JMIR Publications

ABSTRACT

Background:

Smartphone usage is widely spreading in the society. Functions and sensors embedded in these devices may play an important role in healthcare monitoring and intervention planning. However, the usage of smartphone as a relevant tool for intrapersonal continuous behavioural and physical monitoring is not yet fully supported by adequate studies addressing technical reliability and acceptance. However, the usage of smartphone-based sensing platform has been widely investigated on clinical side to cope with the need of unobtrusive and continuous data collection while reducing biases in the patients’ behaviour.

Objective:

The objective of this paper is to identify and discuss technical issues that may impact on the wide usage of smartphones as clinical monitoring tools. The main focus is on the quality of the acquired data and the transparency of the data acquisition process to the user daily life.

Methods:

QuantifyMyPerson (QMP) is a complete platform for continuous acquisition of smartphone usage features and embedded sensors data. The platform consists of a mobile app running on the smartphone for data acquisition, a backend cloud server for data storing and processing and a web-based dashboard for users' management and data visualization. The data processing is aimed at the extraction of meaningful features for the description of the daily life according to the existing literature. These features include phone switching on and off, calls, apps usage, GPS and accelerometer data. 12 health subjects installed on their personal smartphones and run the mobile app for and overall period of 7 months. The acquired data were analysed to assess their quality in terms of impact on the daily life of the smartphone (i.e. battery consumption and anomalies in functioning) as well as data integrity. Data integrity was computed as the ratio of app running time and total smartphone working time. Relevance of the selected features in describing changes in daily life was assessed through the computation of a k-NN Global Anomaly Score aimed at detecting days that "differ" from the others with the aim of using the platform to trigger potential modification in daily lifestyle.

Results:

The effectiveness of smartphone-based continuous monitoring depends on the acceptability and interoperability of the system to increase users' retention and on integrity of the acquired data. Acceptability was confirmed by the full transparency of the app once installed and the absence of any anomaly in smartphone usage when the mobile app is running. The only negative issue pointed out in acceptability was the battery consumption. Even if the trend of battery drain with and without mobile app running was comparable, the users complain about this. Referring interoperability, the app was successfully installed and run on several brands and models of Android-based smartphones. The study shows that some smartphone manufacturers implement power saving policies that do not allow a continuous acquisition of data from embedded sensors so impacting on data integrity. Data integrity was 96% on smartphones whose power saving policies have no impacts on the embedded sensors management and 84% overall.

Conclusions:

The results of the testing pointed out that the two main issues that impact on spreading of continuous behavioural and physical monitoring in free living conditions, i.e. battery consumption and power saving policies of manufacturers, may be overcome. Referring to battery consumption, it is mainly due to GPS triangulation and may be limited. Brand-related operative system policies still have the most important effects on the data integrity due to the observed fragmentation of the Android services. Referring the power saving policies of some manufacturing, the embedded sensors are re-activated by any event happening on the smartphone, the missing data demonstrated to be related to periods of non-usage of the phone in which the activity data would have been anyway null. According to the results of this study, smartphone-based passive sensing techniques and platforms seem to be fully feasible and scalable on wider pool of users despite the strong Android policies fragmentation.