Designing and implementing telemonitoring for early detection of deterioration in chronic disease: Defining the requirements

Proposed sample

We selected participants using a purposive approach. Informatics and telehealth conferences, professional networks and snowball sampling were used to identify key opinion leaders in the following fields: health informatics and technology; policy leads in this field; health service researchers; and clinicians familiar with the exemplar conditions.

Data collection

The interview schedule addressed the following issues: which are the best indicators (signs or symptoms) for early detection of deterioration; when to initiate monitoring; how to collect data, analyse and store it; how to act on detected events (alerts); what are the patient or clinical issues with the use of such technologies; which research questions remain unanswered? The interviews were audio recorded, transcribed and summaries provided to interviewees for endorsement.

Analysis

Thematic analysis was conducted on the transcripts: inductive coding and categorisation were undertaken before the development of identifiable themes, using a constant comparison method. ¹⁴ A collaborative approach was taken: discussion amongst the research team was facilitated by outlining themes using concept maps ¹⁵ until consensus was reached regarding synthesising and summarising the data.

The requirements for early detection telemonitoring were identified by approaching the results of the thematic analysis with the following questions:

We then outlined a set of research questions by asking:

Participants

Twenty-six key informants from across the UK were interviewed (Table 1). Eighteen had a clinical background, seven of these in nursing, of which four worked as community-based nurses providing care for patients with chronic diseases. Eight informants had experience in health informatics, four were policy leads in the National Health Service (NHS) and nine had experience with recent telehealth programmes. Two informants were academics involved in medical technology development. The interviews took place between June and November 2008 and lasted approximately 90 minutes each.

OPEN IN VIEWER

Table 1.

Informant characteristics.

Category	ID	Background
Clinical	C1	Community Matron working with telemonitoring
	C2	Community Matron working with telemonitoring
	C3	Community Matron working with telemonitoring
	C4	HF Specialist Nurse introducing telemonitoring to managed care
	C5	HF Specialist Nurse, previously worked with telemonitoring, now at a policy level
	C6	GP conducting academic research in primary care
	C7	Cardiologist specialising in HF and engaged in academic research
	C8	Cardiologist specialising in HF and engaged in academic research
	C9	Diabetologist and Clinical Professor
	C10	Diabetologist engaged in academic research
	C11	Respiratory Consultant specialising in COPD and engaged in academic research
	C12	Respiratory Consultant specialising in COPD and engaged in academic research
Health Informatics	I1	Paediatrician working in HI academic research
	I2	Professor in computer science specialising in learning algorithms
	I3	HI researcher specialising in telemonitoring and health records
	I4	Biochemist working in HI academic research
	I5	Professor in computer science specialising in AI and Grid computing research
	I6	Professor in computer science specialising in data mining
Policy	P1	Nurse working with projects in chronic disease
	P2	Telecare programme director
	P3	GP working at a national policy level for health informatics
	P4	Nurse with interest in HI working at government level
Other	O1	Evaluation of telecare projects for social services
	O2	Professor of medical engineering
	O3	Anaesthetist with interest in health IT working as healthcare sector consultant
	O4	Diabetologist working in academic biomonitoring research

AI, artificial intelligence; COPD, chronic obstructive pulmonary disease; GP, general practitioner; HI, health informatics; HF, heart failure; IT, information technology.

Thematic Analysis: structure

Interviewees emphasised the complexity and diversity that characterised patients suffering from chronic conditions. It became apparent that a successful technological solution must take account of these factors in order to be clinically effective and widely adopted. Also, there were wide variations in current care practices; the analysis identified three existing models of care—traditional, managed and telemonitoring— which can be considered as evolving towards early detection monitoring.

The results of the analysis were therefore categorised as three broad themes (Figure 1): ‘Theme 1 - The Diversity of Patient and Disease Factors’, describes the characteristics of chronic diseases and the patients who suffer from them; ‘Theme 2 - Current Healthcare Models’, describes existing healthcare models and their drawbacks; and ‘Theme 3 - Early Detection Telemonitoring’, describes participants’ ideas for the characteristics of an ideal future system.

OPEN IN VIEWER

Figure 1.

