Heart failure (HF) represents a growing social burden that significantly impacts the quality of life of individual patients and imposes escalating costs on healthcare systems (1). Recent data in the US show that an estimated 6.2 million people have HF, but the projections are worrisome since it is expected that by 2030 more than 8 million people will have this condition (2). The rise in HF incidence is not a problem unique to the US; globally, the absolute number of HF cases increased by 91% between 1990 and 2017 and reached 64.3 million (3).

Although the introduction of new drugs and devices led to significant progress in the treatment of HF in recent years, the mortality in HF patients remains high, reaching up to 65% at 5 years after diagnosis (4, 5). One reason for this is that only a small proportion of patients with HF achieve optimal doses of recommended HF therapy (6), and treatment adherence remains a persistent challenge in this population (7). In addition, HF typically evolves by bouts of hemodynamic deterioration, triggered by a vast number of factors leading to frequent and high-mortality hospitalizations (8). Therefore, the traditional in-clinic monitoring of HF patients is often inadequate to capture early signs of deterioration in time. Given that the earliest detectable changes in physiological measurements might occur prior to the onset of symptoms (6), timely detection of impending crisis is crucial to avoid hospitalization and premature death.

The widespread adoption of mobile technologies offers an opportunity for implementing solutions that can capture the early signs of cardiac decompression and provide timely intervention. Non-invasive remote monitoring and two-way, real-time communication between the clinic and home-based patients (9, 10), are already being tested in pilot trials in the HF population.

In the literature, these solutions are alternately referred to as telehealth, telemedicine, eHealth, mHealth, and several other creative names, reflecting lack of consensus on a proper terminology in the field. This review will adopt the term “virtual healthcare” (VHC), broadly defined as the remote delivery of healthcare via connected devices, mobile phones or tablets, and related internet technologies. The efficacy of VHC interventions is still in the early stages of investigations, with several studies reporting encouraging results in HF populations (11, 12), although questions regarding the comparative effectiveness of specific devices remain open (13).

While defining the objectives of this review, the authors noted that one neglected research topic is the challenge related to upscaling of the results of small-size clinical trials into real-world settings. In contrast to established procedures of real-world drug and medical device development, approval, and distribution, widespread utilization of VHC solutions poses additional challenges. The logistics of VHC deployment among the patients and caregivers require a much more active role of patients as well as an increased level of cooperation and shared decision making among healthcare professionals. In addition, the large amount of data gathered by mobile devices necessitate new approaches to data acquisition, storage, and analysis. Here, the adaptation of artificial intelligence algorithms (AI) will be necessary not only to process the real-time data inflow but also to aid in advanced diagnostics, personalization, and decision-making. These issues are largely neglected in existing clinical trials, where the focus is mainly on the comparative efficacy and utility of VHC interventions.

This review surveys the current state of VHC in HF with a focus on the practical aspect of VHC implementations and its challenges in real-world situations, summarizing existing evidence on methodology, efficacy, and integration into routine clinical settings.

Standard methodology for conducting and reporting systematic reviews recommended by the Cochrane Collaboration's Handbook for Systematic Reviews of Interventions was adapted to conduct a literature review (14). The review surveyed VHC solutions in patients with HF with the research objective to describe and characterize the landscape of evidence of the past 5 years (2015–2020) of the approved and in-development virtual or mobile healthcare solutions, defined as a care team + connected devices + a digital solution (e.g., a smartphone app and/or wearable devices). The results were then summarized in a narrative form.

Using predefined search strategies, Embase was searched via the Ovid platform from inception until December 6th, 2020 (Supplementary Table S1). All abstracts identified by the search were reviewed by a single reviewer according to predefined, PICOS-framed eligibility criteria (Supplementary Table S2). All studies identified as eligible during title/abstract screening were then screened at the full-text stage by two reviewers. A Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) diagram was generated for complete transparency and reproducibility of the search and screening process (15).

A standardized data extraction table was generated to define the study characteristics, patient characteristics, intervention characteristics, and outcomes that were extracted from eligible studies. Quality control procedures were undertaken during data extraction to verify the accuracy and completeness of each collected data point.

An additional search targeting studies with keywords related to scaling up VHC solutions in HF within the real-world setting was performed on September 27th, 2021 to supplement the evidence base (Supplementary Table S3). The need for additional search arose post hoc after summarizing the main body of evidence. Evidence from this search was incorporated narratively into the results.

One of the results of reviewing the included studies and observing shared methodological patterns was identification of three distinct paradigms of VHC interventions. These three paradigms were distinguished by the clinical objectives targeted by the VHC interventions: ○

Telemonitoring: remote monitoring of cardiac and extra-cardiac variables with regular uploading and evaluation of the data at the centre in order to detect early signs of cardiac decompensation.

