This is an open-access article distributed under the terms of the Creative Commons Attribution License (https://creativecommons.org/licenses/by/4.0/), which permits unrestricted use, distribution, and reproduction in any medium, provided the original work, first published in JMIR mhealth and uhealth, is properly cited. The complete bibliographic information, a link to the original publication on http://mhealth.jmir.org/, as well as this copyright and license information must be included.
There is growing interest in the potential for wearable and mobile devices to deliver clinically relevant information in real-world contexts. However, there is limited information on their acceptability and barriers to long-term use in people living with psychosis.
This study aimed to describe the development, implementation, feasibility, acceptability, and user experiences of the
A total of 15 outpatients with a diagnosis of schizophrenia used a consumer wrist-worn device and smartphone to continuously and remotely gather rest-activity profiles over 2 months. Once-daily sleep and self-rated symptom diaries were also collected via a smartphone app. Adherence with the devices and smartphone app, end-of-study user experiences, and agreement between subjective and objective sleep measures were analyzed. Thresholds for acceptability were set at a wear time or diary response rate of 70% or greater.
Overall, 14 out of 15 participants completed the study. In individuals with a mild to moderate symptom severity at baseline (mean total Positive and Negative Syndrome Scale score 58.4 [SD 14.4]), we demonstrated high rates of engagement with the wearable device (all participants meeting acceptability criteria), sleep diary, and symptom diary (93% and 86% meeting criteria, respectively), with negative symptoms being associated with lower diary completion rate. The end-of-study usability and acceptability questionnaire and qualitative analysis identified facilitators and barriers to long-term use, and paranoia with study devices was not a significant barrier to engagement. Comparison between sleep diary and wearable estimated sleep times showed good correspondence (ρ=0.50,
Extended use of wearable and mobile technologies are acceptable to people with schizophrenia living in a community setting. In the future, these technologies may allow predictive, objective markers of clinical status, including early markers of impending relapse.
Approximately 80% of those treated for the first episode of psychosis experience at least 1 further episode within 5 years [
Disturbances in rest-activity patterns are frequently reported in the early stages of relapse in schizophrenia. Sleep disturbance is the most common relapse indicator to be identified by family members [
Rest-activity profiles across phases of illness are poorly characterized, in part due to the challenges of sampling these rhythms over long periods in individuals with psychosis. To date, objective estimation of sleep under free-living conditions has been derived from wrist actigraphy [
Recent advances in consumer wearable technology offer several advantages over actigraphy for the remote, continuous, and unobtrusive sampling of rest-activity patterns across phases of illness, in ecologically valid settings. Marketed as lifestyle devices for improving health and well-being, they are designed to be aesthetically and functionally appealing for everyday use, and might, therefore, be more compatible with extended use than currently available actigraphy. Worn on the wrist, such devices measure motor activity via accelerometry and some sample mean heart rate using photoplethysmography. Heart rate may provide an additional predictive signal as cardiovascular regulation also expresses a circadian oscillation [
Crucially, most devices now communicate automatically and wirelessly with smartphones, thereby reducing the burden of data collection by allowing information to be uploaded over the mobile network and delivered to a researcher, clinician, or patient in near real-time. Some manufacturers provide an application programming interface (API), allowing researchers and software developers to access data directly from the company’s servers. The average cost of consumer wearable devices is a quarter to a third of that of currently available research actigraphs.
Accelerometers, gyroscopes, ambient light, and other sensors within smartphones can also be harnessed to infer rest-activity patterns through the
Although several studies have explored the feasibility and acceptability of smartphone interventions in schizophrenia [
A theoretical overview of the approach, where continuous passive variables that have been shown to co-vary with severity of psychopathology, including rest-activity profiles and heart rate variability, are captured using digital technologies (bottom). Disturbances in these variables may be detectable in the early stages of relapse (top), thus providing a window for preventative intervention (w).
This pilot study represents the first stage in an ongoing program of research for applying wearable and smartphone technologies for the identification of early signs of deterioration in schizophrenia, with a particular emphasis on passive sensing of rest-activity rhythms (
The
User-centered design: People living with psychosis were involved throughout the development and testing cycle. Consultation groups consisted of patients who advised on the selection of wearable and mobile devices, the design of the software app, and aspects of the study design including recruitment strategy, feedback of data, and incentivization. The subsequent acceptability-feasibility study comprised a separate group of patients living in the community.
