The future of digital mental health?

An increasing number of people, psychologists among them, argue that the future of health is digital. A 2012 review by Elizabeth Murray at UCL’s e-health unit noted our ageing population, and pointed to the convenience, accessibility and anonymity of web-based interventions. They ‘have the potential to combine the tailored approach of face-to-face interventions with the scalability of public health interventions that have low marginal costs per additional user’. So why, in the past seven years, have online and app-based health interventions not become the standard?

The reality is that these interventions have high dropout rates, even in clinical trials (Eysenbach, 2005), which reduces the benefits that can be derived from them (Stellefson, Chaney et al., 2013). Many of the existing digital health interventions need considerable ongoing engagement from the user to record symptoms and complete educational modules. Users report that the main reason they stop using these interventions after short periods, is because they do ‘not fit into their everyday lives’ (Anhøj & Jensen, 2004).

What’s the answer? Interventions that can be easily integrated into existing routines and ‘disappear from view’ (i.e., become part of routine activity, or largely operate in the background) may result in more sustained engagement (Murray, Treweek et al., 2010). The emerging Just-In-Time Adaptive Intervention (JITAI) approach holds promise for achieving this.

Help when it’s needed most

JITAI has been made possible by evolving mobile and sensing technologies that can be used to monitor an individual’s internal state and behaviour in real time, to detect changes and determine when an intervention is needed (Nahum-Shani, Smith et al., 2017). The individual’s changing state and context is used to adapt the support provided, aiming to deliver interventions when the person needs it the most and is most receptive (Spruijt-Metz, Wen et al., 2015). The use of passive detection of behaviours, and the delivery of interventions only when needed, reduces the burden on the user. This, in turn, increases the possibility that the user will stay engaged over in the longer term and therefore benefit from the intervention.

The low participant burden of JITAIs means they have potential to increase engagement in behaviour change interventions across a range of health conditions, but may hold particular promise for mental health interventions (Ben-Zeev, Kaiser et al., 2013). Some of the common symptoms across mood, anxiety and psychotic disorders (e.g., major depression and bipolar disorder) are challenges with fatigue and concentration (Segal, 2010). These symptoms may make it particularly challenging for those with mental health conditions to remain engaged with interventions that require daily or sustained input to be beneficial. Therefore, this patient group may find JITAIs more manageable in the long term, as the cognitive load of engagement is lower, particularly if they employ passive detection of behaviours rather than active logging of symptoms by the user. They also only require attention from the user for a short time when the intervention is needed, as opposed to long and cognitively taxing stints engaging in psychoeducation modules.

Sensing mood and anxiety

A combination of sensing technologies and machine learning offers promise for passive detection of behaviours and states associated with mental health symptoms (Garcia-Ceja, Riegler et al., 2018). Sensors have been found to be effective for detecting episodes of low mood and elevated anxiety while they are occurring (Garcia-Ceja, Riegler et al., 2018). For example, Miranda and colleagues (2014) found that anxious episodes could be detected in people with social anxiety disorder using a wearable heartrate monitor – higher heart was associated with anxious episodes (Miranda, Calderón et al., 2014). Accelerometer sensors (Garcia-Ceja, Osmani et al., 2016), and smartphone touchscreen interactions (Carneiro, Castillo et al., 2012) have been used to detect stress.

A study among psychiatric inpatients in Austria found that depressive and manic episodes in bipolar patients could be detected using movement tracking, voice analysis and phone records (Grünerbl, Muaremi et al., 2014; Osmani 2015). Manic episodes were characterised by physical hyperactivity (identified using accelerometers and GPS), rapid talking (identified using voice recognition software), and increase in frequency of telephone contact with others. Depressive episodes were characterised by slower movement and fewer locations travelled to, slower speech, and fewer telephone calls (Grünerbl, Muaremi et al., 2014; Osmani, 2015). This aligns with research by Alghowinem and colleagues (2013), who found that temporal features of speech could be used to detect participants with depression.

Sensing trouble ahead

Sensing data and machine learning are even being used to predict imminent symptoms before the patient is aware of them (Garcia-Ceja, Riegler et al., 2018). A 2013 study from Nakamura and colleagues found that by analysing patterns of change in movement, an activity monitor could predict transitions from manic to depressive episodes 10 days before they occurred.

This type of sensing technology can be used in two ways within a JITAI: 1) using the passive detection of a symptom change to initiate an intervention (which could be either reactive or proactive), or 2) learning to associate symptom changes with an environmental trigger, and initiating the intervention when the trigger occurs (before the symptom change is detected).

In the first case, reactive JITAI would deliver the intervention as the symptom state was occurring (e.g. when an anxious or depressed state was detected the user could be provided with pre-selected coping strategies, such as breathing exercises or directions to a local drop-in support service.) A proactive JITAI would use early markers of a symptom change to trigger an intervention before the user may be aware that a change is occurring. For example, people with bipolar disorder could be provided with an early warning of manic and depressive episodes (Nakamura, Kim et al., 2013), which could prompt them to seek support before the episode occurs, or put strategies in place to mitigate against unhelpful behaviours associated with the episode.

In the second case, GPS signals and mapping data could, for example, be used to detect a situation or location that has previously triggered a symptom change. The JITAI can deliver the intervention when the person enters the location or situation, rather than waiting for the symptom change to be detected. However, this mode of delivery has a higher risk of false positives, which might become off-putting to the user.

