The role of statistics in behaviour change

Recent advances in mobile health and sensor technologies have led to increased scientific interest in the development and use of just-in-time adaptive interventions (JITAIs) [1]. JITAIs are designed to monitor the dynamics of an individual's internal state and context in real time and hold great potential to promote health behaviour change. Specifically, they use digital devices to monitor patients, relying on an algorithm that analyses the ongoing data to "decide" if a person needs an intervention, and if so, JITAIs deliver the appropriate health support (intervention) at the time and place where it is needed. The potential of JITAIs is particularly relevant to chronic disease management where timely interventions are critical to help patients stay healthier and prevent disease relapse while not weighting much on healthcare costs.

To date, JITAIs are used across many health fields, including physical activity promotion, smoking cessation support, HIV medication adherence support, and in general illness management support where we look to avoid over-treatment and provide increased levels of support to those who can benefit. Below, a toy example of a simple JITAI mechanism which uses ongoing data and contextual information to encourage physical activity through push smartphone notifications.

centered image

So far, methodology to inform the construction and evaluation of such complex interventions is still underdeveloped. The major challenge lies in building health and behaviour theories which are able to consider the heterogeneity both across individuals and across time, that is, theories more dynamic in nature. In addition, an optimal JITAI needs to consider how to keep the individuals engaged in the long-term, but at the same time avoiding over-burdening them with too many messages [2].

From a statistical perspective, designing methods to support an optimal JITAI construction needs careful consideration of the fact that: (a) response to treatment is often heterogeneous across individuals and thus, treatment decisions should be personalised at individual-level, and (b) the ‘personalisation’ happens at multiple stages of intervention; that is, the personalised decision can change over time. Data for constructing JITAIs can be obtained either from sequentially micro-randomised clinical trials or longitudinal observational studies. Observational data provide a great source of information for therapeutic evaluation but they are often highly heterogeneous, making estimates more subject to confounding bias. Furthermore, the high frequency data collected from wearable devices are fostering development of sophisticated real-time JITAI algorithms, but it is the combination of the right treatment and right delivery time to be likely most critical. Indeed, JITAIs are inherently multicomponent interventions. So, at a population level it is interesting to compare the effectiveness of the different intervention components (being either multiple message suggestions or message suggestions plus other behavioural/pharmacological interventions) with the ultimate aim to disentangle the possible interactions between components and find which component(s) are most effective. This can guide development of future JITAIs. Meta-analysis and network meta-analysis techniques will come in handy for this purpose [3].

[1] Nahum-Shani Inbal, Smith Shawna N, Spring Bonnie J, Collins Linda M, Witkiewitz Katie, Tewari Ambuj, and Murphy. Susan A 2016 Just-in-Time Adaptive Interventions (JITAIs) in mobile health: key components and design principles for ongoing health behavior support. Annals of Behavioral Medicine (2016), 1–17.

[2] Nahum-Shani, I., Shaw, S. D., Carpenter, S. M., Murphy, S. A., & Yoon, C. (2022, March 17). Engagement in Digital Interventions. American Psychologist. Advance online publication.

[3] Metelli S., Chaimani A. Disentangling effects of complex health interventions. Manuscript, 2022.