The Extent of Engagement With Telehealth Approaches by Patients With Advanced Cancer: Systematic Review

Introduction

Background

Cancer ranks as a leading cause of death worldwide and is a leading cause of premature death in most countries [1]. For people living with advanced cancer, fluctuating unmet needs can be experienced over time with disease progression [2]. Common symptoms include pain, experienced in approximately two-thirds (66.4%) of patients with advanced disease [3], alongside breathlessness, nausea and vomiting, and fatigue [4]. Typically, individuals experience more than one symptom, with an average of 14 symptoms for those with advanced cancer [5]. Such physical symptoms often exist alongside deterioration across physical, psychological, social, spiritual, and overall quality of life (QOL) trajectories [6]. There remain gaps in supporting care delivery for patients with cancer, including barriers in health communication with health care providers, lack of care coordination, and challenges in accessing care [7].

Telehealth and telehealth interventions refer to a method in which the patient and health care professional can communicate clinical information remotely via a number of different mediums such as telephone, web-based methods, and mobile apps [8]. This method is increasingly used to deliver cancer care as it provides opportunities for efficient and flexible service delivery and enables clinicians to maintain involvement independent of the physical location of the patients or clinicians [9-12]. These characteristics have also driven their increased application to support delivery of care during the COVID-19 pandemic, enabling avoidance of direct physical contact while contributing to provision of continuous care in the community. Telephone-based approaches have been highlighted as a possible means of overcoming gaps in service delivery for patients with cancer [7], including reducing the travel required to access support services that can lead to physical, psychological, and financial stress [13,14]. Examination of telehealth approaches for patients with chronic diseases has found varying effects, with improved self-management of diabetes and reduced mortality and hospital admissions in heart failure, but these improvements have not been observed across other conditions, including cancer [8]. Emerging evidence is mixed, with a recent review that focused on all cancer stages demonstrating clinical equipoise, with no discernible difference between telehealth and usual care in improving QOL [15]. However, a recent systematic review focusing specifically on patients with advanced cancer and diverse web and technological interventions (largely providing psychosocial, self-management, and expert-guided support) found that most approaches suggested some degree of efficacy relating to QOL and psychosocial well-being [16]. However, we do not know how well people with advanced cancer engage with these interventions.

With emerging clinical validation demonstrating the potential of digital technology approaches to improve care and outcomes of patients with advanced cancer, usability must also be considered [17]. Subjective aspects of usability require a better understanding, specifically regarding user satisfaction and engagement [17]. Patient engagement can be an important factor in the success of health interventions, leading to better intended health outcomes for the patient and lower health care costs [18]. As such, the effectiveness of telehealth interventions in improving health outcomes is heavily dependent on patient engagement. However, patient engagement is a broad term that can cover multiple levels of how a patient interacts with an intervention. For the purposes of this review, with a focus on technology-based interventions, engagement will be used to refer to the specific quantitative measures of behavior of engagement as defined by Perski et al [19] (ie, comprising the frequency, amount, duration, depth of use, and other measures of use and interaction with a digital health intervention). A previous systematic review found that information technology platforms (eg, mobile phone devices, internet-based interventions, social media, and other web-based communication tools) can help engage patients in health care processes and motivate health behavior change [20]. However, interventions with the intention to help support patients in managing chronic conditions can be complex. There is a need to understand whether different aspects of telehealth interventions uniquely influence patient engagement, especially for patients with advanced cancer who often experience a high symptom burden and functional impairment [21]. Understanding patients’ engagement with telehealth interventions is necessary to further evaluate and refine the implementation of these emerging and promising approaches for patients with advanced cancer. Therefore, there is a need to understand how patients with advanced cancer engage with telehealth interventions and which aspects of these interventions may influence engagement.

Objectives

Past systematic reviews have sought to synthesize the evidence of telehealth interventions among patients with cancer and survivors but have not explored interventions solely intended for and tested on patients with advanced, incurable cancer [15,16]. Understanding patient engagement can help us evaluate and refine further design, development, and evaluation of telehealth approaches for people with advanced cancer. A companion review [22] explored the clinical and cost-effectiveness of the interventions on health and health system outcomes, whereas this review synthesizes the data on patient engagement with the interventions. The aims of this review are as follows: (1) to characterize the extent of behavioral engagement of people with advanced, incurable cancer with telehealth interventions and (2) to explore factors that influence engagement with telehealth interventions.

