearing tests and measures of cardiovascular performance. Our enrollment of a large and diverse cohort of participants facilitates the characterization of listening and exposure patterns and hearing health behaviors on a national sample, which will, in turn, accelerate our understanding of headphone audio, environmental sound, and total sound exposures and the impacts of these exposures on hearing and cardiovascular health in a way not previously possible. Future manuscripts will provide detailed results for each of the aspects of the study described here.

The authors wish to thank the participants of the Apple Hearing Study, whose support made this research possible. Funding for this study was provided by Apple Inc. Salaries for R.L.N., L.S., L.W., and G.G. were supported by study funding provided by Apple Inc. J.B., M.C., K.M., R.A., G.D., and B.V. are employees of Apple Inc.