Assessing the feasibility of using smartwatches for low burden, high temporal density capture of longitudinal health data.

Longitudinal health data enable researchers and clinicians to explore the causes, progression and treatments of diseases in ways that are not possible with cross-sectional data. Prospective longitudinal cohort studies enable these to be studied at the population level, while N-of-1 studies enable their exploration in individuals. In both cases, there is a trade-off between how frequently data are recorded, and the burden to the participant: higher temporal density provides greater insights into detailed patterns of health but places a higher burden on participants.

For some time, researchers have used smartphones for low-cost, high temporal density data capture. Smartphone-based ecological momentary assessment (EMA) prompts participants several times a day in free-living conditions to answer questions about their health. This approach reduces recall biases and errors, increases ecological validity, and can provide very high temporal density data. However, when capturing high temporal density data, smartphone EMA can be highly disruptive. While we perceive smartphones as always nearby, on average smartphones are only within reach ~50% of the time1. This means participants need to disengage from their current tasks to respond. Also, as the smartphone is not necessarily on the body, audible rather than less distracting haptic prompts are required. These disruptions lead to poor compliance rates2, and are problematic in longitudinal cohort studies as increased burden is a concern for participant engagement3.

Smartwatches enable us to retain the benefits of EMA, while reducing disruption. As they are worn on the wrist, they are never beyond reach, so time taken to access the device is significantly reduced. Less intrusive haptic prompts can also be used. The recent development of microinteraction-based EMA (uEMA) goes further in reducing disruption: requests are reduced to a single question with a limited set of answers that can be responded to with a single tap. Recent pilot testing4 found that when these elements were combined to form smartwatch-based uEMA, compared with smartphone-base EMA over a 4-week period, the smartwatch system had better compliance (82% v 64%), completion (92% v 67%), and lower levels of disruption (38% v 53%).
While this pilot work is supportive of the wider use of smartwatch-based data collection, and the potential for these techniques in public health research has been acknowledged5, they have yet to be used in this context.

The feasibility work proposed in this fellowship is essential in enabling the uptake of these methods, helping researchers understand; how these methods perform over longer periods, whether specific groups are more likely than others to find them acceptable, whether their feasibility varies by the type of health data collected, and the feasibility of combining multiple health data collection approaches in a single smartwatch device. The fellowship will also explore how best to analyse the data capture with these methods, which could make a step change in understanding the evolution of health and disease, enabling better causal understanding, earlier intervention and ultimately health improvements. The nature of this work, exploring how new digital technologies can be used to capture health data, aligns strongly with MRC and HDR UK priorities in the areas of Digital Technologies and Informatics for Health.

From a personal development perspective, the fellowship will provide the time and resources necessary for me to focus on building my skills and profile in the field of digital phenotyping, moving me closer to my goal of becoming a fully independent researcher leading a digital phenotyping team with an international reputation for excellence.

References
1.Dey,A,...(2011).UbiComp:163-172.
2.Courvoisier,D...(2012).Psychol Assess,24:713-20.
3.Lucas,P,...(2013).BMC Med Res Method,13:56
4.Intille,S,...(2016).UbiComp:1124-1128
5.China,G,...(2017).CHI-EA: 2767-277

Technical Summary of Activities

1.Develop smartwatch-based micro Ecological Momentary Assessment (uEMA) application (months 1-6 & 13-18, objectives A,B,C).
This application will support all feasibility testing in the project. In addition to various self-report measures it will capture metrics for assessment of feasibility, including completion, compliance, and disruption.

2.Test feasibility of smartwatch-based uEMA in ALSPAC (months 7-30, objective A).
The uEMA application will measure the Alcohol Urges Questionnaire in 20 participants from the ALSPAC study 6 times a day over 3 months. Completion, compliance, and self-reported disruption will be assessed, and in-depth interviews will highlight positive and negative aspects of the system. Differences in these by gender and socioeconomic background will be examined.

3.Test feasibility of combining smartwatch-based uEMA with passive detection of behaviour (months 19-30, objective B).
The uEMA application will record mood and cigarette craving in 20 smokers 6 times a day over 3 months. Participants will be recruited from the Bristol Tobacco & Alcohol Research Group participant database. On the same smartwatch, the StopWatch system for passive detection of cigarette smoking (developed by applicant) will record details of cigarettes smoked a day. Completion, compliance, and disruption will again be assessed, and interviews will highlight experiences of the combined system.

4.Explore feasibility of combining smartwatch-based uEMA with cognitive testing (months 30-36, objective C).
Working with collaborators Cambridge Cognition, the technical and regulatory aspects of combining cognitive assessment methods with other forms of measurement on smartwatches will be explored.

5.Develop analytical approaches (months 1-36, objective D).
Appropriate methods will be developed for analysing longitudinal health data captured using Smartwatches, calling on time series analysis and multilevel modeling.