The study generated a rich and heterogeneous qualitative dataset comprising several hundred pages of interviews, ethnographic field notes, email exchanges and extracts from documents. Below, we present how the trial was operationalised differently in each setting, leading to different impacts on the care pathway. We then describe the experiences of patients and health professionals and consider specific material, technical and clinical issues in the trial.
Differences in how the intervention was implemented across study sites
Study sites joined the trial with different existing service models, staffing levels, technical and clinical capabilities, and distribution of professional responsibilities. Different local demographics, histories, cultures and past experiences with telehealth and other technology projects (Table 2) led to variability in trial implementation.
In Site A (the main ‘hub’) some of the patients initially recruited by the CCM lead research nurse were also receiving support from a nurse-led community heart failure service. Due to professional tensions, the clinical teams eventually arrived at a policy of keeping trial participants separate from this community-based service and recruiting patients from areas with less intensive community-based services. Site D also had a nurse-led community heart failure service; these HFSNs who appeared less interested in the trial and made few referrals (in a few cases, they actively resisted cooperation). But the hospital-based team in site D actively recruited patients to SUPPORT-HF, partly because they saw remote monitoring as a solution to long geographical distances.
In Site B, neither community HFSNs nor GPs recruited patients or particiopated actively in the study (probably because of a negative experience with a previous telehealth trial); the hospital research nurse and principal site investigator (a consultant cardiologist) took on responsibility for recruitment and prescribing. In Site F, hospital-based HFSNs had a large number of complex or unstable patients which they deemed unsuitable for the trial (given lack of active management in the control arm) and decided to refer only the more straightforward patients whom they would normally have referred to community HFSNs. In this site, the trial served as an alternative rather than an add-on to the usual care pathway, which may also explain why recruitment stalled part-way through the trial.
In site C, there was no consultant involvement; the community HFSN team actively recruited patients whom they were about to discharge to their GPs for follow-up, because they saw the intervention as a useful supplement to existing care. Although some of these community HFSNs continued to see patients who were participants in the trial, no interprofessional conflicts were described with the CCM team (unlike in site A). In sites E and G, there was no specialist community nursing to augment GP management of heart failure; SUPPORT-HF2 was welcomed by local clinicians as it aligned well with pressing service needs. In Site E, the principal site investigator (consultant cardiologist) deliberately sought to recruit patients in localities with substantial pressures on hospital outpatient appointments and lack of community HFSNs. In Site G, the hospital outpatient clinic seized the opportunity to “hand patients over” to the trial due to lack of resources for specialist care post-discharge.
“It fits in really nicely with the existing infrastructure. We need a way of surveilling all those being up-titrated. And to monitor the sicker ones such as those receiving home IV diuretics. So we can use it for disease management but also to see progression, and see when they’re falling off their perch.” Consultant, PI, Site E [SUPPStaff11]
This is consistent with the SUPPORT-HF2 protocol, which hypothesised the value of digital health interventions to be higher “in contexts where quality of care is (on average) suboptimal with substantial unwarranted variability at the provider-level.” [8, p.62]. The different experiences across sites, however, highlight the challenges of running an RCT alongside existing standard care that may be historically embedded and involve staff with varying degrees of buy-in to the research.
Impact of SUPPORT-HF2 on the care pathway
The rationale behind SUPPORT-HF2 includes the importance of a tight feedback loop in deteriorating patients. This was sometimes achieved successfully, with local input from site staff via the SUPPORT-HF2 dashboard and active intervention by staff at the central support unit (see Figure 1).
Sometimes, however, the feedback loop did not work as intended, mainly due to challenges in communicating recommendations to GPs, establishing whether these had been actioned (either by GPs or by patients themselves), and confirming timeframes for up-titration. The complexity of the wider system (for example, the unavoidable use of traditional ‘snail mail’ letters with GPs who at the time used neither email nor faxes) caused considerable delays.
