The COVID-19 pandemic has caused disruptions to the way people go about their daily lives. From the changing nature of work and employment, economic factors, the isolation brought about by stay-at-home orders, and the uncertainty of ever-changing health advice and medical directives, it is anticipated that these stresses will lead to an increase in mental health service demand beyond the current capacity [1]. The adoption of digital health technologies can potentially alleviate this burden.

Wearable devices are electronic sensors that are designed to be placed onto, or near to, the skin to measure signals from the body. Such devices can include wrist-worn devices similar to a watch or wristband, which can pair wirelessly with a mobile phone. Such devices have become a popular behavioral intervention for monitoring physiological activity to promote a healthy lifestyle [2]. Early forms of health monitoring include pedometers that would track daily steps and derive basic energy expenditure (EE) [3]. The potential of wearable devices for the monitoring of health has become particularly attractive to health care innovators seeking to enable new models of telehealth. However, these devices monitor physiological signals or physiologically related proxies (such as physical activity) of the user rather than mental health. Such devices may take the form of fitness trackers, which are typically simpler, lower-cost, and fixed-function devices with limited capabilities. Such devices often cannot support third-party apps, have limited user interactivity, and focus on fitness monitoring as their primary goal. By contrast, smartwatches are usually higher-end devices with a richer mix of sensors and user interfaces and a flexible, extensible software architecture permitting third-party software access and extended features such as voice calling, media control, and messaging. As the market matures, there are some products that may blur the lines; however, it is the richer suite of sensors, user interfaces, and support for third-party apps and data access, which makes these devices attractive for mental health research and monitoring purposes.

Mental health can be defined as “a state of wellbeing in which the individual realizes his or her own abilities, can cope with the normal stresses of life, can work productively and fruitfully, and is able to make a contribution to his or her community” [4]. This state is intimately connected with physical health and forms an integral part of general or overall health [5]. A mediation study examined the effects of physical health on mental health and vice versa, finding significant direct and indirect effects and cross-effects [6]. Studies have also indicated the effectiveness of physical activity in improving anxiety and depressive symptoms [7]. The measurement of signals from wearable devices that allow for an understanding of physical activity may also allow mental health status to be inferred.

Apple Inc has emerged as an industry leader in health technology and wellness tracking devices [8]. The Apple Watch, first introduced in 2015, has retained the largest market share since its introduction and has continually advanced the capabilities of smartwatches [9]. These devices are primarily intended as wellness tools, garnering additional personal health monitoring for the wearer, typically for physiological activities such as heart rate (HR), HR variability (HRV), respiration rate, and physiologically related measures such as EE and fall detection. Some capabilities of these devices, such as the electrocardiogram (ECG) function, including a supporting app, have received Food and Drug Administration (FDA) clearance [10], whereas other aspects of their sensors and app capabilities have not yet been independently validated or received regulatory clearances. Monitoring of stress using these devices has been less studied but appears to be a promising avenue for application, particularly in the mental health sphere.

As digital health provides a novel model of care through the use of intelligent data, computing, and telecommunications, it holds promise for meeting the challenges of increased mental health demands. It can also enable precision medicine, which provides treatments bespoke to the patient’s needs [11]. There is interest in digital health across a number of industry sectors, including health care providers, insurers, and businesses [12-17], that may desire access to information on personal health through wearable devices such as the Apple Watch.

Wider adoption of devices for mental health monitoring is, in part, hampered by a lack of clarity regarding the devices’ capabilities, the accuracy and validity of the data that are collected, and their applicability to mental health monitoring and diagnosis [18-20]. This research aimed to fill this knowledge gap by examining the embedded sensor capabilities within the Apple Watch range, the physiological and physiologically related metrics recorded and made available for analysis, the validation status of these metrics within the literature, the connections (where they exist) between relevant health conditions associated with each metric, and implications for treatment. This analysis was performed both in a “top-down” approach focusing on reviewing published literature regarding the Apple Watch and a “bottom-up” approach focusing on the hardware and software capabilities of the Apple Watch to identify both currently available features and potential features that could be operationalized through the creation of customized apps using the Apple WatchKit, CareKit, and ResearchKit frameworks.