Integration with everyday life: Initial user-group testing with a range of currently available research wearables (GENEActiv, Activinsights, Cambridge, UK; ActiGraph GT9X Link, Actigraph corp, Pensacola, USA; Empatica E4, Milan, Italy) suggested that they would not be acceptable for extended use due to their design and limited functionality. To enhance acceptability and minimize user burden and stigma, widely available consumer-oriented technologies were therefore considered. The user groups favored the wrist-worn Fitbit Charge HR (Fitbit Inc, San Francisco) due to its appearance as a lifestyle device that is acceptable to both younger and older users and the ability to view metrics relating to sleep and activity via the Fitbit app.
Wireless functionality: The Fitbit provided wireless data transfer to a smartphone and allowed access to minute-level activity, sleep, and heart rate data via API calls to the Fitbit server [
Remote and real-time: Each participant was provided with a 4G mobile data plan, which allowed data to be continuously uploaded to the research server and adherence with the wearable device, sleep, and symptom diary to be monitored in real-time. Participants with a home wireless network could also upload data via this route.
Secure: Safeguarding privacy and data security was a cornerstone of the platform. Usernames were obfuscated using an MD5 hash algorithm [
Open source software: Code for the platform architecture is publicly available to promote replicability, refinement of the software, and assessment of external validity through studies in other centers.
The
Interoperability with existing local electronic patient clinical record systems was not supported at this stage, though future integration is planned.
Sleepsight platform architecture.
Passive variables provided by the
Device and sensor(s) | Output from sensor(s) | |
|
||
Tri-axial accelerometer | Raw data not availablea | |
Photoplethysmogram (optical sensor) | Mean heart rate/minute | |
|
||
|
Tri-axial accelerometer | Acceleration/g sampled at 10 Hz |
|
Light sensor | Ambient light intensity/lux |
|
Battery probe | Battery level and charging events |
|
Screen event probe | Whether or not the screen was active |
aInstead, several derived variables computed by Fitbit available: steps per minute; lightly, moderately, and very active minutes; sleep onset, offset, number of awakenings, and total sleep time.
Two sample variables from the researcher-facing dashboard for a single participant. Upper panel: accelerometer output from smartphone showing periods of rest-activity over a continuous 36-hour period; lower panel: smartphone screen state showing screen doze state (lower bars) and active screen state (upper bars) over 7 days.
A total of 15 adults with a diagnosis of schizophrenia were recruited through clinical teams from community psychiatric services in South London (
Screening for eligibility and all clinical assessments were undertaken by an experienced psychiatrist (NM). Participants were asked to use the system continuously for 8 weeks, during this period, passive sensing and symptom rating variables were collected remotely.
On enrollment, informed consent and psychiatric, medical, and sleep histories were obtained. The PANSS, Pittsburgh Sleep Quality Index [
Clinical characteristics of participants at baseline.
Clinical characteristics | Statistics | |
|
||
|
Male | 9 (60) |
|
Female | 6 (40) |
Age, mean (range) | 44.1 (30-54) | |
Duration of illness in years, mean (range) | 16.6 (5-33) | |
|
||
|
Positive subscale | 13.5 (7-23; 4.3) |
|
Negative subscale | 19.7 (8-36; 7.5) |
|
General subscale | 25.1 (16-40, 6.4) |
|
Total score | 58.4 (32-81; 14.4) |
|
||
|
Clozapine | 7 (47) |
|
Oral antipsychotic | 5 (33) |
|
Depot antipsychotic | 3 (20) |
|
||
|
Nontouchscreen mobile phone | 7 (47) |
|
Touchscreen smartphone | 8 (53) |
They were asked to wear the Fitbit on the nondominant wrist at all times, except for when bathing or swimming, and they were advised to charge the smartphone once a day when going to bed and to charge the device every 4 to 5 days. It was explained that the purpose of the study was to assess feasibility and acceptability; therefore, putative markers of relapse would not be monitored or responded to during the study.
In addition to the provision of a mobile data plan, it was explained to participants on enrollment that they would be given the option of keeping the wearable device and smartphone on study completion to provide an incentive for engagement and completion and compensation for their time. However, the equipment would remain the property of the research team for the duration of the study, and they would be asked to return them if they withdrew before study completion. This strategy was formulated with user groups, and it was thought to be a fair and appropriate approach.