What people with depression want from digital health

For those with depression, the challenge of using digital interventions comes not only from the cognitive load of engaging, but also the need to constantly reflect on their mood. This issue was raised in a co-design workshop our group conducted, undertaken to explore what people with depression need from a digital intervention.

Attendees talked about the potential to become obsessional about mood scores, and feeling like a failure if scores were low or not improving over time. They felt that if good support was not provided, it could be easy for feedback on mood to cause rumination and worry. They also reflected that daily logging of mood is challenging when they did not want to think about depression.

Attendees also talked about the importance of focusing on positive things that help them, rather than negative things that affect them. JITAIs avoid the need for constant reflection on mood changes by using mobile and sensing technologies to monitor mood changes.

Attendees also raised issues about ‘digital balance’ – the importance of maintaining relationships with others, rather than disappearing into the digital space. JITAIs require minimal input, which means the user is not required to spend as much time in the digital space.

Attendees also wanted a digital intervention to learn who they are, and reflect their individual needs and experiences, providing them with support when they were going through a challenging time. JITAIs can learn the individual’s behaviours, when they need support, and the type of support that works best for them.

Finally, attendees also talked about the importance of being provided with information about local resources and groups that could help. GPS on JITAI users’ telephones could be used to support the identification of local support services.

A promising solution

JITAIs are a promising, low-burden solution for a range of health conditions, but may be particularly beneficial for those with mental health conditions. The use of sensing technologies and delivery of an intervention just when it is needed, means that the amount of engagement required by the user is lower than in traditional digital health interventions. For people with mental health conditions struggling with fatigue and issues concentration, the reduced cognitive load of JITAIs may result in prolonged engagement and therefore a greater benefit being gained from the intervention.

- Sophie Turnbull is in the School of Psychological sciences and Population Health Sciences, University of Bristol. [email protected]

References

Alghowinem, S. (2013). From joyous to clinically depressed: Mood detection using multimodal analysis of a person's appearance and speech. Humaine Association Conference on Affective Computing and Intelligent Interaction, IEEE.

Anhøj, J. & Jensen, A.H. (2004). Using the internet for life style changes in diet and physical activity: A feasibility study. Journal of Medical Internet Research, 6(3), e28.

Ben-Zeev, D., Kaiser, S.M., Brenner, C.J. et al. (2013). Development and usability testing of FOCUS: A smartphone system for self-management of schizophrenia. Psychiatric Rehabilitation Journal, 36(4), 289-296.

Carneiro, D., Castillo, J.C., Novais, P. et al. (2012). Multimodal behavioral analysis for non-invasive stress detection. Expert Systdms with Applications, 39(18), 13376-13389.

Eysenbach, G. (2005). The Law of Attrition. Journal of Medical Internet Research, 7(1), e11.

Garcia-Ceja, E., Osmani, V., & Mayora, O. (2016). Automatic stress detection in working environments from smartphones' accelerometer data: A first step. IEEE Journal of Biomedical and Health Informatics, 20(4), 1053-1060.

Garcia-Ceja, E., Riegler, M., Nordgreen, T. et al. (2018). Mental health monitoring with multimodal sensing and machine learning: A survey. Pervasive and Mobile Computing, 51, 1-26.

Grünerbl, A., Muaremi, A., Osmani, V. et al. (2014). Smartphone-based recognition of states and state changes in bipolar disorder patients. IEEE Journal of Biomedical and Health Informatics, 19(1), 140-148.

Miranda, D., Calderón, M., & Favela, J. (2014). Anxiety detection using wearable monitoring. Proceedings of the 5th Mexican Conference on Human-Computer Interaction, ACM.

Murray, E. (2012). Web-based interventions for behavior change and self-management: potential, pitfalls, and progress. Medicine 2.0, 1(2), e3.

Murray, E., Treweek, S., Pope, C. et al. (2010). Normalisation process theory: a framework for developing, evaluating and implementing complex interventions. BMC Medicine 8(1), 63.

Nahum-Shani, I., Smith, S.N., Spring, B.J. et al. (2017). Just-in-time adaptive interventions (JITAIs) in mobile health: Key components and design principles for ongoing health behavior support. Annals of Behavioral Medicine, 52(6), 446-462.

Nakamura, T., Kim, J., Sasaki, T. et al. (2013). Intermittent locomotor dynamics and its transitions in bipolar disorder. 22nd International Conference on Noise and Fluctuations (ICNF), IEEE.

Osmani, V. (2015). Smartphones in mental health: Detecting depressive and manic episodes. IEEE Pervasive Computing, 14(3), 10-13.

Segal, D. L. (2010). Diagnostic and Statistical Manual of Mental Disorders (DSM-IV-TR). The Corsini Encyclopedia of Psychology, 1-3.

Spruijt-Metz, D., Wen, C. K., O’Reilly, G. (2015). Innovations in the use of interactive technology to support weight management. Current Obesity Reports, 4(4), 510-519.

Stellefson, M., Chaney, B., Barry, A.E. et al. (2013). Web 2.0 chronic disease self-management for older adults: A systematic review. Journal of Medical Internet Research, 15(2), e35.