Ethical Considerations in Using Sensors to Remotely Assess Pediatric Health Behaviors

Abstract

Sensors, including accelerometer-based and electronic adherence monitoring devices, have transformed health data collection. Sensors allow for unobtrusive, real-time sampling of health behaviors that relate to psychological health, including sleep, physical activity, and medication-taking. These technical strengths have captured scholarly attention, with far less discussion about the level of human touch involved in implementing sensors. Researchers face several subjective decision points when collecting health data via sensors, with these decisions posing ethical concerns for users and the public at large. Using examples from pediatric sleep, physical activity, and medication adherence research, we pose critical ethical questions, practical dilemmas, and guidance for implementing health-based sensors. We focus on youth given that they are often deemed the ideal population for digital health approaches but have unique technology-related vulnerabilities and preferences. Ethical considerations are organized according to Belmont principles of respect for persons (e.g., when sensor-based data are valued above the subjective lived experiences of youth and their families), beneficence (e.g., with sensor data management and sharing), and justice (e.g., with sensor access and acceptability among minoritized pediatric populations). Recommendations include the need to increase transparency about the extent of subjective decision making with sensor data management. Without greater attention to the human factors involved in sensor research, ethical risks could outweigh the scientific promise of sensors, thereby negating their potential role in improving child health and care.

Sensors, including accelerometer-based and electronic adherence monitoring devices, are important digital instruments for assessing health behaviors that are linked to psychological well-being, such as sleep, physical activity, and medication-taking. The use of sensors in psychological research has dramatically increased in the last decade, including in research involving children and adolescents. This recent increase in scholarship may be because of the proliferation in the commercial availability, popularity, and widespread adoption of wearable health sensors. As of 2019, 21% of adults in the United States reported that they regularly wear a smartwatch or wearable fitness tracker (Vogels, 2020). Between the years 2014 and 2020, the number of wearable fitness trackers shipped worldwide increased by approximately 1,444% (Laricchia, 2022). While ownership data are more limited among youth (Ridgers et al., 2016), children and adolescents are often described as the ideal population for digital health tools like sensors because they are digital experts who have grown up with and frequently use smartphones and other technologies (Rideout & Fox, 2018). As health sensors are increasingly available and integrated into the fabric of daily life, they offer many strengths for gathering “real-world” health data in “real time,” which can ultimately inform the delivery of personalized health and mental health interventions.

A primary appeal for sensors in pediatric health research is that these devices have the capability to gather a tremendous amount of health data passively, in either natural or laboratory environments, without a significant user burden. This passive data collection may be especially advantageous for certain children and adolescents who cannot provide self-report about their own behaviors due to their age or developmental level. Additionally, sensors have demonstrated strong reliability and validity compared to other measures. For example, electronic medication monitors capture more missed medication doses than self-reported adherence (Stirratt et al., 2015). Most accelerometer-based devices, with appropriate data processing, show strong sensitivity in detecting the sleep period compared to polysomnogram (Ancoli-Israel et al., 2015; Meltzer, Montgomery-Downs, et al., 2012) and in detecting physical activity levels compared to indirect calorimetry (Lynch et al., 2019). These measurement strengths are relevant for selecting clinically meaningful primary outcomes for health behavior change interventions, which can in turn promote downstream health and well-being.

Because of their capacity to collect daily data, sensors also provide novel opportunities for understanding and intervening in the temporal relationships between contextual states and health behaviors. For example, by combining accelerometers to measure physical activity and ecological momentary assessment to repeatedly measure affective and physical feeling states, Dunton et al. (2014) determined that physical feeling states (more energy, less fatigue) predicted greater physical activity in children, which in turn predicted subsequent positive physical and affective states. In another study, accelerometer-derived and self-reported nightly sleep disturbances predicted subsequent increases in youths’ suicidal ideation (Bernert et al., 2017). Just-in-time adaptive interventions use incoming contextual data from sensors and/or ecological momentary assessment (i.e., repeated surveys) to make individual-level decisions about delivering the most effective intervention, at the right time, while minimizing unnecessary or burdensome intervention components (Nahum-Shani et al., 2018).

While sensors offer many strengths, their use introduces numerous ethical challenges to the researchers and clinicians who employ them, including with respect to respect for persons (e.g., when sensor-based data are valued above the subjective lived experiences of youth and their families), beneficence (e.g., with data management and sharing), and justice (e.g., with sensor access and acceptability among minoritized pediatric populations). Compared to adults, these ethical considerations are amplified for children and adolescents, who have unique technology-related vulnerabilities and preferences, and other caregivers involved in their technology access and use, health behaviors, and research consent procedures (Psihogios et al., 2022). Still, there has been limited discussion of the ethics of health sensors in general or specifically in pediatrics.