Methods

Information Sources

This systematic review was registered with PROSPERO (International Prospective Register of Systematic Reviews; CRD42018117232). A systematic review of the literature was conducted in the following databases: MEDLINE, Embase, CINAHL, PsycINFO, Cochrane Library, Web of Science, and Sociological Abstracts, with studies included from the inception of each electronic database up until December 31, 2020. No lower cutoff date was chosen as there has not been a previous review looking into engagement with telehealth interventions in this population. An example search strategy used for MEDLINE can be found in Multimedia Appendix 1 and includes keywords and medical subject headings. The development of the search strategy was supported by information specialists at the University of Leeds. This search was supplemented by forward and backward citation searching of key papers. This review was reported in line with the PRISMA (Preferred Reporting Items for Systematic Reviews and Meta-Analyses) 2020 guidelines. The Centre for Reviews and Dissemination guidelines directed our process for conducting this systematic review and the decisions made [23].

Eligibility Criteria

Studies were eligible for inclusion in the review if the following applied:

1. They involved a telehealth intervention, which is defined as “any intervention in which clinical information is transferred remotely between patient and health care provider, regardless of the technology used to record or transmit the information” [8]. This could include symptom measuring or monitoring (eg, Patient-Reported Outcome Measures); education, information giving, and support, including decision aids and advanced care planning; psychological interventions; or medical consultation (telemedicine or teleconsultation). Participants could be located anywhere as long as the intervention that was carried out conformed to the telehealth definition.
2. They included participants of any age who were living with cancer of any type that could not be cured (advanced, metastatic, or terminal). This included people who had been treated with curative intent but whose cancer had recurred or progressed, those not being treated with curative intent, and those at or near end of life.
3. They included a measure of engagement as an outcome or reported as part of the study findings. In this review, we used the measures conceptualized as behavior that were identified by Perski et al [19]: frequency, amount, duration, and depth of use.
4. The studies were carried out in any country at any time.
5. Risk of bias was not used as a selection criterion for inclusion in the review.

Studies were excluded when the following applied:

1. The participants included patients with cancer currently being treated with curative intent, and the studies had mixed populations (ie, not 100% of the sample were people with cancer that could not be cured), unless findings pertaining to our population of interest were presented separately in the results section.
2. The studies did not report primary data (eg, systematic reviews, study protocols, conference abstracts, editorials, and commentaries).
3. The studies were not in the English language.

Study Selection and Data Collection Process

In total, 2 authors (WG and MA) reviewed titles, abstracts, and full-text papers, assessing them for eligibility independently. Any disagreements were resolved through discussion.

Data from the included studies were extracted into a predesigned form by WG and verified by MA to capture study characteristics (design, sample size, cancer type, gender, age, and outcomes). Data were also extracted based upon the items included in the Template for Intervention Description and Replication checklist (why, what, who provided, how, where, when and how much, tailoring, modifications, and how well) [24].

Quality Assessment

The included studies were assessed for methodological quality and risk of bias independently by 2 authors (WG and MA), with any disagreements resolved through discussion. The risk of bias for randomized controlled trials (RCTs) and nonrandomized studies was assessed using the Mixed Methods Appraisal Tool [25].

Data Synthesis

A narrative synthesis [26] was conducted to outline the design, population, and context (mode of delivery, health care provider, and intervention intensity) of the individual studies. Studies were categorized by their approach to examining intervention effect, differentiating between those exploring pure intervention effect (eg, using blinded RCT designs) and those exploring effect in the context of routine health care [27]. For the primary outcome of engagement, a deductive and inductive approach was taken using the definitions of engagement behavior outlined by Perski et al [19] while also ensuring that other engagement-related data were captured. Engagement data were identified and split into categories based upon the type of engagement the studies measured: frequency (how often contact was made with the intervention over a specified period), the amount or breadth (the total length of each intervention contact), duration (the period over which participants were exposed to an intervention), and depth (variety of content used) [19]. Across these 4 measures, studies were grouped together based upon how they measured the outcome, which was then summarized.