The 8-week data collection period then began, and participants were free to use the smartphone for making calls, messages, and Web browsing. Participants were asked to use the study smartphone as their primary device, and where necessary, their pre-existing telephone number and personal information were transferred to the new device. Participants received a personalized encouragement text message once weekly, with the aim of promoting motivation and engagement with the research team, and participants were instructed not to perform any updates to the Fitbit firmware during the course of their participation. Data from each user were monitored on a daily basis. Where there was evidence of continued nonadherence for 2 or more days—for example, no signal from the wearable device—a text message prompt was sent to the participant, followed by telephone contact if necessary.
Feasibility was defined as the proportion of participants using each element of the system (wearable device, sleep log, and symptom log) for at least 70% of the 8-week study period. “Definitely feasible” was defined as ≥70% participants meeting the criteria; “possibly feasible” as 50% to 69% participants; and “not feasible” as <50% of participants. Removal of the wearable device was detected by the absence of the heart rate signal. In a test conducted separately (see
The symptom diary items were developed and validated in schizophrenia in the Clintouch study [
Acceptability and usability were assessed at the end of the study using a questionnaire conducted in person, adapted from previous studies [
Quantitative analyses were undertaken using R software [
End-of-study questionnaires are reported descriptively. Interview transcripts were cross-checked for accuracy by a second researcher. Analysis of qualitative end-of-study user experiences followed a grounded theory approach [
Ethical approval was obtained from the Dulwich Research and Ethics Committee, London, United Kingdom. As a research study that did not alter routine clinical care and had no intention of commercialization, national regulatory approval from the Medicines and Healthcare Regulatory Agency was not required at this stage.
Of the 15 participants, 14 completed the study, with varying levels of adherence to each element of the system. One participant withdrew from the study at the end of week 1 due to finding the wearable device uncomfortable, particularly in bed, and returned the study devices. That participant did not report suspicion toward the devices as a cause of discontinuation and had a low total PANSS score of 47. Rather than adopting the study phone, 1 participant wished to continue to use her own smartphone, and it was therefore agreed that she would use this in parallel with the study device. The mean monthly data volume generated by the system per participant was approximately 1.1 GB, not including personal data use. One participant used the monthly data allocation within the first 3 weeks and was provided with an unlimited data plan for the remainder of the study. There was no loss or damage to devices.
For the wearable device, all the participants exceeded the 70% threshold for feasibility, with a mean wear time of 21.8 hours/day or 91% of the total study duration.
For the
Spearman correlation demonstrated a significant negative relationship between PANSS positive score and sleep diary completion (ρ=−.49,
Although adherence to the wearable device remained relatively stable throughout the 8-week study period, sleep and symptom log completion dropped off toward the latter quarter of the study, though remaining mostly above 70% (
Overall adherence to wearable device, sleep, and symptom diaries for each participant.
Mean longitudinal adherence to wearable device, sleep, and symptom diaries for all participants over the 8-week study period.
Overall, responses to the end-of-study questionnaire relating to the
All participants found the personalized encouragement text messages to be helpful in improving motivation. In addition, 10 users were motivated by the prospect of receiving devices at the end of the study, and a similar proportion of participants reported frequently using the Fitbit app to view their sleep and physical activity patterns. All participants expressed a wish to keep the study devices after the study had ended. Moreover, 3 participants reported feeling occasionally suspicious toward the technology and worried that their personal information was at times being monitored. However, this did not lead to discontinuation.
Of the 14 participants, 13 reported an inverse association between their sleep and mental well-being, and 12 of the 14 participants reported significant sleep disturbance around the time of their first episode of psychosis or subsequent relapse.
Further themes extracted from the end-of-study interview clustered around 3 major themes: attitudes and beliefs toward the concept of remote sensing for predicting relapse, factors relating to the technology, and factors relating to the users’ illness in interacting with digital technologies.
All but 1 participant reported experiences of sleep disturbance as an important problem, either during periods of stability or around episodes of acute illness, and welcomed the emphasis on sleep as a possible predictor of deterioration. Many felt that measuring sleep patterns would be an important element of self-management and could imagine how psychological or pharmacological interventions delivered in the early stages of relapse could be therapeutic. Overall, 5 participants commented on how they would have concerns over false alarms—for example, whether sleeping poorly in the absence of deterioration in symptoms would trigger a response. One preferred the terms sleep “tracking” or “logging” rather than “monitoring,” which held connotations of surveillance. Moreover, 4 participants felt that their engagement in the study was influenced by the knowledge that their data were not being analyzed and responded to in real time, and if this were to have been the case, they would have greater levels of engagement.