A PRISMA flow diagram of the study selection procedure is presented in Supplementary Figure S1. A total of 171 studies were included for data extraction and evidence synthesis: 96 studies related to VHC efficacy, and 75 studies related to AI in HF. In addition, 15 publications were included from the search on studies scaling up VHC solutions in HF within the real-world setting (PRISMA flow diagram in Supplementary Figure S2).

Below is the summary of the results for the efficacy data set of VHC in HF. Table 1 shows the distribution of the study types for the 96 included studies.

The population sizes ranged from 10 patients (40, 41) to 3,449 patients in one study retrospectively analysing adherence to a telehealth program in US Veteran Administration centres (42). The mean population size in the included studies was 311 patients, and the median was 110 patients. This discrepancy is caused by the one outlier study, including 3,449 patients. The summary of study characteristics and population characteristics can be seen in supplementary materials (Supplementary Figures S3, S4).

The inclusion criteria for patient enrolment in most of the studies were based on the New York Heart Association (NYHA) functional classification (43). Seventy out of the 96 included studies used this classification for patient inclusion. Twenty-one of the 96 included studies enrolled only patients with reduced ejection fraction defined as less or equal to 40%. Forty studies enrolled a mixture of patients with both reduced and preserved ejection fractions. Thirty-five studies did not include information about ejection fraction (EF) status among the patients. Nineteen studies reported average EF of the enrolled population ranging from the minimum average EF of 21.5% (44) to maximum of 58% (45) with overall mean of 34.1%. Four studies enrolled patients with wearable cardioverter defibrillators which were used for remote data acquisition (44, 46–48).

The three VHC categories (telemonitoring, RPM, self-care) were used to compare clinical efficacy of the interventions. Table 2 shows the total number of studies and the number of RCTs within each of these categories. Both the telemonitoring and remote care studies used some form of remote data capture and upload. The types of patient data captured varied across the studies. The most commonly monitored data were self-reported patient symptoms (40, 49, 74, 75, 124), physical activity measured by accelerometers (18, 46, 125–127), body weight, blood pressure, and heart rate (50, 76–79), data transmitted by cardioverter defibrillators (44, 46–48, 76), and others.

The technologies used in VHC solutions are relatively new, and consequently, their fit into the existing models of HF patient care is still being explored in a variety of settings. When considering the results reported in the reviewed studies, the challenge is in translating them into the real-world clinical practice. Unfortunately, few studies addressed the issue of scalability of the VHC solutions and potential issues that may impede their acceptance. In order to carefully assess the available pathways towards the wide deployment of VHC in HF populations one need to start by looking at what works in the context of clinical trials and then try to identify potential roadblocks in the scaling-up those solutions.

From the standpoint of treatment efficacy, the RPM approach seems to be most effective among the three intervention categories (Figure 1). When looking at the endpoints where VHC showed the most success (rehospitalization, self-care, all-cause hospital visits), the majority of interventions used RPM paradigm. The multi-component intervention included telemonitoring devices with the daily upload of clinical measurements as well as communications with patients' general physician and cardiologist (80, 81). A similar approach integrating automated data monitoring with personalized communications was found in a majority of the studies (see Table 2). The RPM intervention could be as simple as a tool to encourage HF patients to work collaboratively with their clinicians to “make one positive change” in medication regimen (97), or a fully integrated system including a multidisciplinary care team consisting of a nurse coordinator, cardiologist, psychiatrist, and primary care physician, home telemonitoring, patient self-management support, and screening and treatment for comorbid depression (74). RPM was also used to interactively guide physical exercise in HF patients (94, 98) or to organize post-discharge remote visits and consultations (45).

The telemonitoring paradigm relies mostly on automatic data upload with limited patient involvement. Direct medical intervention is only triggered when the uploaded data cross predefined threshold signalling and impending crisis. The data upload can contain clinical variables such as lung fluid measures (16), medication taking (60, 62), vital signs (including, in some cases ECG) (50–52, 63, 64, 99), and self-reporting (55). A special case of remote monitoring concerns wearable cardioverter defibrillators, where the intervention is in part automated and implemented in the software of the wearable device (44, 46–48, 76). Historically, this paradigm pioneered the use of wearable devices and the concept of remote monitoring, however it is currently being integrated as part of the RPM model.

The concept of self-care in VHC includes a variety of approaches to patient empowerment, such as improving self-reported symptoms via mHealth apps (131), engaging the elderly patient population in technology-guided self-care (17, 124, 126, 132), and increasing self-awareness when managing patients' condition (130, 133).