The issue of professional responsibility for patients across the care pathway, especially in terms of nursing care, was a recurring finding. The SUPPORT-HF study team envisaged that the intervention would run alongside and complement usual care and not challenge the work of local heart failure teams, and the trial protocol assumed that data generated by the decision support dashboard and the advice given by the CCM would be uncontested and unproblematic. In reality, clinical disagreements sometimes led to conflicts between nurses from different teams and required time-consuming efforts to resolve differences and negotiate professional boundaries:
“What we have all found is that co-managing doesn’t work. Because the numbers are saying one thing but we have visually seen the patients and we may have known those patients for a long time, know their complexities and when the SUPPORT-HF team give advice we sometimes disagree with some of the decisions. We’ve had to unpick that, it’s very complex. It meant a lot of sitting in a back room making calls” Community HFSN, Site A [FACEHFSN3]
These conflicts appeared to stem from two things: an implicit knowledge hierarchy in which the technology’s decision support system appeared to override their personal knowledge of the patient; and the fact that patients in the control arm were sent a generic alert but no specific support.
Principal investigators in the study sites (who were all cardiologists) did not perceive role conflicts with the central remote support team, with the one exception of the cardiologist in site B, who vetted recommendations from the central support team. He said he knew the patients personally and could take account of exceptions, and he felt the GPs (whose views on telehealth had been influenced by adverse past experiences) would have greater confidence in implementing recommendations from him than from the CCM.
Patient experiences of SUPPORT-HF
The patients we interviewed (sometimes together with family members) came from different socio-economic backgrounds, from rural and urban settings, had different experiences living with heart failure and other medical and life-related complexities, and had varied views on the technology. As trial participants, they were not necessarily representative of heart failure patients in general. With those caveats, our data suggest that many patients used the SUPPORT-HF2 tablet devices without too much difficulty and saw great value in monitoring their condition, especially in terms of gaining reassurance and legitimising help-seeking when they needed clinical care.
Patients generally valued the SUPPORT-HF tablet device and regretted having to hand it back at the end of the trial. But a few described it as “intrusive” and found it impossible to get into the routine of regular monitoring; a few were skeptical about the clinical value of the technology (“they don't actually see the patient, they're just based on figures”: interview 13, heart failure patient and wife, Site C). Some patients felt they could not rely on the machine and assumed a responsibility for confirming their data had been received by the central support team and their medicines were listed correctly.
The care model assumed that users would largely remain static in their homes where they would have good access to the devices on an everyday basis. More active patients (especially those who travelled away from home) found this too restrictive, and some withdrew from the study. Others saw the technology as a constant reminder of their ill health. For instance, one man with severe fluid retention awaiting a heart transplant could not bear the anxiety engendered by a data trend indicating weight increase.
A few patients, especially sicker ones, found regular monitoring physically burdensome; some were physically unable to use the weighing scales (one was an amputee; another had difficulty with balance). Many did not grasp that the main purpose of weighing was to monitor fluid overload rather than weight gain; they needed support from study staff to interpret the readings and adjust medications.
Staff gave examples of patients who (for various reasons including comorbidity, anxiety or wider life issues) declined to participate in the trial, withdrew after enrolment, or expressed relief when the trial ended. Such cases were relatively rare, though withdrawal rates varied across sites.
Staff experiences of SUPPORT-HF
Despite some challenges with implementing the intervention (see above), most staff expressed positive views on telehealth technology enabling remote support in heart failure care. HFSNs felt the technology was well-designed, easy to use, useful and fulfilling for patients, whom they felt had benefited from both the technology and the day-to-day input from the central support team.
There were, however, tensions in the way different professional groups made sense of the potential of the technology in heart failure care. For example, some consultants thought that the reluctance of HFSNs to engage with the technology stemmed from a belief that the technology would replace their jobs. The nurses themselves had a more nuanced view. They worried that if routine uptitration in relatively stable patients was carried out by a telehealth service, they would lose the valued reference point of the straightforward, treatment-responsive patient and be left with a case load of unrelentingly complex and unstable cases.
Another concern was that the structured and algorithmic element of the telehealth intervention would miss important aspects of quality care. Technology, HFSNs felt, provided narrow and decontextualized information and could not replace the nuanced and holistic assessments that experienced clinicians undertook on their patients – including, for example, home visits which gave them rich information about the patient’s the environment and allowed them to observe how the patients approached activities of daily living and medication management. The Chief Investigator, however, rejected the nurses’ characterization of the SUPPORT-HF2 intervention as crudely algorithmic. On the contrary, he argued, the intervention design recognised and accommodated the need for human input and judgement where necessary, while attempting to streamline redundancy and repetition.