Thematic analysis results

Theme 1 - Diversity of Patient and Disease Factors

The informants emphasised that patients with chronic disease have complex needs and often suffer multiple-morbidity. Many patients are elderly, and physical and cognitive health problems may pose significant challenges to the tasks of self-management. A patient’s psychological status and level of social support also affects their ability to manage or monitor their health. For example, many patients with chronic disease suffer from anxiety or depression:

“[they] become dependent very quickly because they’re frightened. They’re frightened by what’s happening to them…it’s not what they planned.” [Informant C3]

In contrast to this are those expert patients who have high levels of self-awareness and prolonged experience of effective self-management. Variability such as this, is characteristic of the patient population (e.g. wealth, health literacy, language, location) and can impact an individual’s access to healthcare and their ability to make use of it.

Patients may alternatively deny their condition, resulting in poor compliance and delay in seeking advice. Informants noted how changes in symptoms that herald deterioration can be missed, ignored or misinterpreted and that it can be difficult for patients who are constantly unwell to distinguish a gradual decline from their usual day-to-day variability.

“Top of the list of things to sort is communication…people not understanding what they were meant to do, not understanding the early warning symptoms…” [P1]

Theme 2 - Current Healthcare Models

There is a spectrum of care models currently in use for patients with chronic disease (Figure 1). ‘Traditional Care’ operates in a fragmented way and the majority of patients are managed by general practitioners (GPs). ‘Managed Care’ describes the use of case-management, typically where community nurses provide co-ordinated care to small numbers of high risk patients. ‘Telemonitoring Trials’ are implemented (in the experience of our informants) primarily as additions to managed care and often include very small numbers of selected patients.

Theme 3: Early Detection Telemonitoring

All the telemonitoring systems discussed in the interviews comprise a common set of components and relationships (Figure 2). Participants’ requirements for future systems that could effectively achieve early detection of deterioration are reported according to these.

OPEN IN VIEWER

Figure 2.

Telemonitoring model components

Sensor: usability and novel modalities

Many existing commercial physiological sensors are cumbersome and their use poses a burden to patients. Usability of telemonitoring for patients was considered to be important. Informants envisaged a future where sensors would be less visible, capable of measuring multiple parameters simultaneously, would require low power or be self-powering, be capable of on-board data storage and analysis and be able to transmit data wirelessly and securely. Such sensors would not restrict freedom of movement (‘free-living’), whereas current equipment often ties the patient to the home.

“For some of those telecare things we are talking about fixing you to your own house…in a world where people are increasingly mobile, or indeed mobility is a sign of wellbeing and independence.” [P2]

Free-living sensors are in development, including those incorporated into garments, but are currently a long way from robust, reliable and commercial application.

Given the emphasis placed on contextual knowledge by case managers, future systems should also be capable of capturing ‘soft’ parameters more effectively (symptoms, behaviour, mood, appearance) and may include detection of events that can lead to deteriorations (medication errors, infections, environmental changes).

Signals: intelligently selected indicators in reusable formats

Clinical participants had clear ideas about parameters that may provide useful early detection information in future for specific conditions: inflammatory biomarkers in COPD, B-type natriuretic peptide (BNP) or electrocardiography (ECG) analysis for heart failure, and blood glucose surrogates or lipid measurements in diabetes. Although given the complexity and variability of many patients with chronic disease there was little confidence that telemonitoring could achieve “distant diagnostics” [C8].

“There’s probably only about 10 vital signs that might be useful, might just give me a bit more clue as to how bad somebody was.” [P3]

Appropriate deterioration indicators may also vary between patients with similar conditions.

Chosen indicators appeared to be monitored because they could, rather than because they should. A consistent view was that data was often collected without a clear aim or consideration of relevance and that the questions that are important to clinicians are not sufficiently addressed.

“What we really want is a very good way that lets clinicians articulate the questions they want answered.” [I5]

Data should also be fit for its defined purpose: authenticated, of appropriate quality (e.g. signal-to-noise ratio), and with context (metadata that allows interpretation of the data). Data collected for one purpose can be reused (e.g. for audit or research) and shared with others if it conforms to a common standard (e.g. HL7 CDA, IHE XDS, EN 13606), permitting interoperability across different devices/databases and facilitating semantic interpretation.