There are several caveats to be considered when trying to generalize the results of the reviewed clinical trials into general practice. Most published studies are conducted in a controlled environment of research hospitals with relatively small groups of HF patients. The efficacy claims of the VHC interventions are valid within the pre-selected HF populations and limited follow-up of the studies. How will these results translate into a standard of care that can be applied to a variety of clinical settings is an open question. Even within the context of controlled studies, some results cast doubt on the efficacy of VHC interventions. For example, VHC interventions seem to have no effect on reducing all-cause mortality. Out of the five RCTs investigating this outcome none achieved significant improvement. Reasons are currently unclear and this issue merits further investigation. An interesting hypothesis to consider is the importance of direct human-to-human interactions aided by technology as reflected in the RPM paradigm. The number of significant results achieved by integrated solutions as compared to purely technology-based interventions may be, in part, explained by the presence of the human component. The technology is best utilized in support of shared decision-making by providing timely information and opening new channels for communications between patients and their care team. The main challenge going forward is to maintain this advantage in large-scale settings of existing healthcare systems.

With a few exceptions, the study durations did not extend beyond a 1-year follow-up. Given that the average lifespan of patients discharged from hospital with HF is 5.5 years (164), extending up to 19.5 years in a younger, low-risk population, the lack of long-term studies in this field represents an unmet need. In addition, HF patient populations span generations of patients starting from middle-aged “Gen-X”, through baby-boomers, to geriatric patients (see Supplementary Figure S4). Technology acceptance varies greatly among generations, hence a personalized approach is needed to achieve desired utilization. Here the studies analysing user experiences and barriers to acceptance are particularly valuable. For example, Woo et al. 2018 analysed factors facilitating/hindering the acceptance of VHC services in HF patients (165). Using a modified Unified Theory of Acceptance Use of Technology, the authors identified several factors associated with HF patients' initiation or use of VHC services in a home setting, some of which are relatively straightforward to implement (e.g., high-risk drugs education by visiting nurses).

One technological challenge to the general adaptation of VHC solutions is the large amount of clinical data the devices can collect and store. Particularly in the context of general use, this can be overwhelming, and clinicians will need assistance with the analysis and interpretation of the patient data. A promising development is the introduction of AI algorithms targeting the diagnosis, phenotyping, and prognosis in HF. The utility of using AI in diagnosing HF may be in aiding the clinician in deciding ambiguous cases (i.e., being a supporting factor to the clinical judgement of the physician). For example, an existing algorithm using a simple wristband achieved 74% accuracy (151) in diagnosing chronic HF (present/absent) from data of cardiology patients undergoing bedside monitoring. More sophisticated algorithms using better data sources (e.g., ECG) may be able to achieve greater diagnostic accuracy. Similarly, algorithms detecting sub-groups of HF patients using a collection of remote data (152) can identify clusters of patients that display different patterns of comorbidities and healthcare outcomes. This information can be then used to tailor the VHC intervention to address specific challenges facing the identified sub-groups. The limitations of AI are mostly in the academic nature of the algorithms with limited availability of these technologies where they are mostly needed—in real-world practice. To achieve a larger utilization, several barriers need to be overcome, such as gaining the trust in the algorithms by clinicians (by illustrating the physiology behind the algorithms), integrating the algorithms with the current workflow, resolving problems with privacy and data sharing, and providing adequate technical support in the field.

The topic of VHC in HF has been subject to several recent systematic literature reviews (6, 8, 162, 166). In this review, the intent was to highlight the comparative efficacy of the three identified categories of interventions: telemonitoring, RPM, and self-care. While summarizing the evidence, the limitations of the included clinical studies prompted additional research, including the questions of AI utilization and the real-world scaling of the results reported by the published studies. One of the findings in this review was that studies leveraging the synergy between AI and VHC in the HF population are limited in numbers. This seems like a potential area for future research. In particular, the AI algorithms can bring expert-level decision support for diagnostics and risk prediction to non-specialist general practitioners and facilitate the scaling up of the VHC solutions into real-world settings. A large-scale study of a distributed AI algorithms aiding clinicians in HF diagnostics and risk evaluations across multitude of real-world clinics could expose the barriers and challenges to AI acceptance.

Additional promising areas of future research include the apparent lack of efficacy of VHC intervention in reducing all-cause mortality in the HF population, long-term adherence and efficacy of VHC with follow-up spanning several years, and age-specific approach to technology adaptation with focus on the elderly HF patients. Additionally, given the relative novelty of these interventions and lack of experience in utilizing them in wider clinical settings, focus on education and transfer of knowledge may help in improving their acceptance and consequently their efficacy. This is particularly true in developing countries where the distances are large, and the number of healthcare professionals are limited.

The main finding of this review is that the promise of VHC in HF has been established in some studies but not all interventions reached their desired outcomes. The solutions with the most consistent positive impact on clinical outcomes are those that address already existing unmet needs such as early detection of warning signs of cardiac decompensation, easy patient access to the healthcare team, convenient access to relevant educational materials, and encouragement of healthy lifestyle. On the other hand, reducing mortality seems to be quite challenging and little progress has been achieved in boosting patient adherence. Artificial intelligence has a potential in interpretation of large data sets, but it is largely confined to academic institutions. Overall, while research studies administered by high-end academic institutions show some success, scaling the VHC intervention to general clinical practice remains a challenge.