A systematic review of lifestyle monitoring technologies

Introduction

In lifestyle monitoring, sensors are installed in the home to monitor behaviour in order to gain an understanding of ‘normal’ activity so that any unusual changes over time can be recognised and responded to. For example, a sudden change in mobility around the home may indicate that the subject has fallen, while a gradual decline in the time spent in the kitchen may suggest illness or malnutrition. In addition, some applications of lifestyle monitoring have been used as assessment tools, i.e. installed for a short period of time to provide data to assist the care assessment process. ¹

The underlying assumptions of lifestyle monitoring, namely that it is possible to determine an individual's health or care status by the remote monitoring of behavioural characteristics, were the focus of the study undertaken by Celler et al. ² in 1995. They concluded that an individual's health status could be determined by monitoring relatively simple variables relating to the interaction between the individual and their local environment. They reported that some 50% of people had undiagnosed medical problems that could be detected by home monitoring. Subsequently the Anchor Trust and British Telecom developed a system to detect changes in a user's lifestyle. ^3,4 The final conclusions of these studies were that the system was generally acceptable, that it enhanced feelings of safety and security, that it increased care choices, and that it supported and enhanced the carer's role.

Despite the growing interest in all aspects of telecare, the evidence base remains relatively weak, especially for lifestyle monitoring, even though there are significant numbers of commercial installations around the world. We therefore conducted a literature review to summarize the current position with regard to lifestyle monitoring based on sensors in the home.

Methods

The review considered articles dealing with lifestyle monitoring available on the electronic databases OvidSP (and including Medline, CINAHL, PsychInfo and British Nursing Index) and INSPEC using the search structures and terms described in Table 1. All articles, including conference proceedings, published in the English language between January 1990 (5 years before the Celler et al. study which was one of the first of its type) and December 2009 were included.

The focus of the review was on home-based lifestyle monitoring systems using the following exclusion criteria:

Articles evaluating user views only;

Two reviewers independently applied these criteria to the papers identified. In the case of any disagreement, a third, independent reviewer made the final decision on which papers were included. The exclusion criteria were then applied again on the full papers and, where appropriate, data extracted. A number of papers described distributed sensors for behavioural monitoring in association with vital signs monitoring, video cameras and body-worn sensors and these were retained in the review.

Results

The database search identified a total of 1994 articles. Of these, 1420 were associated with OvidSP and its subsidiary databases and 574 with INSPEC. On review, it was established that there were a total of 159 duplicates between the databases, giving a total of 1835 unique articles.

The initial screening of these 1835 articles resulted in a total of 133 potentially relevant articles, with the two reviewers agreeing on 79 of them (59%). The remaining 54 articles identified by only one of the reviewers were then independently evaluated by a third reviewer and it was decided that 28 of these articles should be included. This resulted in a total of 107 articles for detailed review of the full texts. Of these, a further 15 were rejected on the basis of the full text while 18 proved to be unobtainable even when the authors were contacted by mail and email, leaving a total of 74 articles to be subjected to full review. The distribution of these 74 articles by year is shown in Figure 1.

OPEN IN VIEWER

Figure 1

Number of publications identified for review and their year of publication. The broken line is the least-squares regression

Of the 74 full articles reviewed, only four ^5–8 were concerned with trials involving more than 20 subjects (Table 2), whilst a further 21 papers reported trials with fewer than 20 subjects. The distribution of reported trials is shown in Figure 2. The remaining 49 papers described only technology development without reporting on any formal user trials.

OPEN IN VIEWER

Figure 2

Numbers of subjects

Motion detection was the most common of the technologies employed, see Table 3. For studies which included other monitoring approaches in addition to behavioural monitoring, the sensor types are summarized in Table 4. The number of studies employing associated monitoring strategies is shown in Table 5, while Table 6 shows additional information associated with the deployment of the monitoring approaches of Table 4.

The information analysis and artificial intelligence methods employed are summarized in Table 7. There is increasing investigation into such techniques. However, on the basis of the papers reviewed, the techniques are not yet robust enough for operational use and performance data is still lacking.

Discussion

The majority of articles provided only an overview of the methods that were applied along with, where appropriate, brief details of the evaluations being undertaken. It is therefore difficult to obtain a complete understanding of what systems and methods have been deployed and in what context. There is a need for all involved in the reporting of research and development findings in this area to provide more information about what they are seeking to achieve and the associated methodologies.

Figure 1 suggests that there has been a steady increase in publications on lifestyle monitoring in the home since 1997. The first paper included in the review was from 1997, confirming that Celler et al.'s study in 1995 was one of the earliest of this type. Only 25 papers reported trials involving one or more individuals, while 49 papers reported on technical development in the absence of trials. This provides only weak formal evidence to support the use of lifestyle monitoring, despite numerous commercial installations around the world. Where evaluation with users has taken place, studies have not yet addressed the clinical and cost effectiveness of the intervention when compared to conventional care delivery. Large scale and longitudinal studies would be helpful, such as that being undertaken by Kaye who has recruited 200 older people, installed lifestyle monitoring systems in their homes, and is gathering data over a 30 month period. ⁹

Many commercial lifestyle monitoring products are now offered for short-term health and care assessments, ^10,11 where equipment is installed for a period of a few weeks and then removed. Since there may be more immediate commercial potential in short-term care assessment, researchers may find it valuable, in order to maximise the impact of their work, to focus their efforts on short-term assessments as well as long-term monitoring.

A wide range of sensor technologies have been used in association with an equally wide range of monitoring strategies (Tables 3–7). For most approaches, the most common sensor technology was the PIR for motion detection, followed by door contacts and electrical supply monitoring. These somewhat crude sensors for data capture may be appropriate for monitoring gross changes in room occupancy and so forth, but for finer analysis the resolution is probably too low. Thus when attempting to link user activity data to formal health and care assessment, researchers may need to acquire data of finer granularity to ensure that accurate assessments can be made.

It is clear from Table 5 that the predominant monitoring strategy is that of detecting changes in activity. Very little attention has been given to determining when or how changes in the profile of activity should be used to signal an important change in health or care status that merits the raising of an alert. Thus, the emphasis has been on detecting changes in the profile of activity, but not on linking this to procedures to determine health and care needs, or of triggering a re-assessment of such needs. Indeed, in many instances there appeared to be a deliberate decision that the focus should be on highlighting changes to activity profiles, and that professionals should then analyse the data and determine when and how to intervene. While this reduces the technical complexity, it may represent a barrier to uptake. Specifically, it is unlikely to be cost effective if clinicians have to spend substantial time in monitoring behaviour that subsequently does not necessitate a health or social care intervention.

Many of the articles reviewed employed information analysis, artificial intelligence and machine intelligence techniques to identify changes in patterns of behaviour. However, these were generally employed in the context of an established profile of behaviour with limited attention given to the introduction of other, long-term factors which may influence behaviour, such as seasonal variations in behaviour. It should also be noted that much of what has been reported was associated with the evaluation of the method rather than its use in an operational environment.

The present review therefore raises a number of questions:

What is the primary purpose and role of lifestyle monitoring (assessment, long-term monitoring or both)?