Analysis: automation, personalisation and diagnostic signatures

Informants stressed the importance of establishing individual baselines, as patients who are chronically ill are unlikely to conform to population norms. Establishing these in complex patients would require clinicians to identify ‘wellness templates’ of data patterns.

“…at the early stages [of learning] it’s not possible to do the automation. You have to have the clinician’s judgement.” [I6]

The parameter changes (signatures) that would reliably indicate an incipient deterioration in conditions such as heart failure and COPD have not been characterised and analysis methods capable of learning individual data signatures for each patient may be required. Algorithms that use multiple parameters, statistical or pattern-matching methods are under development, but such automation is lacking in commercial systems. This is partially because of the difficulty of encoding clinical expertise. Informants sometimes referred to being ‘familiar’ with a patient, to expertise that can’t be taught and to ‘intuition’ telling them that the patient’s condition has worsened. In addition, the difficulty of achieving true interdisciplinarity and cooperation between informatics scientists, system engineers and clinicians is a significant challenge to developments in this field.

“…there’s still a disparity in, sort of, beliefs between physicians and engineers..[engineers] think they can make the round peg go through the round hole and that’s….that’s the problem solved.” [O4]

Alerts and information representation: accuracy, patient involvement and usability

Accurate analysis is insufficient by itself. Trust must be established that the analysis leads to reliable alerts and that health professionals can use it to provide appropriate responses. This is more difficult to achieve in automated systems than for human-controlled systems.

“To trust it you have to understand it. Or do the trials....and see if it works.” [I2]

The accuracy of alerts is a critically important issue for automated monitoring systems. False positives can result in increased clinical workload, unnecessary treatments and reduced trust in the telemonitoring system and should be minimised. However, there will always be situations which have not been anticipated and allowed for, but which are not clinically significant. False negatives are potentially more hazardous, whereby patients remain untreated for a real health crisis. These are handled in current systems by attempting to ensure that patients do not become dependent on the technology. Patients should understand that they are not under continuous observation and that they must not rely on the monitoring to detect and notify health problems in lieu of self-reporting.

In current telemonitoring technology, the patients are relatively passive and do not often have access to their own monitoring information. Informants suggested an alternative to this in which the data and alerts can be delivered direct to the patient, either with or without a health professional being informed. This has the advantage that it can increase patient responsibility and potentially their motivation for self-care.

Usability of hardware, analysis and outputs is key if a telemonitoring system is to find widespread use. Clinicians and patients are more likely to use a system if it is robust and intuitive to use.

“The idea is to make the system complex, but make it simple to use…” [I6]

Information should be presented in a way that is comprehensible and meaningful and does not increase workload or anxiety. Multiple levels of complexity and detail in a single system could be employed to meet the needs of multiple users. More sophisticated analysis methods would also enable hierarchies of alerts to be delivered to different actors depending on the severity or urgency of the situation.

Response: planned and integrated into care structures

Several informants believed that the first response to an alert should always be telephone contact with the patient to rule out poor measurement technique, forgetting to monitor or lifestyle factors that don’t require clinical action. Responses to alerts should be planned in advance, but clinical judgement should be able to override protocols. However, the telemonitoring model does not fit comfortably with the current structure of UK care organisations where the role of responder is not defined. Most previous and current evaluations of telemonitoring are small-scale and exist in isolation from the normal service model.

“Generally they were quite poorly integrated, unfortunately. Just within that unit where they were happening, or just within that hospital. There were no links to primary care or to social services or to anywhere else.” [I3]

However, large-scale implementation into existing services would require detailed specification of who should be responsible for receiving an alert and how to take and document the most appropriate action. Some informants indicated that it could be necessary to re-organise care systems to allow implementation of the telemonitoring model.

“I can’t emphasise enough this business about getting the service model right first…” [P3]

Some informants had concerns about the medico-legal unknowns of clinical governance and accountability for the detection pathway and patients’ welfare.