Various types of published studies and editorials were included. The types of studies were extended to some unpublished (gray) literature that was evaluated and reviewed for its suitability to close gaps in knowledge. Other gray literature sources included developer documentation for the HealthKit application programming interface for storing and managing data collected on the devices. Several opinion pieces were reviewed contextually to further provide a professionally informed perspective or illustrate further points of consideration. This literature review was structured to include the literature concerning the monitoring of physical conditions that may present with psychological stressors and the implementation of the Apple Watch for such monitoring.

The electronic databases selected for this literature review were PubMed, Scopus, and Google Scholar. A list of secondary keywords (Textbox 1) was developed with an emphasis on “Apple Watch” and truncated keywords combined using Boolean operators. Publication dates were restricted to 2015 onward, coinciding with the announcement of the first Apple Watch. Other recent literature that included wearable devices and novel developments to monitor or detect depression, anxiety, or stress was also included in the search process, in addition to reviews and systematic reviews.

Published literature was included based on its use of the Apple Watch for either physiological data validation or psychology or mental health studies. Areas of interest for applications in monitoring physiological stress and mental health included HR monitoring, sleep tracking, respiration monitoring, and EE. Other inclusion criteria included studies performed on the suitability of wearable devices for monitoring physiological stress and their impacts on mental health. Only publications in English were included in the review. Screening was performed by a primary researcher and reviewed by other authors. Duplicate studies were removed.

Data extraction was performed using a spreadsheet that synthesized the findings and grouped the studies. Data management was achieved using EndNote (Clarivate Analytics) as the bibliographic management software. Where studies did not specify the Apple Watch Series, it was inferred by comparing the date of publication with the Apple Watch Series release dates.

The literature search strategy resulted in 5583 paper titles being identified. Screening of titles and abstracts resulted in 2.06% (115/5583) of papers being selected and reviewed in full. Of these 115 papers, 19 (16.5%) were identified as related to Apple Watch validation or comparison studies, which are summarized in Table 1.

Several published reviews focusing on wearable devices, smartwatches, and associated physiological measurements were also identified as part of this search (Textbox 2). These reviews provide a contextual background in a number of areas; however, this review was focused on Apple Watch–specific research.

The Apple Watch is a sensor-rich, well-constructed, and connected device. It uses a large range of apps and has significant potential for applications in mental health (Figure 1).

Apple Watch sensors typically include a 3-axis accelerometer, a gyroscope and magnetometer, optical PPG-based HR sensors, altimeters, ambient light sensors, temperature sensors, ECG, and capacitive (touch) sensors [3]. Across each iteration of the Apple Watch, sensor inclusions and capabilities have increased, matched with software updates aimed at increasing the overall accuracy of the collected data. Figure 2 presents a timeline of the development of the Apple Watch, summarizing the changes in sensor inclusions over time. The latest version of watchOS (version 8.0.0) is supported by Series 3 to Series 7 models. The models currently available for purchase include Series 3, SE, and Series 7. The Apple Watch Series 3 does not include fall detection as the 6-axis inertial measurement unit containing the gyroscope and accelerometer was modified for later-generation Apple Watches [49].

One of the primary sensors in all generations of the Apple Watch is the optical HR sensor, which is used to collect HR data. The scientific principle that these sensors rely on is PPG to detect the amount of blood that is flowing through the wearer’s wrist at any given moment. The reflection of green and infrared light-emitting diode (LED) light is measured with photodiodes that allow for the determination of HR as a periodic variation in the signal. By flashing hundreds of times per second, the optical HR sensor can measure HR across a range of 30 to 210 bpm [48]. Infrared light is used to measure HR in the background and for HR notification systems as infrared light can penetrate the skin better; however, this makes it more susceptible to motion artifacts. Green LEDs are used for workouts and to calculate HRV [48]. The Apple Watch will automatically detect when there is an increase or decrease in motion from the inertial measurement unit and change the LED light color accordingly. Variations have been made in the design and layout of the LED and photodiode arrays with each iteration of the Apple Watch to improve accuracy [79]. These optical HR sensors are used by the Irregular Rhythm Notification Feature (IRNF), which can assist in the detection of AF [80,81]. A red LED was added in Series 6, enabling blood oxygen saturation calculation by comparing the ratio of infrared light and red light. Reflectance oximetry is noted as being less accurate than clinically used transmittance oximetry [79], and we did not identify any literature validating the accuracy of the Apple Watch blood oximetry.