Thus, using examples from pediatric health behavior research, the aim of this article is to critically discuss ethical challenges and recommendations for implementing health-based, human-enabled sensors in pediatric populations. This work builds upon the Digital Health Checklist for Researchers (Nebeker et al., 2020), which delineates broad ethical challenges in research with any digital health tool (e.g., sensors, apps), for any participant population, and related guidance for study procedures (e.g., informed consent). Here, we concretize the ethical considerations for sensors specifically with an expanded discussion about the unique considerations for implementing these technologies with youth. Ethical considerations are organized according to Belmont principles of respect for persons, beneficence, and justice (National Commission for the Protection of Human Subjects of Biomedical & Behavioral Research, 1979), with relevant examples of sensors commonly used in pediatrics (i.e., accelerometer-based devices that measure sleep and/or physical activity, electronic monitoring devices to measure medication adherence). We pose ethical case examples, questions, and recommendations across different stages of sensor implementation (see Table 1). We additionally provide recommendations and future directions for integrating both sensor-based and self-reported pediatric health data in research and practice.

Table 1.

Stages of Sensor Implementation With Ethical Case Examples, Key Questions, and Recommendations

Sensor implementation phases and ethical case examples	Key ethical questions	Ethical recommendations
Device selection  Case example: An adolescent expresses a preference to store their medication in a standard pill box rather than utilize an electronic adherence monitoring device, as the pill box has helped to foster adherence.	Does the sole reliance on sensors undermine the lived, real-world experiences of youth/families? Do recommended sensors undermine potentially more beneficial and/or preferred methods to track health behaviors? Has the sensor/device been tested for acceptability, feasibility, and validity in the researcher’s population of interest, including with marginalized populations?	Assess youth/family barriers to sensor use at baseline, including if sensor use would interfere with preferred adherence strategies (e.g., a standard pill box). When possible, select sensors/processing algorithms with validation evidence in the sample of interest, including considerations of race, ethnicity, SES, and developmental level. If the sensor device itself is part of an intervention (not a research tool), aim to ensure equity in who can access the sensor outside of the research context (e.g., by selecting lower cost device, allowing users to keep the device).
Data collection and management  Case example: A few weeks into a physical activity study, an enrolled adolescent expresses a belief that the accelerometer-based device can detect speech like other wearable devices (e.g., Apple Watch), so she removed the device periodically to maintain privacy.	Do youth and their consenting caregivers fully understand and consent to sensor-based data collection, including data sharing? Are adaptations needed to enhance data collection methods for individuals with lower digital health literacy and/or competing social–environmental stressors?	Create readable and multimethod sensor education materials (e.g., handouts, demonstrations, practice opportunities) to facilitate a more accessible informed consent document and process. Standardize the procedures/algorithms for periodically assessing sensor device use and accounting for periods of nonuse in data cleaning.
Data sharing and dissemination  Case example: A pediatrician questions the child’s subjective report of sleep difficulties and prolonged night awakenings as the accelerometer data show very high sleep efficiency (>90%). The pediatrician is unaware that the accelerometer device is less valid for identifying overnight awakenings.	Are sensor methodologies transparent to minimize harmful misconceptions and/or misuse by participants, their clinicians, and society at large? Are sensor data-sharing practices consistent with the goals and expectations of youth and other relevant individuals involved in their health/health care?	Increase transparency about sensor data processing, security, privacy, sharing, and integration with the health record. When shared with clinicians, sensor data should include a clinical interpretation. Incorporate youth and caregiver perspectives into data-sharing plans. Support continuing education efforts related to using and interpreting sensor-based data.

Note. SES = socioeconomic status.

Types of Health Sensors

Before discussing the ethical challenges related to sensor research, we first introduce two common sensors employed in pediatric clinical research.

Accelerometer-Based Devices for Sleep and Physical Activity Assessment

Sleep and physical activity are critical contributors to optimal child physical, neurocognitive, and social–emotional development (Chaput et al., 2020; Matricciani et al., 2019; Short et al., 2018), and there has been a significant increase in research using accelerometer-based devices to measure these behaviors in children and adolescents (Ancoli-Israel et al., 2015; Meltzer & Westin, 2011; Rowlands, 2007). Accelerometer-based devices are small, unobtrusive, and usually worn on the wrist, ankle, or waist for children ≤3 years of age to measure sleep and physical activity. Sleep research often refers to these devices and the methodology as “actigraphs” or “actigraphy,” respectively. We use “accelerometer-based devices” to refer to actigraphs and other sensors that measure accelerations to infer body movement and estimate sleep and physical activity patterns.

