This week's Hatching Ideas journal club focused on articles by Torous et al. (2015) [i] and Olff (2015) [ii], both of which discussed the unique methodological potential, and challenges, of mobile mental health research.
Your smartphone continues to learn more and more about you. It knows who you talk to, and for how long. It knows how long people take to reply to you, and it knows where you are at all times. It has also begun to learn more and more about your body. It can know your pulse rate, your gross motor movements and even your sleeping patterns.
This increasing sensory sophistication, along with a related proliferation of different types of data produced by the phone, confronts researchers with unique methodological challenges. For one thing, is the current gold-standard, the randomized controlled trial, which is bound by time and a pre/post-test design, appropriate for data that is continuous over time? And how does a mental health researcher grapple with obvious risks around confidentiality and privacy when dealing with a technology uniquely suited to individualized medicine and the personalized extraction of data? How can we use such data to provide better care for mental health service-users?
Much of our discussion at the Hatching Ideas journal club focused on how changes in the user’s mental state might be reflected in the use of their phone, and thus what kinds of data markers might indicate what kinds of shifts. As a simple example, a user who entered a period of clinical depression might reflect this in their calling or texting behavior, or in a sudden alteration of their movement patterns. This reasoning is related to what Torous et al. refer to as “Hidden Markov Modelling,” an approach which aims to combine various different types of data outputs to describe the user’s condition or situation. A simple example: researchers have been able to predict whether people were walking, standing or driving by combining pieces of information such as GPS-registered distance from home and the activity or inactivity of the phone’s accelerometer. As Torous points out, such modelling could be used to ascertain what the “digital signatures of mental illness” are, to predict incipient changes in mental state, and to improve the understanding of mental illness under controlled conditions.
Ultimately, this discussion led us to the problem of surveillance, and when (if ever) surveillance for the purpose of intervention is warranted. A case in point, I would be perfectly fine with tracking my young child’s whereabouts using a cellphone or similar device, because I know they are incapable of managing a return trip home if they were to become lost. On the other hand, the only ethical situation where I could imagine an adult being similarly followed would be one where they had given explicit (and revocable) consent to be monitored. This is not at all a simple matter, and raises yet more questions about autonomy, loss of control, state-specific decision-making and privacy; all of which go beyond the ambit of this article. In the case of the BEACON application, however, these questions are crucial, and must be considered in the development of any mental health smartphone application. In line with an emphasis on individual liberty and self-directed care, our solution has been to make service-user consent and control the basis of any tracking or monitoring functions. However, this said, any commitment to the autonomy of the mental health service-user must be accompanied by regular reflection on and evaluation of that autonomy. Without including the perspectives of service-users through regular surveys, qualitative interviews, and inclusion into the research team itself, I believe that we cannot justifiably claim that these commitments have been met.
[i] Torous J, Staples P, Onnela J-P. Realizing the Potential of Mobile Mental Health: New Methods for New Data in Psychiatry. Curr Psychiatry Rep. 2015 June; 17:61.
[ii] Olff M. Mobile Mental Health: A Challenging Research Agenda. Eur J Psychotraumatol. 2015; 6(1):27882.