In addition to the optical HR sensors, from Series 4 onward (not including the SE model), an ECG electrode was integrated into the back face of the watch and the digital crown. When engaged by the user’s finger, a closed circuit is created to measure the electrical potential across the heart, similar to a 1-lead ECG. An ECG measurement takes 30 seconds. The ECG sensor is exclusively used with the ECG classifier to categorize heart events as AF, normal sinus rhythm, high or low HR, or inconclusive [48,82]. Version 2 of the ECG app also includes additional classifications of AF, high HR, and poor recording. For the earlier Apple Watch Series, a third-party accessory (Kardia Band) could be used to provide a 1-lead ECG that achieved a sensitivity of 93% and a specificity of 84% when compared with a standard tachograph [83].

A clinical study compared the ECG app developed by Apple Inc with an FDA-cleared clinical ECG device (GE Healthcare CardioSoft ECG device), with recordings verified by 3 independent board-certified American cardiologists in each of the ECG app categories [50]. The app received clearance by the FDA as a De Novo class II device as it was proven to perform similarly to the comparator device [82]. The same approval was also given to the optical HR sensor IRNF software in 2018 [80]. Some limitations exist in the use of both apps, which are not intended to be used on persons aged <22 years. Depending on the country in which the Apple Watch user resides, they may not have access to the software and, as such, may not be able to use these notification features. In Australia, both the ECG app (version 2.0) and the IRNF software were approved by the Therapeutic Goods Administration of the Australian Government in early 2021 [84,85].

Health data collected from the Apple Watch could complement smartphone data collection and self-reported measures to provide additional context and assist in determining and tracking a user’s affective and emotional health. Advancements in the sensing technologies available within wearable devices and enhanced user interfaces have removed some of the previously limiting factors of monitoring mental health using wearable technology. However, the current general consensus for using wearable device sensors is that they should be paired with traditional screening and diagnostic tools and not be considered as a replacement [33,83]. Wearable devices can assist in clinical diagnosis and application of therapy if the findings are consistent with the patient’s complaints or concerns or if the patient is unsure of their physiological level of stress [86]. Indeed, a systematic review of digital health interventions for depression and anxiety in young people has shown that such interventions may only be of clinical significance when their use is highly supervised [87].

An article compared several wearable devices, including the Apple Watch (series unspecified), and their applications for “advancing resilience and mental health of employees that experience high workload” [21]. The study noted that an increase in psychological disabilities in the modern workplace requires the development of new and emerging technologies to measure and monitor physical or mental status. As such, these tools are being implemented to assist in the diagnosis and treatment of stress within professional workplaces and in a performance review. A potential issue with workplace inclusion for monitoring mental health and wellness is regulations and access to technology.

The use of the Apple Watch as a source of data may address problems with patient recall bias as most assessments are reliant on patient self-reporting. This could reduce the reliance on patient memory and continued questioning to ensure consistency. In addition, it could be a relatively low-cost method for better long-term tracking of symptoms and trends in the data [69]. The use of these data permits the construction of an ecological context that could empower a more cohesive diagnosis and application of therapy or assist in refining threshold values used in algorithms toward a validated measure.

Although there are potentially great benefits of wearable devices in improving mental health, there are some potential drawbacks, including concerns about abandonment rates. Approximately 11% (2/19) of individual studies commented on the long-term use of electronic wearables, one noting that 20% of consumers stop using their wearables after 3 months, and <50% continue to use them after 1.5 years [83,88]. This is compounded by the need to provide enough contextual information regarding the data collected, which requires some level of active user participation. For a clinical diagnosis of a mental disorder, clinicians must make a decision based on weighing the mix of potentially contradictory evidence according to their expert judgment, which could require symptom tracking over a period of months to come to a clear conclusion. Symptom tracking for the validation of several mental health diagnoses against the Diagnostic and Statistical Manual of Mental Disorders can require the presence of symptoms over a period of weeks, months, or even years for mood disorders, anxiety disorders, and schizophrenia [59].

A validation study was completed on the effectiveness of using the Apple Watch to collect passive sensor data with “ecological momentary assessments” from a watch-based questionnaire app recording patient feedback to assess and monitor substance abuse in young adults [89]. The response from participants on the perceived burden of engaging with the app was low; however, it was noted that the relative ease of completing the surveys was easier on an iPhone than on the Apple Watch. Burdensome interactions within wearable devices could reduce uptake and willingness to use technology for mental health monitoring. However, the benefits of engaging users through health notifications and alerts can assist in seeking medical assistance or outpatient care [29]. A longitudinal observational study using cognitive assessment delivered through the Apple Watch in patients with major depressive disorder noted excellent adherence for both mood and cognitive tests (95% and 96%, respectively) over the 6-week study period, and it was not influenced by symptom severity or cognitive function at the study onset and did not deteriorate over time, supporting the feasibility of this approach [90].