Data from the included studies relating to frequency of use by patients, where reported, were transformed to a percentage of actual patient engagement compared with intended engagement with the intervention to provide a standardized statistical comparison. When overall engagement percentages were calculated, these were weighted by sample size.

To draw associations between the calculated percentage of actual patient engagement, the intensity of the intervention (for the patient and health professional), and mode of delivery, we had to simplify these characteristics. The intensity of the interventions for both the patient and the health professional was coded by a member of the research team (WG). WG reviewed the intervention description in each included study to determine the expected engagement with the intervention for patients and health professionals. This referred to any interaction (both scheduled and unscheduled) that was anticipated or planned with the intervention (eg, a patient having a telephone consultation with a health professional or submitting data via a web-based system). For articles where a second opinion was requested by WG, a second reviewer (MA) discussed the study with WG until a consensus was achieved on the expected engagement reported. The expected engagement was simplified into categories of high, medium, and low expected engagement to make comparisons across studies. For patients, low expected engagement referred to only having ≤3 contacts with the intervention, a medium level of engagement was 4 to 7 expected contacts, and a high level of engagement was ≥8 expected contacts or more than daily reporting of symptoms. A previous study of engagement with a web-based mindfulness intervention identified similar levels of high and low participant engagement (low: 0-4 and high: 5-7); however, a third category was added for this review to account for the studies with >7 contacts [28]. For health professionals, the categories mirrored those for patients if the health professional was required to make contact with the patient (eg, low was ≤3 contacts, medium was 4 to 7 contacts, and high was ≥8 contacts). If the health professional was required to only make contact with the patient when prompted to do so by a patient’s entry on a system or survey, it was coded as low contact on the part of the health professional. For each intervention, we also coded the mode of delivery (eg, telephone, smartphone, or web-based), including interventions where multiple modes were used. We were then able to look at associations between the mode of delivery, expected level of engagement (high, medium, or low for the patient and health professional), and the percentage of actual patient engagement with the intervention.

Results

Search Results

Of the 19,676 papers that were identified in the database search, 0.2% (40/19,676) of papers covering 39 different studies were eligible for inclusion in the systematic review [29-68]. Figure 1 outlines the PRISMA flow diagram for the included studies and the reasons for exclusion of studies.

‎View this figure

Study Characteristics

Table 1 includes a summary of the characteristics of the included studies. Table 2 outlines the characteristics of the included interventions and the engagement outcomes. The included studies had a sample size ranging from 6 [61] to 766 [31] and included multiple RCTs (16/39, 41%) [30,31,33-35,37-39,43-45,48,50,57,63,67,68], with most studies being conducted in the United States (22/39, 56%) [29-31,33,34,37-41,43,45,46,48,50,54,57,59,60,66-68]. Of the 39 studies included in the review, 13 (33%) explored intervention effects in the context of routine care implementation [29,32,36,41,42,49,51,52,58,60-62,64], with the remainder exploring intervention effects often using a blinded controlled trial design.

Engagement

The engagement outcomes for all studies are outlined in Table 2.

Frequency

Across most studies (36/39, 92%), the frequency of times contact was made with the intervention was reported [29-43,45-47,49-58,60-68]. There was substantial heterogeneity in the measurement of frequency across studies. Of the 39 studies, 13 (33%) reported the percentage of contacts either with the whole intervention or with each individual intended session [30,31,35,40,48-50,54,56,63,65,66,68]. The number of contacts with the intervention overall or each individual session was reported by 69% (27/39) of the studies [29,33,34,36-39,41-43,45,46,49,51-55,57,58,60-67].

Across 44% (17/39) of studies, it was possible to create a standardized percentage of actual patient engagement compared with intended engagement [29,30,32,33,38-40,42,45,46, 51,53,54,56,63,66-68]. This ranged from 51% [53] to 100% [29,30,64], with an average across all 17 studies of 75.4% (SD 15.8%). In the remaining 49% (19/39) of studies, it was not possible to create this standardized statistic because of a lack of reported data, and the design of the intervention meant there was no intended engagement and it was instead tailored to the patients’ needs.