Half of the users commented that the wearable was at times uncomfortable: the strap would feel tight, which was particularly bothersome in bed. However, most of the users reported initial discomfort, which improved as the study progressed. One participant reported intermittent concern over radiation emerging from the optical heart rate monitor. There were no reports that the sleep metrics were inaccurate; however, 3 participants felt that the Fitbit overestimated step count. A third of participants found value in using the Fitbit for setting and attaining physical activity goals. Charging of the Fitbit or smartphone did not raise any concerns. Two-thirds of participants found that completing the daily symptom diary helped them reflect on mental state and gain greater insight into the links between their symptoms and behaviors, especially sleep. A sense of being “acknowledged” as a specific consequence of using the symptom diary was reported by 3 individuals. Moreover, 2 mentioned how it encouraged family to be involved and became a focus for education. In addition, 6 participants stated that the symptom diary became repetitive and tedious, particularly when the users felt their symptom burden was low, and that if there were no incentive, they would be unlikely to answer the questionnaire on a daily basis. Two users found some questions difficult to understand. A wish to have greater flexibility in the timing of diary completion was raised by 2 users.
A quarter of participants expressed concern that adherence would diminish during relapse. First, users’ cognitive function and therefore the ability to interact with the technology might decrease as symptoms escalate. Second, users may develop suspicion toward the devices, leading to discontinuation. Generally, users felt that the wearable device would be better tolerated in this scenario than the smartphone diaries.
Correspondence between self-reported (sleep diary) and objectively (Fitbit) determined daily time in bed for all participants showed good overall correlation (ρ=.50,
More detailed visualization of these data in 2 participants suggests that the Fitbit captured both within and interindividual variability in rest-activity profiles.
Comparisons of subjectively (sleep diary) and objectively (Fitbit) determined daily time in bed for each participant, fitted with robust bisquare regression to account for outliers. *, ** Significance at the .05, <.01 levels, respectively, for the 2-tailed test.
Rest-activity profiles from 2 participants—one with a regular rest-activity profile (A-C) and another with a highly variable, free-running circadian rhythm, not entrained to the day-night cycle (D-F). A and D: heart rate data from the wearable device, with darker shading indicating higher mean heart rate. B and E: subjective sleep times from sleep diary. C and F: subjective sleep times superimposed upon heart rate data (see main text for further details).
However sophisticated the engineering and analytic solutions, the success of digital interventions in improving clinical outcomes will depend crucially on whether they are adopted by patients for long-term use. We collaborated with patients in the design and deployment stages and asked participants with severe mental illness, under the care of secondary level psychiatric services with a range of positive and negative symptom severity, to use wearable and smartphone devices over 2 months in a naturalistic, ecologically valid setting. Our findings suggest that people with schizophrenia are enthusiastic about the concept of sleep disturbance as an indicator of relapse, and they are willing to adapt existing consumer technologies in managing their condition. Consistent with previous research examining smartphone interventions [
The relatively high rates of adherence in this study may be attributable to the incentives that were offered. Though all participants reported this as a motivating factor, the majority also reported intrinsic interest in the system and wished to continue using it after the study ended. Nonetheless, enhancing engagement and motivation by providing users with information that they value is a key challenge for the field. Each user is unique in which elements of the system they find engaging. Some participants may value feedback on physical activity, others on sleep duration, and others on their self-rated symptoms. Adaptable systems that can be tailored to each user may be a solution. Given that this was an observational study, we opted against providing explicit feedback of passive sensor or diary data, due to its liability to influence the underlying behavior we aim to examine; however, this may have reduced levels of adherence. Alternative approaches such as “gamification” [
Furthermore, promoting motivation by providing relatively low-cost mobile communication technologies, which many patients may otherwise be unable to afford, may have the added benefit of enhancing social interaction, communication with the care team, and functioning. In a clinical scenario, should mobile technologies be shown to be effective in reducing relapse, provision of equipment and a mobile data contract would represent a relatively insignificant financial outlay in comparison with the personal, social, and economic costs of relapse [
Cost is also an important factor in considering the scalability and reach of digital interventions. The cost of the Fitbit Charge HR was £100, the Moto G smartphone £140, and the mobile data plan £14 per month. The overall cost for consumables per user for the study was, therefore, £270, which equates to approximately just over half the cost of a day on an acute psychiatric ward [
