Danius T. Michaelides’s research while affiliated with University of Southampton and other places

Push notifications offer a promising strategy for enhancing engagement with smartphone-based health interventions. Intelligent sensor-driven machine learning models may improve the timeliness of notifications by adapting delivery to a user’s current context (e.g. location). This exploratory mixed-methods study examined the potential impact of timing and frequency on notification response and usage of Healthy Mind, a smartphone-based stress management intervention. 77 participants were randomised to use one of three versions of Healthy Mind that provided: intelligent notifications; daily notifications within pre-defined time frames; or occasional notifications within pre-defined time frames. Notification response and Healthy Mind usage were automatically recorded. Telephone interviews explored participants’ experiences of using Healthy Mind. Participants in the intelligent and daily conditions viewed (d = .47, .44 respectively) and actioned (d = .50, .43 respectively) more notifications compared to the occasional group. Notification group had no meaningful effects on percentage of notifications viewed or usage of Healthy Mind. No meaningful differences were indicated between the intelligent and non-intelligent groups. Our findings suggest that frequent notifications may encourage greater exposure to intervention content without deterring engagement, but adaptive tailoring of notification timing does not always enhance their use. Hypotheses generated from this study require testing in future work.

Choices in the design and delivery of digital health behaviour interventions may have a direct influence on subsequent usage and engagement. Few studies have been able to make direct, detailed comparisons of differences in usage between interventions that are delivered via web or app. This study compared the usage of two versions of a digital stress management intervention, one delivered via a website (Healthy Paths) and the other delivered via an app (Healthy Mind). Design modifications were introduced within Healthy Mind to take account of reported differences in how individuals engage with websites compared to apps and mobile phones. Data were collected as part of an observational study nested within a broader exploratory trial of Healthy Mind. Objective usage of Healthy Paths and Healthy Mind were automatically recorded, including frequency and duration of logins, access to specific components within the intervention and order of page/screen visits. Usage was compared for a two week period following initial registration. In total, 381 participants completed the registration process for Healthy Paths (web) and 162 participants completed the registration process for Healthy Mind (app). App users logged in twice as often (Mdn = 2.00) as web users (Mdn = 1.00), U = 13,059.50, p ≤ 0.001, but spent half as much time (Mdn = 5.23 min) on the intervention compared to web users (Mdn = 10.52 min), U = 19,740.00, p ≤ 0.001. Visual exploration of usage patterns over time revealed that a significantly higher proportion of app users (n = 126, 82.35%) accessed both types of support available within the intervention (i.e. awareness and change-focused tools) compared to web users (n = 92, 40.17%), χ2(1, n = 382) = 66.60, p < 0.001. This study suggests that the digital platform used to deliver an intervention (i.e. web versus app) and specific design choices (e.g. navigation, length and volume of content) may be associated with differences in how the intervention content is used. Broad summative usage data (e.g. total time spent on the intervention) may mask important differences in how an intervention is used by different user groups if it is not complemented by more fine-grained analyses of usage patterns over time. Trial registration number: ISRCTN67177737.

PROV-Template is a declarative approach that enables designers and programmers to design and generate provenance compatible with the PROV standard of the World Wide Web Consortium. Designers specify the topology of the provenance to be generated by composing templates, which are provenance graphs containing variables, acting as placeholders for values. Programmers write programs that log values and package them up in sets of bindings, a data structure associating variables and values. An expansion algorithm generates instantiated provenance from templates and sets of bindings in any of the serialisation formats supported by PROV. A quantitative evaluation shows that sets of bindings have a size that is typically 40\% of that of expanded provenance templates and that the expansion algorithm is suitably tractable, operating in fractions of milliseconds for the type of templates surveyed in the article. Furthermore, the approach shows four significant software engineering benefits: separation of responsibilities, provenance maintenance, potential runtime checks and static analysis, and provenance consumption. The article gathers quantitative data and qualitative benefits descriptions from four different applications making use of PROV-Template. The system is implemented and released in the open-source library ProvToolbox for provenance processing.

Push notifications offer a promising strategy for enhancing engagement with smartphone-based health interventions. Intelligent sensor-driven machine learning models may improve the timeliness of notifications by adapting delivery to a user’s current context (e.g. location). This exploratory mixed-methods study examined the potential impact of timing and frequency on notification response and usage of Healthy Mind, a smartphone-based stress management intervention. 77 participants were randomised to use one of three versions of Healthy Mind that provided: intelligent notifications; daily notifications within pre-defined time frames; or occasional notifications within pre-defined time frames. Notification response and Healthy Mind usage were automatically recorded. Telephone interviews explored participants’ experiences of using Healthy Mind. Participants in the intelligent and daily conditions viewed (d = .47, .44 respectively) and actioned (d = .50, .43 respectively) more notifications compared to the occasional group. Notification group had no meaningful effects on percentage of notifications viewed or usage of Healthy Mind. No meaningful differences were indicated between the intelligent and non-intelligent groups. Our findings suggest that frequent notifications may encourage greater exposure to intervention content without deterring engagement, but adaptive tailoring of notification timing does not always enhance their use. Hypotheses generated from this study require testing in future work. Trial registration number: ISRCTN67177737

In this paper, we propose a representation for PROV in JSON-LD, the JSON format for Linked Data, called PROV-JSONLD. As a JSON-based format, this provenance representation can be readily consumed by Web applications currently supporting JSON. As a Linked Data format, at the same time, it also represents provenance data in RDF using the PROV ontology. Hence, it is suitable for usages in both the Web and the Semantic Web.

