Integrating smartwatches in community mental health services for severe mental illness for detecting relapse and informing future intervention: A case series

This case series explored the integration of smartwatches in a community mental health service to support severe mental illness (SMI) management and intervention. We examined whether biometric data provided by smartwatches could help to predict relapse and inform treatment decisions.


| BACKGROUND
Severe mental illness (SMI), characterized as a chronic relapsing condition with profound impact on the individual, family, and society (Pratt, 2014), is estimated to affect 5% of the population (Steel et al., 2014).Since the 1960s, there has been a worldwide trend in treating SMI from hospital settings to a more community-focused approach (Fakhoury & Priebe, 2002).In community settings in Australia, multi-disciplinary teams provide intensive case management of patients, including a range of treatment and support services (O'Donnell et al., 2021).Clinicians within these services face several challenges, from patient level (limited insight into emerging symptoms) and job requirements (high patient loads) to systemic issues (working within a highly complex fragmented system).Technological innovations may offer a means to overcome substantial challenges clinicians face within this space.
Smartwatches are devices with the ability to capture key indicators of health including a person's activity level, sleep, heart rate, and electrodermal activity (EDA) (Reeder & David, 2016).Such markers have been shown to be useful predictors of mental health; predictive of relapse (Fonseka & Woo, 2022), treatment efficacy (Winkler et al., 2014), and diagnostically useful in identifying circadian disruptions (Jones et al., 2005).For clinicians, smartwatches may be an appealing prospect for both monitoring and treating patients.They are generally unobtrusive, affordable, and can provide a comprehensive analysis of a patient's overall health (Reeder & David, 2016).
Importantly, these devices have high levels of community acceptance (Hunkin et al., 2020).They also hold the benefit of being appealing to young people, who are a population characterized by reluctance to seek help from formal mental health services (Rowe et al., 2014).Despite these potential advantages for treating SMI with smartwatches, no study to date has examined their utility for predicting and helping to prevent psychological deterioration in a youth community mental health service.
Our case series illustrates the feasibility and potential utility of integrating smartwatch devices within a youth community mental health service for predicting relapse and informing treatment decisions.Patients (aged between 19 and 24) were asked to wear the mHealth Empatica Embrace2 device on their wrists for 6 months as part of a randomized controlled trial (the unWIRED study) (Byrne et al., 2020).This device has a biosensor that captures, stores, and transmits EDA and motion data, and can detect autonomic events and categorize changes across the sleep-wakefulness cycle.This device was chosen due to its capacity to monitor sleep, activity, and stress (see Table 1 for more detail), all of which have been shown to be sensitive in predicting and indexing SMI psychopathology and relapse (Iniesta et al., 2016;Long et al., 2022).Case managers were able to observe patient data through a dedicated website.Our case series of four participants who enrolled in the study highlights the potential utility of integrating this technology within existing services.
Please note that patient names are pseudonyms.

| BIOMETRIC INDICES TO INFORM RISK OF RELAPSE CASE STUDY 1: RILEY
Riley was a 19-year-old female with a history of psychosis.She was previously scheduled as an involuntary patient when she acted on her persecutory delusions.Riley was placed on a Community Treatment Order (CTO) and had been prescribed 25 mg Risperdal, 2.5-5 mg diazepam, and 2 mg risperidone.Nine months post admission, Riley participated in the unWIRED study.Two weeks later (6th January), Riley contacted the mental health line reporting a 'mild crisis', then two days later was scheduled as an inpatient due to an increase in persecutory delusions.
Sleep.A significant decline was observed 3 days prior to her call to the crisis service, with the average amount of sleep decreasing from 8.97 (SD = 0.96) to 3.57 (SD = 1.39) hours per night (Figure 1).Physical Activity.Riley's physical activity displayed the opposite trajectory to her sleep patten (Figure 2).Prior to the 4th of T A B L E 1 Metrics and method of calculation of motion and electrodermal activity data acquired from the mHealth Empatica Embrace2 device.EDA.The most substantial changes noted in Riley's EDA were observed during her wake period, with less autonomic activity detected in the days prior to her admission compared to the week prior (Figure 3).