The gold standard procedures for measuring overnight sleep and total energy expenditure, which are polysomnogram or doubly labeled water (i.e., an isotope-based technique that measures carbon dioxide production), respectively, are often costly and can be challenging for research participants and patients to access. Alternatively, except for recharging, bathing, or swimming, most accelerometer-based devices can be worn continuously, offering unique behavioral insights into child sleep and physical activity patterns across contexts (e.g., home, school) and over multiple days or weeks (Meltzer, Montgomery-Downs, et al., 2012). Device algorithms, most of which are proprietary and/or brand-specific, typically aggregate the raw acceleration data to “counts” in 15-s, 30-s, or 1-min epochs. To assist in data scoring, some devices also include a light sensor and/or a photoplethysmography sensor to measure heart rate (Ancoli-Israel et al., 2015; Meltzer, Montgomery-Downs, et al., 2012), as well as an event marker that the user can press to indicate bed and wake times. Increasingly, consumer wearable devices (e.g., Fitbit) include wireless data streaming capabilities, which can provide “real-time” data and related feedback users.

Across brands, models, and studies, accelerometer-based devices generally have strong sensitivity in detecting the overnight sleep period but poor specificity in accurately identifying wake after sleep onset (Meltzer, Walsh, et al., 2012). Accelerometry is generally considered sensitive and specific for assessing time spent in sedentary, light intensity, moderate intensity, and vigorous intensity physical activity; however, the degree of sensitivity and specificity depends on the scoring approach and the child’s age. Validation research suggests that at least 5 days of data are necessary to reliably measure child sleep (Acebo et al., 1999). At minimum, 2 days of >10 hr/day of data are needed to reliably estimate child physical activity, with more recent data suggesting 3–4 days to enhance these estimates (Antczak et al., 2021).Overall, there is considerable variability in accelerometer data collection (i.e., duration of use, device placement) and management (i.e., scoring procedures) across study teams and disciplines, with limited standardization and reporting of these crucial procedures in accelerometer-based pediatric research publications (Meltzer, Montgomery-Downs, et al., 2012; Schoch et al., 2021). Variability in accelerometer-based research practices raises several ethical issues relevant to pediatric research.

Electronic Adherence Monitoring Devices

Each month, an estimated 22% of children in the United States are prescribed medication(s) to treat a medical or psychological condition (Hales et al., 2018). Despite the benefits of adhering to a prescribed medication, up to 50% of children and 75% of adolescents demonstrate nonadherence, or medication-taking behavior that does not align with agreed-upon recommendations from the clinician (Rapoff, 2010). Nonadherence is one cause of treatment failure (Sabaté, 2003) and is associated with preventable disease complications (Walsh et al., 2002), unnecessary increases in medications (Carmody et al., 2019), lower quality of life (Wu et al., 2014), increased urgent health care utilization (McGrady & Hommel, 2013), and health care spending totaling more than $300 billion per year (DiMatteo, 2004). A critical component of efforts to increase medication adherence includes the routine and valid assessment of adherence (Pai & McGrady, 2015).

“Real-time” medication adherence can be assessed using sensor-based electronic adherence monitoring devices—pill bottles, pill boxes, and inhaler attachments that contain a computer chip that records the date and time of device manipulations (e.g., bottle/box openings, inhaler actuations; McGrady et al., 2018). Electronic adherence monitoring devices can passively sense day-to-day patterns of adherence, a feature not possible with infrequently administered surveys or bioassays that provide an aggregate estimate of adherence across a period (Whalley Buono et al., 2017). Adding to their utility, adherence measured from electronic monitoring devices has converged with biological data (e.g., blood levels of medication or medication metabolites) and is related to important clinical outcomes (e.g., Rohan et al., 2017). For example, a multicenter Children’s Oncology Group study found that youth with acute lymphoblastic leukemia who took less than 95% of their prescribed oral chemotherapy (6-mercaptopurine) per an electronic adherence monitoring device had a 2.5 times greater risk of cancer relapse than individuals whose adherence rates were ≥95% (Bhatia et al., 2012).

Electronic adherence monitoring devices are typically more accurate than self-report measures because they are not subject to recall bias and can detect efforts to inflate adherence due to social desirability (e.g., repeated openings the day before a clinic visit), often making them a preferred measurement strategy (e.g., Vrijens et al., 2017). In an independent evaluation of commercial electronic adherence monitors, seven of 10 devices performed favorably, though each varied with respect to cost (McGrady et al., 2018).