The use of consumer wearable devices in clinical contexts raises several important considerations. These devices are marketed as products that claim to enhance fitness and well-being, often supported by bold but unsubstantiated claims. Although some validation data are published from healthy populations showing acceptable agreement between consumer devices to actigraphy for some physical activity measures [
We suggest that they do, depending on the question being asked [
Though we welcome further validation studies of consumer against gold standard devices, consumer devices follow a rapid product cycle, meaning they are rapidly superseded by newer models. Deciding on which device to study, and in which clinical groups, is therefore challenging. Another issue is the closed nature of these devices, due to reasons of intellectual property, manufacturers do not publish the algorithms through which activity and sleep outputs are calculated, and the researcher generally lacks control over the implementation of algorithm updates during the course of a study. In addition, preprocessing of data on the consumer device limits their granularity to minute-level data, at best. In turn, this might constrain the choice of subsequent analysis, for example, it is uncertain whether nonlinear patterns of motor activity [
In navigating these challenges, greater collaboration between medical and consumer device manufacturers (each of whom operate a fundamentally different business model), researchers, and consumers will be necessary for driving innovation in this area. It is worth reiterating that research grade, wearable accelerometers that are acceptable for long-term use do not currently exist, and further research is urgently needed to evaluate the reliability and validity of consumer wearable and mobile devices as clinical tools [
This small pilot study was designed primarily to evaluate feasibility and acceptability in patients in remission, and at this stage, we are unable to draw any conclusions about its potential for identifying early markers of relapse or offer evidence of the validity of the signal from the wearable device. Our supplementary trial suggested that the absence of HR signal is highly correlated to nonwear; however, movement of the device may activate the HR sensor, and there remains the possibility that spurious readings were produced while the device is removed from the wrist. Participants were in a relatively stable phase of illness, and the influence of worsening psychopathology on adherence was not examined directly. Future studies should aim to test adherence over longer periods, in more unstable populations such as in the period following hospitalization [
We found enthusiasm and high levels of engagement with passive monitoring in people with schizophrenia, which exceeded our acceptability thresholds, across a range of ages, symptom dimensions, and severity. However, mHealth technologies are unlikely to be acceptable to all patients, and it will, therefore, be important to further understand which groups are most likely to engage with and benefit from these interventions, while maximizing their reach by developing adaptable, tailored tools. An inherent property of participatory interventions is that they are likely to improve participants’ outcomes and relapse risk, for example, by encouraging the user to reflect on their symptoms and monitor variables such as their sleep [
A further under-researched area is how the adoption of digital interventions by health care professionals can be facilitated [
Extracting clinically meaningful information in real time from high-volume, multidimensional data represents another major challenge [
A long-term ambition of this approach is to enhance clinical care by identifying early signs of deterioration and thereby facilitate self-management and preventative psychological or pharmacological intervention. As research in this area continues to advance, future efforts should focus upon how clinicians, patients, technologists, and the industry can work together effectively to maximize the clinical utility, validity, and implementation of these novel tools.
Responses to end-of-study questionnaire; validity of heart rate signal for adherence.
application programming interface
Global Positioning System
Positive and Negative Syndrome Scale
The authors thank the participants for their valuable contributions and the Clintouch investigators (University of Manchester) for kindly sharing the validated questionnaire items. The authors acknowledge the contributions of the service user advisory group (SUAG) and Young Person’s SUAG, provided by the National Institute for Health Research (NIHR) Maudsley Biomedical Research Centre, comprising people with experience of mental health problems and their carers who have been specially trained to advise on research studies. The authors thank Dr Sara Simblett for developing the coding frame and for reviewing the manuscript. This research was funded by the Medical Research Council Confidence in Concept Fund. NM is supported by a Medical Research Council Clinical Research Training Fellowship, MR/P001378/1. MK, AF, and RD are supported by the National Institute for Health Research Biomedical Research Centre at South London and Maudsley National Health Service (NHS) Foundation Trust and King’s College London. This paper represents independent research part funded by the National Institute for Health Research Biomedical Research Centre at South London and Maudsley NHS Foundation Trust and King’s College London. The views expressed are those of the authors and not necessarily those of the NHS, the NIHR, or the Department of Health.
NM wrote the manuscript and collected and analyzed the data; MK and DWJ analyzed the data; and MK, AF, RJ, and CK developed the smartphone app and data handling infrastructure. RD and JM provided study oversight. All authors read, contributed to, and approved the final manuscript.
RJ is an employee of AstraZeneca.