Amount

A total of 31% (12/39) of studies measured the amount of contact with each intervention or with the intervention overall [32,35,36,38,39,43,44,46,47,50,53,57,58]. Of the 39 studies, 3 (8%) measured the average amount of time of each intervention contact (10.5 to 85 minutes) [35,47,57], 2 (5%) reported the average amount of time across all intervention contacts (16 to 65 minutes) [38,39,46] and 1 (3%) reported the total amount of call durations, which could be averaged across all intervention participants to 35.3 minutes [44]. In total, 5% (2/39) of studies reported the median amount of time for each intervention contact (10 to 20 minutes) [32,58], and 5% (2/39) of studies reported the median amount of time across the whole intervention (38 to 146 minutes) [50,53]. A total of 3% (1/39) of studies reported the number of intervention contacts that fell into a range of minutes (eg, 16-30 minutes: 58 contacts) [36]. A total of 3% (1/39) of studies did not report time but, as it was a web-based intervention with communication with the health professional through posts on a message board, instead reported the average length of each post at 260.5 words [43].

Duration

A total of 15% (6/39) of studies that had open-ended interventions reported the length of time that each participant was exposed to the intervention [43,52,56-59]. A total of 10% (4/39) of studies reported the average time of exposure to the intervention, ranging from 62 to 195 days [43,52,57,59]. A total of 3% (1/39) of studies reported a median amount of exposure to the intervention of 6 months [58], and the final study (1/39, 3%) reported the number of participants exposed for >4 weeks (n=5) and >12 weeks (n=1) [56].

Depth

A total of 10% (4/39) of studies reported on the variety of components of the intervention that the participants accessed [36,43,50,64]. Each study measured depth in different ways. A total of 3% (1/39) of studies reported the percentage of time that each health professional was on the teleconference calls [64], and another study (1/39, 3%) simply reported that 75% of patients had completed all elements [43]. The number of different interventions that all participants received was reported by 3% (1/39) of studies [36], and the final study (1/39, 3%) reported that patients had viewed a median of 243 webpages [50].

Association With Intervention Level of Intensity

Expected levels of engagement for both patients and health professionals were reported across low (≤3 contacts), medium (4-7 contacts), and high (≥8 contacts) categories. A total of 13% (5/39) of studies could not be categorized as there was no expected engagement with the intervention, and the extent of engagement was determined at the patient’s discretion [32,36,43,55,63]. Table 3 shows the number of studies with the expected interaction of both the patient and health professional with the intervention. Most studies expected a similar level of interaction from both the patient and health professional in an intervention, but no studies expected more interaction from the health professional than from the patient.

Figures 2 and 3 are graphical representations of the association between expected levels of engagement for the patient (Figure 2) and the health professional (Figure 3) and the percentage of actual engagement with the intervention by the patient. Figure 2 shows that the studies that had low expected engagement for the patients had a combined actual patient engagement of 64% (SD 14.8%); for medium expected engagement, this was 66.9% (SD 16.4%); and, for high expected engagement, this was 87% (SD 8.2%). Figure 3 shows that the category with the highest level of combined actual patient engagement was the studies that expected the health professionals to have a high level of engagement with the intervention (86.6%, SD 8.3%). The studies in the categories of low and medium expected engagement from health professionals had lower levels of combined actual patient engagement (71%, SD 15.2% and 62.3%, SD 15%, respectively).

‎View this figure

‎View this figure

Association With Intervention Mode of Delivery and Health Care Providers

Figure 4 [29-68] shows the modes of delivery of each intervention and where interventions use multiple modes, with the names in bold involving multiple health professionals. The figure also shows, where available, the percentage of actual patient engagement by way of color, with blue showing 90% to 100%, purple showing 70% to 89%, and red showing <70%. Of the 39 studies, 17 (44%) used multiple modes of delivery, whereas the remaining 22 (56%) used 1 mode. The telephone was the most popular mode of delivery (28/39, 72%) followed by web-based delivery of the intervention (17/39, 44%). The use of only a tablet or smartphone app for the intervention appeared to be associated with the most actual patient engagement with an intervention, with 8% (3/39) of studies showing between 90% and 100% engagement [54,56,65]. The use of a telephone was more mixed, with actual patient engagement ranging from 54.5% [51] to 100% [29,30]. Figure 4 also shows broadly how many health care providers were involved in delivering the interventions, with those involving multiple health care providers shown in bold. Those interventions that involved multiple health care providers reported higher patient engagement than those with only 1 health care provider (79.3%, SD 18.5% vs 70.5%, SD 11.5%).