Electronic adherence monitoring devices have been present and widely used in clinical research for decades. One commonly used device, the Medication Event Monitoring System (MEMS), was introduced in 1986 (Haberer, 2013). Still, methodological details describing their use have largely been neglected in publications. More commonly, publications include a summary of the device, such as patients were instructed to take all medication doses from the MEMS bottle, and then MEMS data were downloaded at the end of the study (Landier et al., 2017). To draw greater attention to the methodological nuances and complexities, McGrady and Ramsey (2020) published a step-by-step framework that helps researchers design, prepare, implement, and clean data from studies that use electronic adherence monitoring devices. We make the case here that ethical considerations underlie many critical decisions embedded in this framework, including selecting an appropriate device among an evolving and crowded marketplace of options and data management and integrity decisions.

Ethical Considerations

Respect for Persons

Respect for persons is an ethical principle that incorporates two convictions: (a) the need to treat individuals as autonomous agents who are capable of self-determination, such as pursuing their own personal goals, and (b) to protect individuals who have diminished autonomy (e.g., because of an illness or disability). Upholding this principle in sensor-based research with youth and their families requires attention to critical questions, including “Does the sole reliance on sensors undermine the lived, real-world experiences of youth and their families?” “Do youth and their consenting caregivers fully understand and consent to sensor-based data collection?” and “Are sensor data sharing practices consistent with the goals and expectations of youth and other relevant individuals involved in their health?”

Device/Measure Selection

A primary consideration related to respect for persons is whether relying exclusively on sensor-based measurement, without accompanying self- or caregiver-reported measures, undermines the lived experiences of youth and their families. To uphold respect for persons is to incorporate participants’ opinions and choices into the research process. An ethical dilemma is present when researchers assign greater value to sensors over self-report/caregiver-report, without considering that sensors also contain bias (because humans use and implement them). For example, a researcher may assume that any missing actuation from an electronic adherence monitoring device represents “true nonadherence,” neglecting the many reasons that youth may have taken their prescribed medication but did not utilize the device as intended (e.g., due to device malfunction, hospitalization, or when the patient “pocket doses” and does not use the monitor as instructed; McGrady & Ramsey, 2020).

Universal prioritization of accelerometer-based data over (or instead of) self-/caregiver-reported data may also lead to biased outcomes. For instance, accelerometers generally have poor specificity in accurately identifying wake after sleep onset, particularly in young children who exhibit increased movement overnight, leading to an overestimation of accelerometer-based night awakenings (Meltzer, Walsh, et al., 2012). Without an accompanying caregiver report of night awakening frequency and duration, the research team may characterize a young child’s sleep as being highly fragmented. Sleep researchers have long recommended using a concurrent sleep diary and/or device removal log to support accelerometer-based data scoring and interpretation (Ancoli-Israel et al., 2015; Sadeh & Acebo, 2002), yet recent research suggests that a sizable proportion of accelerometer-based studies do not report whether such measures were included or integrated into scoring procedures (Schoch et al., 2021).

Raw sensor data should be integrated with other sources (e.g., self-/caregiver-reported surveys, data from the electronic health record) to help contextualize sensor data, elucidate periods of device nonuse, and triangulate health behaviors. Of note, self-/caregiver-reported quantitative measures and qualitative data also play a critical role in both centering the experiences of marginalized youth and countering dominant and inequitable narratives (e.g., that medication adherence is a primary function of individual-level behavior change, with less attention to systemic factors that influence adherence, such as experiences of racism and access to affordable health care). Moreover, not every health behavior or health behavior change mechanism can be assessed reliably via a sensor (Li et al., 2020). Although accelerometers can measure many dimensions of sleep health, including nighttime and total (24-hr) sleep duration, sleep timing, and regularity, these devices cannot assess sleep quality (i.e., child and/or caregiver perception of “good” or “poor” child sleep) or perceived difficulties falling and/or staying asleep (Meltzer, 2018; Meltzer et al., 2021). Furthermore, self-report of youth health (also referred to as patient-reported outcomes) can yield contextual data that sensors cannot, such as adherence barriers (Plevinsky et al., 2020) and discrepancies in how youth and their caregivers perceive disease management (Miller & Drotar, 2003), sleep, and physical activity. For these reasons, at the point of device selection, we advocate for careful consideration of self-/caregiver-reported measures that will be collected alongside sensor-based data collection.