Material and technical issues
Interviewees raised a number of practical issues with the technology. Some patients needed extra support to feel confident with it. When technical difficulties emerged (quite commonly with WiFi or Bluetooth connection, but also due to malfunctioning equipment) input from the trial team and local nurses was often crucial for addressing issues quickly, though patients sometimes found solutions themselves.
It was necessary to visit every participant at home, usually on two occasions, to set up the equipment and show the patient how to use it. This was a time-consuming process (typically two hours for the first visit and one for the second), though it is not clear how much of this time was spent setting up the equipment and how much on requirements specific to the clinical trial.
In one site (B), half of all potentially eligible patients could not be randomised due to lack of broadband coverage in the area where they lived. This was not a problem in other sites, though sometimes synchronisation of readings via Bluetooth between the monitor and the tablet was slow because of fluctuation in signal, leading to problems with data transmission; this usually resolved if a Wi-Fi connection could be established. These transmission issues sometimes led to patient confusion as they could see that the data had not been ‘sent’.
Research nurses used a workaround to help manage these synchronization issues. They gave participants a separate phone number to use if the messaging facility of the tablet was not working, and if they had not received any data or messages from a participant for a day or two, they would phone them to check that all was well. In most cases, such problems were due to patients not connecting properly to their own Wi-Fi. Occasionally the CCM nurse was contacted to help troubleshoot.
“Another person had somehow changed the setting so it was now in a foreign language. And we couldn’t read it to correct it. Had to get [CCM research nurse] involved and she talked us through screen by screen.” Research administrator, Site E [SUPPStaff15]
The trial allowed for some iterative changes in the technology. Most of these were made to facilitate data collection for research, though the bioengineer’s input was occasionally sought to make adjustments, aiming to improve the clinical care of participants. As software changes were resource-intensive and costly, this seems to have constrained the number of co-designed modifications.
Clinical safety issues
One of the SUPPORT-HF2 trial’s secondary objectives was investigating the clinical safety of technology-supported medicines management as captured by significant event reporting and the trial’s stated secondary objective of a composite clinical safety endpoint [8]. Our interviewees described some safety issues. In the case of one patient, the system did not flag up a clinically important weight increase because it did not occur rapidly enough (i.e. over a 3-day period - the conventional assessment time frame). The central support team worked with IT colleagues to adjust the system to change the alert in this case, but such changes were labour-intensive and hence expensive and unscalable. The CCM Lead Nurse commented that further refinement of the dashboard, enabling safety alert flags to be easily individualised, could improve the safety of the technology.
It was occasionally possible to personalise clinical management without modifying the technology. For example, one patient had profound and symptomatic Parkinson’s Disease-related postural hypotension. At the GP’s request, the CCM nurse asked the patient to follow each seated BP with a standing BP; from the paired readings, it was possible for her to discern the standing BP and base her up-titration directions on that value. In this example, it was the GP’s knowledge of patient, and the GP’s initiative in contacting CCM nurse that led to this safety-critical workaround. The CCM nurse, who was an expert in heart failure not Parkinson’s disease, was unaware of the level of clinical concern. Notably, the patient did not disclose their Parkinson’s Disease to the research team.
As anticipated (and as stated in the SUPPORT-HF2 protocol), critically unstable or end-stage heart failure patients were generally deemed unsuitable for management through the telehealth technology. HFSNs described needing to actively look after these patients, including co-ordination with a complex set of additional services, depending on the case. As the patients themselves observed, being severely unwell with the complications of heart failure reduced their ability to carry out the necessary tasks. In most sites, clinical staff excluded such patients from the study:
Site E, however, in the absence of community structures to support heart failure patients, deemed SUPPORT-HF2 as filling a gap in service provision, and deliberately included “patients who needed closer attention” (PI, Site E [SUPPStaff11]). Variations in the definition of instability and difficulties identifying when patients are ‘end-stage’ combined with perceived deficiencies in usual care may have prompted creative interpretation of this exclusion criterion.
In relation to safety, our results illustrate, on the one hand, the potential for the SUPPORT-HF2 intervention to ‘tighten the feedback loop’ and detect clinically important deterioration promptly. On the other hand, our findings also illustrate how the physical effects of heart failure (or its comorbidities) can create a safety challenge when patients are simply not well enough to self-monitor or engage effectively with the system. Staff in the different sites balanced practical and logistical issues to achieve the safest use of SUPPORT-HF2 in the circumstances.