‎View this figure

Study Quality

The included studies could be grouped into two broad categories to be assessed using the Mixed Methods Appraisal Tool: quantitative RCTs and quantitative nonrandomized trials. The RCTs were of a broadly high quality; however, a number of studies did not provide enough information to assess whether the randomization procedure was conducted adequately or whether the groups at baseline were comparable. There were also 15% (6/39) of studies that did not have complete outcome data at follow-up. Among the nonrandomized trials, study quality was again high, apart from the included studies that did not control for confounders in their analysis. This is likely because most of these studies were feasibility or pilot studies and were not powered to detect significance, which would have been inappropriate. A breakdown of how each study was rated can be found in Multimedia Appendix 2 [29-68].

Discussion

Principal Findings

This systematic review is the first to synthesize engagement data from telehealth interventions for people with advanced cancer. This review found that people with advanced cancer were able to successfully engage in telehealth interventions with variable types of telehealth modalities, including telephone, mobile phone–based apps, and web-based interventions, albeit largely in the context of research studies. This review found that the frequency of engagement with the intervention was the most commonly reported measure of engagement, although there was heterogeneity in the method of reporting across the studies. Where standardized comparison was possible across the studies, actual engagement as a proportion of intended engagement was at an average of 75.4% (SD 15.8%). The level of engagement was found to vary based on the expected interaction of both the patient and health care professional and the mode of delivery. Actual patient engagement was higher in studies that expected higher levels of engagement from both the patient and health care professional but was noticeably lower in studies that expected only a low or medium level of engagement. Furthermore, the use of only a tablet or smartphone app for an intervention appeared to be associated with the highest levels of actual patient engagement with an intervention. This could in part be explained by the immediacy of access and reduced steps for accessing an intervention through a mobile phone app when compared with an intervention hosted on a website.

This review is in line with previous reviews that looked at engagement with interventions involving digital technology among people with chronic diseases, which found broadly that there are high levels of engagement with interventions [20]. However, this review provides an overview and critique of existing reporting of engagement for telehealth interventions in patients with advanced cancer and found wide disparities in metrics for engagement used and reported across the included studies. The frequency of interaction with an intervention was reported widely, but other measures of engagement, such as the amount of time spent engaging with the intervention, were not reported as well. Furthermore, the duration and depth of engagement with the intervention were reported by only one-quarter of all included studies (9/39, 23%). This may be due to the design of interventions with a set duration or only 1 component that patients could engage with, but this was not clear across studies. In addition, few studies reported the expected levels of engagement for an intervention, limiting the interpretability of any subsequent reporting of actual patient engagement. Refining and using measures to better understand factors driving digital engagement, including for telehealth, could inform the development of approaches from design through monitoring as part of routine care. For example, the application of engagement measures could serve as a progression criterion in feasibility studies of emerging telehealth approaches. Future research may need to define and develop meaningful and context- and condition-appropriate measures of digital engagement for palliative care to facilitate measurement of digital engagement. Although this review focused on the quantitative measures of behavioral engagement, the future development of a measure should attempt to incorporate components that provide a broader understanding of subjective experiences and aspects of engagement, potentially through qualitative approaches. There is also scope to develop and refine the dimensions comprising the digital engagement framework used to guide the synthesis of data in this study. For example, there is scope to incorporate a temporal element to consider the intensity of the intervention (eg, whether the intervention is spread over a week or months) alongside refining the underpinning definitions of terminology used for each dimension as the framework continues to evolve.

Through this review, we can conclude that there is no standardized method to report engagement in telehealth interventions for people with advanced cancer. The frequency of interactions with the intervention was presented most commonly, although the way in which this was done varied greatly across the studies, and there is a limited ability to understand what this means in the context of the intervention and the proposed and expected engagement needed for clinical utility. For example, people with advanced cancer have fluctuating needs, and a higher level of engagement with an intervention may not relate to the success of the intervention itself but be reflective of worsening outcomes for the patient [69]. In addition, patients may have their symptom management needs met early on in the intervention and may not need further follow-up, which may not be indicative of poor engagement with the intervention per se. With regard to mobile health interventions, the Mobile Health Evaluation, Reporting and Assessment checklist has been developed to help standardize the methodology for reporting the content and context of an intervention to support reproducibility and comparison of interventions [70]. Future iterations of the tool could include, for example, reporting of the expected and actual patient engagement levels of intended users of telehealth interventions alongside frequency of use—the most widely reported measure in this review. These data could complement and contribute to emerging evidence regarding the feasibility and acceptability of telehealth approaches as part of care for people with advanced cancer.