Data Collection and Management

While thoroughly explaining data collection procedures is essential in any informed research consent process, there are unique nuances for sensor-based research in pediatric populations. First, these technologies may be entirely new to users or confounded with other similar technologies that are more widely available in commercial markets (e.g., the differences between “research-grade” accelerometers and consumer devices), making it difficult for participants to have a fully operational understanding of how the sensor will work and how data may be transmitted during the study (e.g., remotely to a cloud via an app or the device itself vs. via manual download by the study team). Second, since youth under the age of 18 typically require a parent or legal guardian to consent for research procedures (even when the sensor use is carried out by youth), there are additional complexities related to explaining sensor-based data collection to both youth and caregivers, who may have differing levels of technology and health literacy and different concerns about data privacy and security. Third, sensors typically collect data passively and continuously, possibly leading users to forget that their data are being tracked over time (Rowe & Lester, 2020). This raises ethical considerations related to whether continuously using the sensor constitutes a choice to continue with the sensor/research procedures (Campbell et al., 2016), or whether there is a possible need for intermittent consent over time since sensor data collection, transmission, and storage can each appear “invisible” to the user (Ulrich et al., 2020). To ensure that consent for digital health research is truly informed, Nebeker et al. (2021) applied their Digital Health Checklist and language readability tools to create an accessible informed consent document. As this research continues to evolve, it will be important for study teams to test these methods with pediatric populations.

Data Sharing and Dissemination

Relevant to respect for persons, sensor-based monitoring of health behaviors may limit an individual’s choice to keep their health data private. Large data breaches have occurred with commercial-grade fitness trackers and wearables, including Fitbit and Apple products, when personal data (e.g., first name, last name, and date of birth) were stored in unsecured databases. Extending well beyond research studies, many consumers know very little about the data brokers and third parties who collect and trade information obtained by digital platforms and tools. Privacy notices that acknowledge these risks are lengthy, often unreadable to the average citizen, and may not reflect all periodic updates to the technology that can impact the specific privacy terms. Here lies a critical ethical challenge related to an individual’s choice to keep their health data private, as data may be shared without their full consent or without awareness from the researchers and clinicians who recommended the sensor in the first place. Responsive to these concerns, the U.S. Federal Trade Commission has indicated its intent to limit commercial surveillance, including soliciting public input on understudied but unique issues that may impact children and adolescents (e.g., whether platforms should attempt to enforce age requirements with users; Rinehart et al., 2022).

There is also concern about the extent to which sensors may be used to monitor individuals without their full knowledge or consent (Campbell et al., 2016), which may be amplified for marginalized populations who face racism and discrimination in health care settings and experience medical mistrust, which is an adaptive, functional coping response (Bogart et al., 2021). Related to data sharing with clinicians, numerous health systems have begun to share data from wearable sensors with clinicians via the electronic health record, with efforts to problem-solve related implementation challenges (e.g., difficulties synthesizing and translating mass amounts of sensor data into a clinically relevant format; Dinh-Le et al., 2019). To ensure respect for persons, we must be transparent about data privacy and sharing so users can make informed decisions about participating in sensor-based data collection (Graham et al., 2020). Additional research is needed to understand youth and caregiver preferences for sensor data sharing with clinicians (Wong et al., 2020), as well as clinician perspectives on the utility of sensor data in pediatric practice. Even when participants agree to share their data with clinicians, ethical challenges related to data use and interpretation can emerge, which we discuss further below.

Beneficence

The ethical principle of beneficence holds that clinical/public health researchers and psychologists should do no harm to participants and/or patients and strive to maximize any benefits while minimizing any harms that could emerge in research and/or treatment (American Psychological Association, 2017; National Commission for the Protection of Human Subjects of Biomedical & Behavioral Research, 1979). Beneficence is closely related to promoting the ethical principle of integrity, or accuracy in psychological research and practice. Related to this principle, we pose two core ethical questions: (1) “Do recommended sensors undermine potentially more beneficial and/or preferred methods to track health behaviors?” (e.g., when a study-issued electronic pill bottle requires a participant to stop organizing their medications with a standard pillbox) and (2) “Are sensor methodologies transparent, including strengths and limitations of data collection, management, and dissemination, to minimize harmful misconceptions and/or misuse by participants, their clinicians, and society at large?”

Device Selection

The potential for harmful and unintended consequences of initiating sensor use is a primary challenge to maintaining beneficence in sensor-based research. It is important that researchers anticipate and assess for these potential harms early in the research process. For instance, as electronic adherence monitoring devices require families to make changes to their medication administration process (e.g., using a study-provided device for medication administration, transferring refills to a study-provided device), it is critical to consider whether these changes may adversely impact medication adherence. One possible solution is to include device options (e.g., an electronic adherence pill bottle and pill box) that allow families to self-select the device that is most closely aligned with their current adherence practices, which also has implications for promoting respect for persons (i.e., autonomy) by providing a choice. Before enrolling families into a study, we recommend that teams discuss the family’s current medication-taking practices and their perceptions about the electronic adherence monitoring device. If families are currently using a strategy that does not align with electronic adherence monitoring device use (e.g., using a standard pill box without sensing capabilities) or express concerns regarding device use (e.g., the sensor-based pill bottle is too big to store in the bag where they usually keep their medication), teams should consider procedures to minimize potential risks and/or whether study participation is appropriate. Once families are enrolled, regular monitoring of child and family perceptions about any negative impacts or unintended consequences of sensor usage is recommended.