| Case study 2: Sinclair
Sinclair was an 18-year-old trans-male with a diagnosis of borderline personality disorder and gender dysphoria.In the year prior to their enrolment in the trial, Sinclair had four presentations to the emergency department (ED).Sinclair was seeing a psychologist weekly and was prescribed fluvoxamine 100 mg; Quetiapine 300 mg oral modified release tablet.A relapse occurred whilst Sinclair was part of the trial.Sinclair attended the ED with somatic complaints (9th December), absconded, then presented the next day with evidence of self-harm and was discharged three days later.
Sleep.Across the initial part of the trial, Sinclair was on average sleeping 7.19 h per night (SD = 0.69) Two days prior to attendance at the ED Sinclair was sleeping 7.39 h per night (SD = 0.52).
Physical Activity.On the day of the initial presentation to the ED, there was a substantial change in Sinclair's level of activity (Figure 4).
Prior to this, Sinclair was on average taking 11 380 steps per day (SD = 4769.33)but on the 9th this increased to 23 340 steps in a day.
EDA. Sinclair's EDA levels highlighted the degree of affective instability (Figure 5).There is marked variability and high levels of EDA activity, which occurred independent of the amount of activity.
Interestingly in the days prior to their presentation at the ED, there was a pronounced lack of activity during their sleep relative to the week prior.

| Biometric indices to inform treatment decisions case study 3: Jasmine
Jasmine was a 24-year-old female with a primary diagnosis of bipolar affective disorder.In 2021, Jasmine was admitted to an inpatient unit F I G U R E 3 Riley's (case study 1) electrodermal activity from 30 December 2021 3 AM to 04 January 2022 9 AM (i.e., the week before readmission due to reporting of a "mild crisis" and increases in persecutory delusion symptoms).
and placed on a CTO after experiencing mania, delusions, suicidality, and auditory hallucinations.Jasmine was prescribed aripiprazole 400 mg IM monthly, Chlorpromazine 75 mg nocte, and Lithium 900 mg nocte.Following her admission, Jasmine was referred to a community treatment team and was seeing an occupational therapist weekly.
Initially, Jasmine's therapy goals were primarily focused on making friends, however Jasmine's smartwatch data revealed that at the time of beginning the trial she was on average sleeping between 2:29 AM and 4:06 PM (M = 14.43 SD = 1.12) and was on average walking 333 steps per day (SD = 186.10).This additional data allowed for significant insight into the extent of Jasmine's problem (i.e., issues with sleep and activity) but also a focus for intervention.The clinician initially implemented a behavioural modification program that sought to increase Jasmine's engagement in outside activities and improve their sleep schedule.The clinician monitored the level of activity and used this to provide feedback and encouragement.Five months into the intervention, Jasmine's step count reflected a 325.8% increase in her level of activity.Jasmine's sleep cycle also progressively changed to sleeping between 10:42 PM and 11:44 AM (M = 12.86, SD = 1.17).
An interesting finding was the normalization and decrease of Jasmine's EDA, which occurred during the intervention (Figure 6).Notably, Jasmine's EDA was still more pronounced during her sleep as opposed to her wake period, highlighting further work is required to increase her level of activity during the day.