The availability of sensor and health data collected from the Apple Watch and patient input relies on the application programming interface frameworks available from Apple for iOS and watchOS. The main frameworks are HealthKit, ResearchKit, CareKit, and SensorKit [91-93]. HealthKit is the most comprehensive as it implements a central repository for all collected health data related to the user. Developers can write apps that request permission to access the HealthKit data store to record, access, and share user health data. SensorKit is used in the event that raw access to sensors is required. ResearchKit may be used to build research study apps, whereas the CareKit framework is suited to the development of ongoing care capabilities. Together, these frameworks allow for the implementation of apps that can collect raw data and store and analyze collected data (including passively collected data) and provide tracking feedback to the end user as well as the clinician.

Within the HealthKit framework, a range of rigid data classes and methods can be used to collect, store, and retrieve data. In this way, virtually all types of health-related data can be stored as numerical data (eg, HR) and categorical data objects (eg, blood type). It categorizes the data systematically, reducing duplication and allowing for straightforward statistical data analysis. HealthKit supports units of measurement within each of these categories such as length, mass, volume, and energy. Conversion between measurement systems is automatically supported within the framework but can also be explicitly defined. Developers cannot create custom data types or units but can use the metadata fields to store additional data.

Most of the identified studies investigating wearable devices collected the activity level (steps and caloric expenditure), HR, and sleep data without indicating how the data were collected from the device, the frequency of data recording, or which measures were extracted from HealthKit. We believe this to be important information to be provided by studies, especially those that develop a custom app, to ensure a comprehensive understanding of the data, allow for comparative analysis with other studies, and inform future developments.

Digital phenotyping approaches have been an active area of development enabled by the popularity of smartphones [94]. By collecting data from sensors in a smartphone on a moment-by-moment basis, it is hoped that information about the user’s behaviors can be inferred to personalize patient care [95]. Active and passive data collection techniques have been explored, including data such as location, activity, app use, phone use, Bluetooth signals, and voice samples [96]. Research has focused on correlating such data with reported and diagnosed conditions to determine the most valid signals for mental health applications; however, this is still considered to be in its infancy.

Early studies suggest that data surrounding activity and geolocation could serve as early signs of mania or depression [97]. Furthermore, the monitoring of movement and light data was able to detect and assess depression severity [98]. Research into schizophrenia shows that digital phenotyping approaches have merit in identifying relapse events [99], that collected accelerometer and GPS data have a good correlation with future patient survey scores [100], and that such an approach was tolerated by outpatients [101].

Issues surrounding noise, privacy preservation, missing data, and data quality have been acknowledged and pose challenges in data analysis as the sensors may not be able to provide a complete context [102]. However, such approaches still require considerations of clinical relevance, social equity, development of common data standards, and multidisciplinary collaboration [103,104]. This may include the need to improve digital health literacy through training programs tailored to the needs of the target population [105].

Although it may be theoretically possible to combine smartwatch data with those collected from a smartphone to improve data quality for digital phenotyping approaches, as a smartwatch is more likely to be worn on the body than to be left behind, such an approach may be incompatible with smartwatches, which are much more resource constrained in terms of computational power, storage, connectivity, and (most importantly) battery power. The continuous collection of sensor data on smartphones has been shown to have a significant impact on battery life, which is a factor against user acceptance [103]. The impact on smartwatches, which typically have smaller batteries and rely extensively on sleep power-saving techniques to achieve all-day battery life, is anticipated to be significant.

As a result, it seems most prudent to identify the relevant physiological and physiologically related signals that relate to mental health and build algorithms focusing on data from those metrics alone rather than taking a dragnet correlation approach as is traditionally used in digital phenotyping. Such an approach will also serve to address some of the concerns regarding privacy and user perceptions that such a system is fated to diagnose users with conditions simply based on overcollection of data and misunderstanding of cause and effect [106].

The issue of personal health information regulation is important for maintaining user trust and privacy. Regulations have usually lagged behind rapid technology development, with concerns about data ownership. As such, there is some suggestion that wearable technology be considered differently from consumer technology because of inherent personal health information concerns.