Data Collection and Management

Data management and integrity in sensor-based research may also present challenges to beneficence. There are many subjective decision points in selecting and applying scoring algorithms, with research suggesting that such decisions can have a meaningful impact on the validity of accelerometer-derived sleep and physical activity data (Meltzer & Westin, 2011). For example, even applying a widely used accelerometer algorithm (e.g., Sadeh or Cole–Kripke for sleep, Cole et al., 1992, Sadeh & Acebo, 2002; Evenson physical activity cut points, Evenson et al., 2008) requires individual decision making around setting the epoch length (Schoch et al., 2021); choosing the best physical activity cut point (Bianchim et al., 2020); determining the sensitivity thresholds for detecting movement during sleep, which vary developmentally; and integrating visual scoring and/or diary data, which may be discrepant from the auto-scored data (Meltzer et al., 2019).

Subjective decisions about data management and the limited empirical consensus in making these decisions are inherent to sensor research (McGrady & Ramsey, 2020), but they have the potential for harm when researchers employ methodologies that result in inaccuracies. For example, in pediatric sleep, scoring accelerometer-based nap data is particularly challenging, with little consistency or consensus available (Ancoli-Israel et al., 2015; Meltzer, Walsh, et al., 2012). As naps are developmentally typical for young children and may occur in older youth who nap to offset shortened overnight sleep duration (Meltzer et al., 2021), any accelerometer-derived assessment of total (24-hr) sleep time should aim to include naps. Still, many studies exclude nap data (Schoch et al., 2021), resulting in less accurate findings about linkages between sleep duration and child outcomes and about the proportion of children in the study meeting age-specific guidelines for total sleep duration. In addition, variation across research groups in applying scoring rules and identifying and defining periods of sensor nonuse can lead to inaccuracies in measuring sleep, physical activity, and medication adherence. That is, if researchers do not collect data on sensor use, data can be misclassified as 100% sleep efficiency and sedentary behavior (since no movement is captured) or a missed medication dose (since there was not an actuation of the electronic pill bottle/box/inhaler). To overcome these and many other subjective decision points, open-source algorithms applied to raw acceleration data are increasingly becoming available to help standardize the accelerometer-derived sleep and physical activity estimates (Belcher et al., 2021; John et al., 2019; Van Hees et al., 2015). However, subjective decision making is still needed to select among available algorithms, especially considering that many algorithms have been trained by samples that do not sufficiently represent racial and ethnic minoritized groups (Raza et al., 2023).

Data Sharing and Dissemination

In keeping with the principle of beneficence, some have argued that researchers have an ethical obligation to share valuable results with participants, which may include health behavior insights derived from sensors (Downey et al., 2018). When a study involves sharing results of sensor-based data collection with participants, as part of an intervention strategy or as an exit product once the study concludes, families may question the validity of the sensor data when the data do not align with their own experiences (e.g., if device use data are not accounted for). In a recent qualitative study of a mobile health pediatric sleep extension intervention, both caregiver and child participants expressed concerns about the accuracy of the Fitbit-based sleep duration feedback they received during the intervention (Mitchell et al., 2021). Few published studies qualitatively explore family views of sensor usage and validity in research.Building thiscomponent into pilot studies may help researchers become more attuned to potential challenges to beneficence when providing sensor-based health behavior feedback.

Unintended challenges to beneficence may also arise when results of sensor-based research are disseminated to youth and families. For example, when accelerometer-based devices are used to measure sleep or physical activity in research, participating families, clinicians, and consumers of the research findings may believe that all such devices and their features are comparable, reliable, and valid measures of sleep patterns and stages or physical activity levels, which is not supported in extant research (Guillodo et al., 2020). Given that consumer devices are readily available (e.g., Fitbit), families might rely on data from their personal wearable devices rather than seeking support from qualified health professionals to address questions about or concerns about their child’s health behaviors. In the case of pediatric sleep, this may be particularly problematic when such devices are incorrectly assumed to accurately reflect sleep stages, fragmentation, and oxygen, which could result in families overlooking symptoms that warrant clinical attention, such as sleep-disordered breathing.