| DISCUSSION
Our case series illustrates how wearable technology could be integrated into community health services including the potential utility of predicting relapse and supporting monitoring and treatment of symptoms of SMI.Notably, clinicians in community health settings are consistently faced with difficult decisions when it comes to managing risk and planning future care.Clinicians are reliant on conducting risk assessments whose predictive properties are notoriously poor.
For example, the strongest discriminator of suicide risk is a current or recent psychiatric hospital patient, yet this represents the majority of clients seen in community mental health services in Australia (Chan et al., 2016).Clinicians therefore are being placed in precarious positions where the tools at their disposal to inform clinical decision making are severely limited.
In our case series, we presented two instances (Case study 1; Riley; psychosis; Case study 2; Sinclair; borderline personality disorder) where biometric indicators preceded the eventual admission to an inpatient unit.In both instances there appeared to be changes in circadian rhythm represented by changes in sleep patterns or changes in EDA.At this stage, it is difficult to know whether these signals are predictive or whether post-hoc reasoning has influenced our interpretation.Further, our biometric indicators of relapse were not necessarily aligned across these two cases (e.g., sleep disturbances preceded crisis in Riley's case, but not in Sinclair's case).However, this underscores the potential importance of personalized medicinean F I G U R E 6 Jasmine's (case study 3) electrodermal activity from 24 December 2021 12 PM to 07 January 2022 9 PM, demonstrating normalization of EDA as she engaged with psychological treatment.
F I G U R E 7 Krystal's (case study 4) electrodermal activity from each instance of exposure to the social event organized by the mental health service.
approach that considers individual differences in patients' genes, environments, and lifestyles with the goal to tailor interventions to the characteristics of each patient (Maughan, 2017).Ultimately, understanding an individual's unique digital footprint offers clinicians the ability to utilize dynamic risk factors to inform their decision making.
Future research may therefore consider exploring the potential utility of using these devices on people post-admission where the incidence of relapse is substantially higher.Additionally, given the amount of data involved, machine learning may offer a novel solution in being able to assist in identifying the personalized relapse signatures of individuals (Sánchez-Reolid et al., 2022).Ultimately, future research should explore the utility of such technology to assist in improving the care that individuals receive in community health services.
Presently, the methods through which treatment effects are measured in community services is largely reliant on patient self-report.
The second two case studies illustrate how integrating smartwatches into a service can introduce objective data for the clinician to use to inform treatment decisions.In case study 3 (Jasmine; bipolar affective disorder), the clinician became aware of how poor the patient's sleep and activity levels were, and then implemented a targeted treatment to address this.In this example, the use of technology was able to overcome the significant lack of patient insight.
In case study 4 (Krystal; agoraphobia), the clinician was able to use biometric data to monitor the implementation of their exposure protocol, and make adjustments to their treatment plan to target safety behaviours.This highlights the usefulness of monitoring arousal through EDA during behavioural therapeutic tasks.Cognitive models explain agoraphobia with recurrent panic attacks as stemming from exaggerated beliefs that anxiety and bodily sensations will result in physical or mental catastrophe (Hoffart, 2016).Remission of symptoms occurs when there is adoption of a more benign and realistic alternative explanations, achieved through repeated behavioural experiments.Whilst there was a reduction in the level of physiological arousal the effects did not generalize beyond the specific event.As noted by the clinician, the generalization of the improvement will be the on-going focus of treatment.
Despite encouraging results across the four presented case studies, the generalisability of our findings is limited due to our small sample size.A more in-depth study with a larger sample size is required to examine in more detail how EDA can be used in community mental health settings for detecting potential relapse and inform timely action.Additionally, there are several ethical and practical concerns, which need to be considered.The wearable devices used collect highly personal data, however currently this is limited to activity, sleep, and EDA.With increasing technological innovation comes increasing potential to capture more data from a person, however it's benefit must constantly be evaluated.In a high-risk population, there is forever the delicate balance between an individual's freedom and safety, and it may be easy to justify the usage of these devices in lieu of the evidence to support its use.This will continue to be an ongoing area of debate.
January, Riley averaged 5551 (SD = 4611) steps per day which increased to an average of 10 200 (SD = 2871) across the next three days.
Krystal was a 22-year-old female with diagnosis of agoraphobia, social and generalized anxiety.These issues onset at age 17, resulting in school refusal and an inability to leave the house.Since this time, Krystal has been housebound.At the time of enrolling in the study, Krystal was prescribed 250 mg of sertraline and 25 mg of clonidine.Krystal was having weekly sessions with her community health clinician focusing on treating her agoraphobia and social anxiety.Krystal's clinician utilized the smartwatch to monitor the exposure protocol that was being implemented, specifically exposure to unfamiliar locations and a weekly social event run at the community mental F I G U R Sinclair's (case study 2) daily step count from 02 December 2021 to 09 December 2021.Source: *day that Sinclair presented at ED with somatic complaints; Sinclair engaged in self-harm on 10 December 2021.F I G U R E 5 Sinclair's (case study 2) electrodermal activity from 03 December 2021 3 AM to 09 December 2021 9 PM (i.e., the week leading up to Sinclair's presentation to ED due to somatic complaints and subsequent self-harm.health service.Krystal's EDA was then explored in parallel with her subjective ratings.

Figure 7
Figure7highlights the changes in EDA across time when Krystal would attend a social event organized by the service (from 12:30 PM to 2:00 PM).During the first visit Krystal showed relatively consistent EDA, however following the visit, Krystal experienced a panic attack at approximately 4:30 PM that same day.The clinician hypothesised that Krystal's utilization of safety behaviours inhibited her arousal, delaying the onset of panic symptoms.A behavioural experiment was then performed during the next exposure involving 'dropping' the safety behaviours.This was mirrored in the EDA where there was an increase in activity during exposure.No panic attack was experienced after exposure, which was also reflected in the EDA.The exposure protocol continued with a focus on safety behaviours until third exposure, where there was little change from baseline EDA, suggestive of the effectiveness of the intervention.
In conclusion, our case series demonstrates the promising potential of integrating smartwatch technology into community mental health services for individuals with SMI.The use of wearable devices allows for the continuous monitoring of key indicators like sleep, physical activity, and EDA.The presented case studies, encompassing diverse diagnoses, provide insights into the predictive utility of these biometric indicators and their role in informing treatment decisions.ACKNOWLEDGEMENTOpen access publishing facilitated by The University of Sydney, as part of the Wiley -The University of Sydney agreement via the Council of Australian University Librarians.
Riley's (case study 1) daily step count from 28 December 2021 to 07 January 2023.Source: *day that Riley contacted the mental health line reporting a mild crisis.