“…If I request that one of my patients has a continuous monitor the onus should be on me to ensure that that data is handled responsibly. That doesn’t necessarily mean that I should be looking at it….I don’t know how it’ll work.” [O4]

These issues will become more challenging as increased mobility of patients and sensors results in telemonitoring crossing more organisational and geographical boundaries and being distributed over many providers.

Data storage: volume, control and interoperability

Telemonitoring has the potential to create large volumes of data, particularly if continuous free-living monitoring is used. The storage requirements will additionally depend on the data quality and format – whether it is kept in its original format or aggregated. There was no consensus amongst informants whether advances in storage capacity and data compression would keep pace with increasing volume demand.

Informants with informatics backgrounds did not favour centralised records, but distributed systems are not currently effective. Many envisaged a future with linked and virtualised multiple health records with easy access and the details of implementation hidden from the user.

“I don’t care how it happens, I just want it to happen. Because it’s very frustrating and time consuming at the moment trying to ferret out all the information.” [C3]

Interoperability of stored data allows for information-sharing and re-analysis, but also raises the issue of consent for secondary purposes. Several informants voiced the opinion that patients should have control or ownership of their telemonitoring data and that there should be processes for clinicians to access it when needed. Increasing information-sharing and integration of care across organisations will require details of control, access, privacy and security of health data to be resolved.

“You don’t want to share everything… sometimes the challenge of what you think the other professional needs and what they actually want is quite difficult to identify.” [P4]

Principal findings

Early detection as a driver for introducing telemonitoring was not a concept our participants recognised. Small scale telemonitoring trials had resulted in improved staff efficiency and patient empowerment but there were concerns that these services were driven and constrained by technological developments and commercial availability rather than by clinical need. Telemonitoring systems had not overcome the drawbacks of traditional care because they existed as an off-the-shelf ‘bolt-on’ and could not be connected with many of the different care services that these patients engage with. A future early detection telemonitoring system should be designed with high levels of user input and therefore be characterised by high usability and integration with multiple services. The technical requirements include more versatility of monitoring capabilities and proven automated methods of pattern detection.

We have defined the components of a generic telemonitoring system and a set of questions that should be considered when designing or implementing such technologies. We also proposed a series of research topics which will address these questions for an early detection telemonitoring system. Suggestions for further research have been outlined, but not prescribed. For example, the analysis to automatically detect changes in patient status could use knowledge-based algorithms or machine learning methods. The important aspect is that the questions are addressed with scientific rigour where appropriate (e.g. in determining the most suitable clinical indicators) and with users’ needs as a major design factor.

Strengths and weaknesses of the study method

One strength of our study was the wide range of participants that were interviewed. Front-line clinicians, including community-based nurses and specialist consultants, were able to describe the daily challenges faced by those providing care for patients with chronic disease. Our sample also included professionals from a variety of backgrounds with current and past experience of telemonitoring projects. However, our study is necessarily limited by the experience of our participants and the nature of the interviewing methodology, i.e. the questions asked and the ideas that participants chose to express. Therefore, although the requirements and research questions appear to be comprehensive, we do not claim that they are exhaustive. In particular, we did not interview any patients, although further work has involved consultations on this topic with patient focus groups.

Our informants were based around the UK and our information is UK-specific, but this does not mean that the questions we have identified are only relevant to the UK. Most countries have fragmentation in their care systems (state- or privately-run)^16–18 and face similar demographical changes. Also, economics and migration require telemedicine solutions that cross geographical boundaries. None of the items in Table 2 are UK-specific and each can be interpreted in consideration of local or national circumstances.

The informants who had direct involvement with telemonitoring were all based in primary care and so the architectures of the telemonitoring systems they described were quite similar. Other participants had indirect experience of trials based in secondary care or university departments. Those reported in the peer-reviewed literature are more often of this type.