Researchers must consider how to avoid data misinterpretation and/or misuse when participants agree to share their sensor data with the youth’s clinical provider(s). For example, although accelerometers can help monitor sleep duration and timing during pediatric sleep treatment, these devices are not indicated for use in diagnosing pediatric insomnia (Meltzer, 2018), which requires an evaluation of youth and/or caregiver perceptions about the child’s difficulty falling and/or staying asleep. Ideally, a clinical interpretation that includes a discussion of the device’s measurement scope, limitations, and use in clinical practice should accompany any sensor data that are shared with clinicians. In practice, however, such interpretations may be difficult to implement because, unlike other medical tests like x-rays or computerized tomography scans, there are few standardized guidelines for adherence monitoring and sleep sensor use in clinical care, while physical activity sensors are rarely used for clinical purposes (Ahmadi et al., 2020).

Ultimately, researchers are responsible for ensuring they have the necessary training to not only use sensors in pediatric research but also to provide accurate interpretations of sensor data results. Across populations (i.e., patients/families, clinicians, researchers, lay public) and dissemination outlets, we encourage researchers to clearly articulate what the device used can and cannot measure. Researchers should also provide sufficient details about device usage in peer-reviewed articles, such as whether open source or other research algorithms were applied and to differentiate study data from that obtained via typical (and often proprietary) device-specific algorithms. Likewise, to avoid causing harm, clinicians also require training around the use and interpretation of sensor data in patient care. Clinical institutions and funding agencies can support clinicians and researchers by including training on the use and interpretation of common sensors in continuing education and Responsible Conduct of Research courses. Institutional review boards can also support these endeavors by integrating ethical considerations from the Digital Health Checklist into their training, review procedures, and templated informed consent/assent documents (Nebeker et al., 2021).

Justice

The Belmont Report and Ethical Principles of Psychologists and Code of Conduct broadly describe the ethical principle of justice as ensuring fair and just access to, participation in, and benefit from human subjects research and/or psychological services. In 2021, the Secretary’s Advisory Committee on Human Research Protections acknowledged the limitations of the Belmont Report’s definition and framing of justice in human subjects research, noting that the Report largely focuses on protecting against egregiously unjust and/or exploitative research practices and emphasizes an “equal” rather than equitable balance of research burdens and benefits (Office for Human Research Protections, 2021). This perspective aligns with recent calls to recognize and dismantle racism and other forms of oppression in psychological research by embedding equity, as well as practices to promote diversity and inclusion, across the continuum of research, from study design to dissemination (Buchanan et al., 2021). Here, we pose two justice-related questions: (1) Has the sensor/device been tested for acceptability, feasibility, and validity in the researcher’s population of interest, including with marginalized populations? and (2) Are adaptations needed to enhance data collection methods for individuals with lower digital health literacy and/or social–environmental stressors?

Device Selection

A crucial justice-related ethical issue in pediatric sensor research is that most validation studies have been conducted with predominantly non-Hispanic/Latine white families and/or those of higher socioeconomic status (SES) backgrounds, with very few exceptions. In addition, studies rarely include children with neurodevelopmental differences, making device validity limited to neurotypical, non-Hispanic/Latine White, and/or higher SES groups. Limited or absent validity data have important implications for ethical and equitable device selection and usage in both research and clinical practice. As one example, research indicates that pulse oximeters are more likely to underestimate oxygen saturation levels and miss hypoxemia in African American/Black patients compared to non-Hispanic/Latine White patients (Feiner et al., 2007), potentially because these devices were validated with primarily non-Hispanic/Latine White individuals. Others have raised similar concerns about the accuracy of accelerometer device research (Colvonen et al., 2020). Researchers should strive to select sensor devices/associated algorithms with some evidence of validation in their sample of interest or, in the absence of validity data, should clearly indicate this limitation.

Researchers should also consider acceptability and comfort of sensor usage in their study sample. For instance, accelerometer-based devices are commonly worn on the wrist using straps made of materials that are not necessarily optimized for all children. Children with a neurodevelopmental condition, such as autism or Down syndrome, may find stiff plastic strapping to be physically uncomfortable and psychologically distressing given heightened sensory sensitivities. Thus, textile-based sensors may be more comfortable and less distressing. Some feasibility studies have been conducted on using accelerometry to assess sleep and physical activity in children with autism (Alder et al., 2022; Fawkes et al., 2015), with related recommendations for use, which can serve as a model for the much-needed acceptability and feasibility studies among other pediatric populations.