Consumer wellness devices are not considered medical devices and, thus, may not be as accurate or reliable for remote health monitoring. Establishing their accuracy would require independent verification or undergoing regulatory approval processes. Constraints surrounding medical device regulation are a source of concern as the long process can stifle innovation and the development of new technologies [107]. However, some features may be able to individually receive clearance from regulators (eg, the ECG app with the Apple Watch) [108]. The ECG app and IRNF are both classified as De Novo within the FDA regulations, which is a marketing pathway for novel devices of low to moderate risk where a predicate device does not exist. In this manner, the FDA creates a classification for the device, which can be used for future premarket approvals of equivalent devices to ensure that new and emerging novel technologies are not held back during classification.

In addition, most device manufacturers provide their own independent platforms, very similar to HealthKit for the Apple Watch, for users’ data storage. These platforms may be limited in terms of data access and sharing, forming a vendor lock-in that prevents users from being able to migrate their personal health information to other platforms and reducing the research value of the devices. There are concerns over the control larger companies may have over the health data of users; this can conflict with informed consent, which is integral to medical practice [69]. Passive data collection is less intrusive and time consuming for the wearer; however, it can capture a large amount of personal data that can be stored unknown to the user, even if they have authorized the data to be recorded. Typically, the average person is more relaxed with security implementation when using personal devices and may be unaware of the level of security that third-party apps provide [13]. Similar concerns surround wearable devices and their use in workplace wellness programs and health insurance provisions if there is ambiguity regarding how the data will be used and the potential for surveillance [13,14]. The ethos behind the Apple HealthKit framework’s rigid type structures and fine-grained authorization process is designed to ensure that only necessary data are collected or accessed [109,110].

The use of wearable technology for health care service provision is still in its infancy, and evidence to support its implementation is still being developed. Known concerns exist regarding passive data collection, data ownership, data use, user trust, and user attitudes toward wearable technologies, leading to potentially high abandonment rates [44,103].

Perhaps the best model for how the Apple Watch can be applied to mental health can be found in the insurance sector, where some insurance providers have embraced wearable technologies to promote healthier lifestyles. Incentive programs involving wearable devices have been used by numerous US health insurance providers, including United Health Care, Anthem, Humana, Health Care Service Corporation, Centene, CVS Health (Aetna), WellCare, Kaiser Permanente, GuideWell, and Molina [17]. AIA Insurance Australia has a specific program using the Apple Watch called the Vitality Apple Watch Benefit, which reduces the monthly loan repayment of the device through the achievement of weekly activity targets [16]. Loss-framed incentivized policies using the Apple Watch achieve a 34% increase in tracked activity days over 1 month in comparison with a standard gain-framed policy [12]. This offers a potential solution to individuals who may not have the financial flexibility to pay the full upfront cost of the Apple Watch device but can still have access to the benefits of the device as a wellness monitor for personal health. Another study investigated the “incentivize and persuade” health-tracking approach from both insurers and employers for enhancing business chain value. It was concluded that persuaded self-tracking, whereby service firms or employers encourage consumers and employees to collect and share data via self-tracking, is heavily influenced by service firm and individual determinants. Understanding consumer perceptions and consumer reactions within a conceptual framework should reflect values in use, privacy and security, and perceived fairness or justice as the technology itself may perpetuate inequalities [15]. Both studies noted the effects of physical activity on physical wellness, as well as mental health, but did not specifically note the impact on policy holders with severe mental illnesses. Investigation into mental health monitoring for insurance purposes could potentially create contention and the consensus that balancing privacy and confidentiality is critical for engendering trust with users and policy holders through transparency [111].

The Apple Watch has presented itself as a capable wearable device that is able to monitor several physiological parameters and track overall health and wellness. Its use within the mental health sphere is encouraging, particularly as more research emerges correlating changes in the emotional and physiological states of the body. Measures of HRV are key indicators of changes in both physical and emotional states. In combination with other sensors to monitor general activity, sleep, and more, health data can be aggregated with user-provided information to assist in the monitoring and even diagnosis of mental health disorders. Particular benefits may be derived through the avoidance of recall bias by providing a more objective, data-driven record of events in a passive manner. The lack of methodologically robust and replicated evidence of user benefits and a supportive health economic analysis, as well as concerns about storage, access, and security of personal health information, remain key factors that must be addressed to enable broader uptake for mental health applications.