The use of statistical modelling by researchers in all disciplines is growing in prominence. There is an increase in the availability and complexity of data sources, and an increase in the sophistication of statistical methods that can be used. For the novice practitioner of statistical modelling it can seem like you are stuck at the bottom of a mountain, and current statistical software allows you to progress slowly up certain specific paths depending on the software used. Our aim in the Stat-JR package is to assist practitioners in making their initial steps up the mountain, but also to cater for more advanced practitioners who have already journeyed high up the path, but want to assist their novice colleagues in making their ascent as well. One issue with complex statistical modelling is that using the latest techniques can involve having to learn new pieces of software. This is a little like taking a particular path up a mountain with one piece of software, spotting a nearby area of interest on the mountainside (e.g. a different type of statistical model), and then having to descend again and take another path, with another piece of software, all the way up again to eventually get there, when ideally you’d just jump across! In Stat-JR we aim to circumvent this problem via our interoperability features so that the same user interface can sit on top of several software packages thus removing the need to learn multiple packages. To aid understanding, the interface will allow the curious user to look at the syntax files for each package to learn directly how each package fits their specific problem. To complete the picture, the final group of users to be targeted by Stat-JR is the statistical algorithm writers. These individuals are experts at creating new algorithms for fitting new models, or better algorithms for existing models, and can be viewed as sitting high on the peaks with limited links to the applied researchers who might benefit from their expertise. Stat-JR will build links by incorporating tools to allow this group to connect their algorithmic code to the interface through template-writing, and hence allow it to be exposed to practitioners. They can also share their code with other algorithm developers, and compare their algorithms with other algorithms for the same problem. A template is a pre-specified form that has to be completed for each task: some run models, others plot graphs, or provide summary statistics; we supply a number of commonly used templates and advanced users can use their own – see the Advanced User’s Guide. It is the use of templates that allows a building block, modular approach to analysis and model specification. At the outset it is worth stressing that there a number of other features of the software that should persuade you to adopt it, in addition to interoperability. The first is flexibility – it is possible to fit a very large and growing number of different types of model. Second, we have paid particular attention to speed of estimation and therefore in comparison tests, we have found that the package compares well with alternatives. Third it is possible to embed the software’s templates inside an e-book which is exceedingly helpful for training and learning, and also for replication. Fourth, it provides a very powerful, yet easy to use environment for accessing state-of-the-art Markov Chain Monte Carlo procedures for calculating model estimates and functions of model estimates, via eStat engine. The eStat engine is a newly-developed estimation engine with the advantage of being transparent in that all the algebra, and even the program code, is available for inspection.

We present a technique to capture retrospective provenance across a number of tools in a statistical software suite. Our goal is to facilitate portability of processes between the tools to enhance usability and to support reproducibility. We describe an intermediate notation to aid runtime capture of provenance and demonstrate conversion to an executable and editable workflow. The notation is amenable to conversion to PROV via a template expansion mechanism. We discuss the impact on our system of recording this intermediate notation in terms of runtime performance and also the benefits it brings.

Whilst a large range of valuable training resources are available to those interested in learning quantitative techniques, few discuss how statistics are conducted in practice by working analysts. Advances in technology, however, have opened up the possibility of using more interactive tools to develop such resources. We have conducted interviews with quantitative researchers from a variety of disciplines, and are collaborating with them to produce interactive eBooks based on a case study each of them have chosen. These are written using our Stat-JR package which can interoperate with a variety of other statistical software, and can thus allow users to explore how a given analysis might be conducted in a range of different packages. The resulting eBooks will form a library of case studies that those newer to the field can use as a learning tool, with the aim of elucidating and demystifying the quantitative research process.

Attrition is a significant problem in web-based interventions. Consequently, research aims to identify the relation between web usage and benefit from such interventions. We have developed a visualisation tool that enables researchers to more easily examine large data sets on intervention usage that can be difficult to make sense of using traditional descriptive or statistical techniques alone. Objectives: This paper demonstrates how the visualisation tool was used to explore patterns in participants’ use of a web-based weight management intervention (POWeR: Positive Online Weight Reduction). We also demonstrate how the visualisation tool can be used to inform subsequent statistical analyses of the association between usage patterns, participant characteristics, and intervention outcome. Methods: The visualisation tool was used to analyse data from 132 participants who had accessed at least one session of the POWeR intervention. Results: There was a drop in usage of optional sessions after participants had accessed the initial, core POWeR sessions, but many users nevertheless continued to complete goal and weight review. POWeR tools relating to the food diary and steps diary were re-used most often. Differences in participant characteristics and usage of other intervention components were identified between participants who did and did not choose to access optional POWeR sessions (in addition to the initial core sessions) or re-use the food and steps diary. Re-use of the steps diary and the getting support tools was associated with greater weight loss. Conclusions: The visualisation tool provided a quick and efficient method for exploring patterns of web usage, which enabled further analyses of whether different usage patterns were associated with participant characteristics or differences in intervention outcome. Further usage of visualisation techniques is recommended in order to 1) make sense of large data sets more quickly and efficiently, 2) determine the likely active ingredients in web-based interventions, and thereby enhance the benefit they may provide and 3) inform (re-)design of future interventions to promote greater use and engagement by enabling users to easily access valued intervention content/tools.