Recent evidence suggests that digital health interventions could provide a degree of efficacy related to QOL and psychosocial well-being [16]. For this review, most included interventions focused on symptom management, with high levels of engagement that suggest potential for its use to support remote monitoring. This approach could facilitate reductions in the required number of in-person visits while enabling continued access to data to inform patient care. However, in order to ensure such an approach is sustainable, there is a need to consider the burden of data entry on patients and the need for review—and potentially response—by health professionals. For patients, emerging approaches provide options for enhancing the richness of data received through remote monitoring without increasing the data burden for patients. For example, wearable technologies can passively collect sensor data on heart rate and activity to inform automatic monitoring and feedback processes [71], augmenting existing approaches without increasing the need for manual data entry. For health professionals, this review found that studies with high levels of intended engagement for both the patient and health care professional were associated with higher levels of actual engagement on the part of the patient. High intended engagement from health care professionals may not be a sustainable approach for digital technology, particularly when considered alongside the additional invisible work that such digital health can create for health professionals (eg, data must be interpreted, made sense of, located within existing knowledge and data sets, and negotiated) [72]. This is important to consider in light of projections of an increasing burden of serious health-related suffering and subsequent demands on palliative care services across geographical regions where demand is increasingly outstripping supply [73,74]. Therefore, for telehealth approaches to be sustainable as part of care for people with advanced cancer, they should seek to balance demands on both the patient and the care team, seeking to achieve maximal information with minimal data burden.

Limitations

There were a number of limitations associated with this review. First, the focus of engagement in this review was on the behavioral aspects that were outlined by Perski et al [19] but not on the subjective measures of engagement, such as interest, attention, and enjoyment. Integrating these subjective measures into a future mixed methods review could allow us to evaluate the experience of interventions. In addition, because of the heterogeneity of the studies and reported approaches to measuring engagement, such as frequency, it is difficult to determine exactly which components of interventions contribute to higher engagement levels. We were only able to draw associations, and future research is needed to better explore causal factors. Furthermore, although this review looked at the extent of engagement, how it was measured across studies, and the association with the study characteristics, we did not assess whether engagement led to an improvement in patient-reported outcomes or experience. A future review should consider how engagement interacts with patient-reported outcomes. In addition, when determining the categories for low, medium, and high expected engagement, we did not take into account the time frame of the intervention; therefore, 2 studies could be grouped together with different levels of intervention intensity. Furthermore, most of the studies included in this review explored the intervention effect through mostly controlled studies, which could bias the recruitment toward those individuals who were motivated and more likely to be technologically literate. The levels of engagement identified in this review may not then translate into routine clinical care if these studies and their intervention effect have to date been confined to exploration in the context of RCTs and similar study approaches. This review also limited the included studies to those written in English; therefore, this review may not contain the entirety of related studies.

Conclusions

This review identified that, where reported, there is a high level of engagement with telehealth interventions among people with advanced cancer. We identified that actual patient engagement is associated with both the expected level of engagement of the patient and the health professional as well as the mode of delivery of the intervention. We highlighted the heterogeneity in the reporting of engagement results across the research and the need to improve such reporting guidelines. As treatment delivery becomes increasingly more dependent on remote or telehealth modalities, the inclusion of a measure of engagement in future telehealth evaluations is essential to enable the comparisons of interaction and use across intervention approaches and to provide further granularity in factors that determine optimal implementation of telehealth approaches. There is a need for consistent measurement and reporting of domains relating to digital engagement (eg, breadth, duration, and frequency) with the scope to amend or develop measures. This will increase the ease of reporting of engagement in future studies, inform which telehealth intervention components are linked to variations in engagement, facilitate evidence syntheses, and support the development of condition-specific benchmarks of digital engagement for people with advanced cancer.