Results in context – published work

Other researchers have conducted work seeking to inform future research in the general area of telehealth or health informatics/e-health and there is inevitable overlap between these and our findings. Many authors have identified the organisational and social challenges to successfully implementing such technologies.^19–25 As telemonitoring does not match up with current organisational structures and has not been designed with sufficient clinical input, it is seen as a disruptive technology that forces users to change their established work practices and roles and imposes inflexible protocols, thus impeding its wider acceptance.^19–25

Clinical usefulness is one of the most important requirements for users of telemonitoring. ²³ Calls for improved usability, ease of use, fitness for purpose, etc. are relatively common in the e-health literature, ^20,21 and it is felt that these issues can be addressed by increased user involvement in the design process.^21,24,25 Our participants were consistently in favour of this approach and the gap between engineers and informaticians on the one side and clinicians on the other was blamed for many of the perceived problems with current systems. The wider literature on the design and implementation of information systems proposes the use of such paradigms as co-design, co-realisation, participatory design and computer supported co-operative work (CSCW) to bridge these gaps.^25–27 We do not propose which method should be adopted, but agree that improved definition and communication of requirements is essential for telemonitoring systems to improve their effectiveness and efficiency. In the experience of our participants many systems were effectively bought ‘off-the-shelf’ with little attempt to adapt them to local needs. In depth consideration of the questions suggested here when planning to introduce telemonitoring would be a significant improvement on such practice.

Previous work on telehealth interventions has not specifically addressed the concept of early detection of deterioration. Our findings are that the requirements for this are related to the technical issues of clinical indicators and pattern detection. There was no consensus amongst our clinical informants regarding the natural history of physiological indicators preceding symptomatic deterioration. Clinical professionals could not describe what changes in the monitoring data they expected to see, only that changes from baseline should be identified and acted upon. It appears that very little research has been conducted to demonstrate whether there are suitable physiological markers for early detection. Wolfel ²⁸ has listed several indicators that can be predictive of incipient heart failure deterioration, only one of which (weight) is currently used in commercial telemonitoring interventions. The specificity and selectivity of these in typical primary care patients is unknown. From the technical perspective we cannot predict at this point, for example, whether statistical pattern matching or machine-learning techniques will prove most effective at detecting deterioration; this is yet to be determined. Personalisation of indicators and analysis methods may well be the most effective means to improve telemonitoring outcomes. ²¹

The local primary care pilots or feasibility studies described by some informants were small-scale and lacked scientific rigour and effective evaluation There have been many published trials of different forms of telemonitoring for chronic disease management,^6,10 but there are limitations in using these as an evidence base for the design and adoption of telehealth; not least that the technology changes faster than controlled studies can be conducted and published. Published trials are often of relatively short duration and lack high quality cost-effectiveness evidence.^29,30 They also tend to conduct complex interventions, ³¹ but assess only simplistic health or resource use outcomes and lack qualitative evaluation. ²⁰ There have been calls for more theory-based interventions ³² and for improved study designs and evaluation methods for telehealth trials.^22,30,33 However, the majority of published data is from systems based in secondary care, whereas the majority of patients are managed in primary care and have complex and variable conditions.

Results in context – current situation

The need for improved communication and closer working between UK care organisations and patient-centred approaches has been accepted for some time ^34–36 and the latest five-year plan for the NHS is especially apposite:

“We will make more use of information-based technologies to design new models of care as well as improving the performance of existing services. We will integrate information around the patient, deliver relevant information at the right time to clinicians and use technology to drive efficiency for both patients and clinicians.” ³⁷

However this was published before the coalition government described its intention to reform the NHS, and the shape and effects of the Health and Social Care Bill ³⁸ on the use of such information technologies are still to be determined.

The Department of Health is currently evaluating a large-scale trial of telecare and telehealth for chronic disease management^13,39 which is due to report in 2011 and has also commissioned 16 Integrated Care Pilot projects ⁴⁰ that will address some of the organisational and informatics questions we have identified. We eagerly await the publication of these results and hope that the evaluation will be more effective and informative than most.

Organisations currently contemplating introducing a telemonitoring service can use the questions we have defined here to ensure that their requirements are well considered, explicitly defined and communicated to system providers. Although apparently simplistic, the concept of such forward planning appears to be absent from some interventions. These questions are brief, but comprehensive, and could improve the effective use of currently available technology. We call for a move away from ‘What can we do with this technology?’ approaches towards ‘What technology can help us do this?’.