Data Collection and Management

The above validity gaps also raise questions about whether sensor-based methods, such as data collection and scoring procedures, can be effectively and equitably applied to generate accurate estimates across samples. It is likely that there are inherent biases and assumptions about users that impact data-related decision making and interpretation (Raza et al., 2023). For instance, to our knowledge, only one study of infants has evaluated impact of shared versus nonshared sleep surfaces (e.g., bed- or couch-sharing) on the validity of accelerometer-based sleep estimates (Camerota et al., 2018), yet this sleep arrangement is common among families with young children, including toddlers and preschoolers, in many parts of the world (e.g., African and Asian countries/regions (Mileva-Seitz et al., 2017; Mindell et al., 2010), including among Black and Latine families with young children in the United States; Barajas et al., 2011). In many cases, it is unclear whether research teams have included an assessment of the sleep arrangement or other aspects of the sleep environment when collecting accelerometer data, particularly when an accompanying child and/or caregiver-reported sleep diary was not administered.

The use of participant/family resources, such as access to smartphones, sufficient data plans, or a consistent internet connection, may further challenge inclusive and just participant enrollment and data collection in sensor-based research. As telemedicine implementation expanded during the coronavirus pandemic (COVID-19), so did concerns about an increasing digital divide limiting access to care for families of lower SES backgrounds (Eyrich et al., 2021; Tully et al., 2021). Many electronic adherence monitoring devices, accelerometers, and other consumer wearables require these resources for continuous data transmission. Lack of smartphone with data plan or internet access may prevent some families from enrolling in or completing a study and/or may contribute to missing data (Stiles-Shields et al., 2020). To provide fair and just access to research, teams are encouraged to budget for necessary devices (e.g., study-provided smartphones including data plans) when a participant does not possess their own. Research teams can also provide resources to support service access, such as by helping participants identify safe and accessible public Wi-Fi hot spots while considering the need to ensure privacy of participants’ health information (Eyrich et al., 2021).

Data Sharing and Dissemination

To uphold justice, the Belmont Report stipulates that publicly funded therapeutic device-related research should not confer advantages only to the populations that could afford to access the devices and/or those who are able to participate in clinical research. Accordingly, researchers using sensors to study whether these devices can enhance health behaviors should also consider whether the devices will be accessible when the research concludes. For example, given that wearable activity trackers can positively influence physical activity across a wide range of metrics and age groups (Ferguson et al., 2022), it may be beneficial to offer study participants the opportunity to keep the wearable device after the research ends. This approach could help increase equity in wearable ownership and related health benefits and could be proactively built into proposed research budgets. Applying the digital health equity framework (Crawford & Serhal, 2020) to identify the interacting and multilevel social and contextual factors that may contribute to sensor usage and dissemination-related inequities can help research teams integrate a health equity lens and uphold justice at the point of study design.

Conclusions and Future Directions

Sensors have dramatically increased the possibilities for unobtrusively collecting mass amounts of health behavior data within naturalistic environments, using devices that are often like or the same as technologies that are widely available to consumers. These attractive technical strengths have captured scholarly interest, with far less discussion of the human factors involved in implementing sensors—including with device selection, data management and integrity, data sharing and dissemination, and aligning recommendations for sensor use with the needs, goals, and expectations of youth and their caregivers. Without greater attention to the ethical challenges of sensor-based pediatric health research, psychological researchers and clinicians risk compromising beneficence, justice, and respect for persons. These risks could outweigh the scientific promise of sensor innovations, thereby negating their potential to improve child health and care.

Cross-cutting our recommendations for future research include (a) greater attention to equitable distribution of benefits and risks related to sensors; (b) the need to combine sensor data with self-reported measures of health behaviors, health behavior mechanisms, and device use; and (c) increased discussion of and transparency in the extent of subjective decision making when collecting, storing, scoring, interpreting, and disseminating sensor-based research findings. These recommendations apply to all psychologists engaging with sensor-based research, including trainees and research coordinators; investigators who design, implement, and disseminate sensor-based research; scholars who review sensor-based research in grant applications and journal outlets; and clinicians who consume research findings and apply them with pediatric patients and families.

Together, this article is a primer for ethics in sensor-based pediatric health behavior research and is not an exhaustive list of all possible ethical challenges and remedies or all types of sensors. The landscape of technological innovation and application of sensors in psychological research is evolving in exciting and unprecedented directions (e.g., combining data from sensors and repeated self-report to create digital phenotypes of psychological concepts, such as child anxiety; Nisenson et al., 2021)—so must our discussion of ethical use of sensors continue to evolve.

Public Significance Statement.

This article identifies and explores ethical considerations for implementing health sensors with pediatric populations. We raise awareness of the human factors involved in sensor implementation, which has been lacking to date, but has implications for the ethical use of sensors within research, practice, and public health initiatives.

Biographies

Alexandra M. Psihogios

Sara King-Dowling

Jonathan A. Mitchell

Meghan E. McGrady

Ariel A. Williamson