Chronic heart failure (CHF) is one of the most prevalent high-cost chronic diseases affecting at least 1%-2% of the worldwide population [1]. Europe and the United States spent approximately 1%-2% of their national health care budget on this chronic disease [1,2]. Worldwide, the economic burden of CHF is estimated to be approximately US $108 billion per annum, of which US $65 billion can be attributed to direct health care costs [2]. CHF is characterized by an erratic and difficult-to-predict course. A high percentage of incurred costs are due to high readmission rates, as well as the high number of comorbidities [3].

Changes in physiological parameters such as weight, heart rate, blood pressure, and pulse oximetry may precede cardiac events. Signaling such changes through telemonitoring may enable physicians to intervene before the patient needs hospitalization or an emergency department (ED) visit [4]. Telemonitoring has been proposed as a possible strategy to tackle the challenges that CHF brings to health systems, most notably, spare use of the utilization of expensive resources. Telemonitoring has the potential to prevent hospital readmissions, thereby saving costs and improving the quality of life of these patients [5].

The technology and quality of care for patients with CHF have been evolving, which has resulted in a variety of telemonitoring programs consisting of different elements for different populations. In addition, studies on the effect of telemonitoring on health care utilization differ widely according to cointerventions, time horizons, and outcome measures. Such parameters may affect the effectiveness of telemonitoring and the comparability of studies [6]. This heterogeneity at many levels results in debates concerning the effectiveness of telemonitoring.

Studies on telemonitoring have shown mixed results with respect to health care utilization. Although some studies showed a decline, others showed no significant differences or even an increase in health care utilization. Many studies have included cointerventions within the telemonitoring intervention. Structured telephone support (STS) has often been incorporated [7-9]. As STS has been reported to reduce the number of heart failure–related hospitalizations [10], the effects found in studies that use these two interventions simultaneously may not be solely attributed to the telemonitoring program. A Cochrane study from 2015 [10] concluded that, although telemonitoring can reduce heart failure–related admissions, telemonitoring programs were not able to reduce the risk of all-cause hospitalization. A variety of systematic reviews have been performed, mostly exclusively including randomized controlled trials (RCTs). Bashi et al [11] performed an overview of systematic reviews and found that 11 of 19 systematic reviews only included RCTs. Although RCTs are considered as the gold standard for research purposes, observational studies also have merits that should not be ignored. For example, they provide a sense of the real world as opposed to experimental RCT settings [12]. Especially for diseases such as heart failure, which is known for a high degree of multimorbidity [13], RCTs may be limited by their strict inclusion and exclusion criteria. In addition, the vast majority of previous systematic reviews have only analyzed the effect of telemonitoring on hospitalizations and mortality [10,11]. Thus, limited information is available on the effect of telemonitoring programs on other outcomes, such as ED visits, length of stay, and other forms of health care utilization.

We therefore performed a systematic review on the effect of telemonitoring for CHF on health care utilization, including both RCTs and observational studies. By including a wide variety of telemonitoring programs and outcome measures, we aim to identify the various aspects and broad impact of telemonitoring programs on the health care utilization of patients with CHF.

PubMed, Web of Science, and Embase (Ovid) databases were searched. All authors were consulted for additional eligible studies. A detailed description of our search strategy for each database can be found in Multimedia Appendix 1. We excluded articles published before January 1, 2010 because technology and insights have rapidly evolved over the past 10 years. Our search was updated until August 4, 2020 according to the method described by Bramer [14]. Removal of duplications was performed according to the method described by Bramer et al [15].

Telemonitoring includes a wide variety of definitions and descriptions. Generally, telemonitoring refers to the use of telecommunication to assist in the transmission of medical information and services between health care providers and patients [16]. This definition also includes STS. Many studies do not explicitly distinguish between invasive and noninvasive telemonitoring techniques. However, this review explicitly focuses on noninvasive techniques exclusively because there are significant differences between implanted monitoring devices and noninvasive telemonitoring in terms of costs and eligibility of patients [17]. Thus, we defined telemonitoring as the noninvasive application– or web-based collection and transfer of physiological data, aimed at improving quality of life or decreasing health care utilization in patients with heart failure, or both.

We included peer-reviewed studies that reported direct or indirect measurements of health care utilization such as hospitalizations, ED visits, length of stay, days of hospitalization, visits, and health care costs. We did not apply exclusion parameters concerning study design to include all available evidence. Therefore, we included RCTs, nonrandomized trials, and studies on observational data. However, we excluded studies that were not based on original data such as Markov models because these studies are based on assumptions rather than empirical data. Studies with regular telephone support initiated by nurses or health care providers (without medical indication) were excluded to focus on the effect of telemonitoring as opposed to the combined effect of telemonitoring and STS. We excluded studies from countries not belonging to the Organization for Economic Co-operation and Development to improve homogeneity in terms of the socioeconomic characteristics of the populations. Finally, studies that were not published in either English or Dutch were included. Textbox 1 presents the inclusion and exclusion criteria.

Titles and abstracts were independently screened by 2 researchers (SLA and TEPR) for eligibility. In cases where no conclusive decision could be made, studies were included for full-text screening. Full-text screening was performed in duplicate and independently by SLA and TEPR. Disagreements were resolved by discussion until consensus was achieved or consultation with a coauthor (RBK or SAVD). Thereafter, data were extracted by one researcher (SLA) for both CHF-specific health care utilization and all-cause utilization because a reduction in CHF-specific health care utilization may not translate into an overall reduction [10]. For each study, the effect of telemonitoring on health care utilization was determined by extracting data on hospitalizations, ED visits, length of stay, days of hospitalization, visits, and health care costs. We determined the statistical significance of these results using a cut-off value of P<.05. If no P value was reported, we deemed the effect statistically significant if the reported CI did not include 1.00.

The risk of bias assessment was performed by SLA and TEPR. For the risk of bias assessment, we used the Cochrane risk-of-bias tool for randomized trials [18] and Risk of Bias in Non-randomized Studies of Interventions tools (Cochrane) [19] for RCTs and all remaining studies, respectively. The studies were assessed in duplicate and independently. Subsequently, the individual assessments were compared for possible discrepancies. When discrepancies were observed, the studies and corresponding assessments were discussed until a consensus was reached.

Our search strategy identified 3770 unique studies, which were then screened in the title and abstract. This resulted in 99 remaining studies. After the full-text screening, 29 studies were included. Figure 1 shows the flow diagram. The main reason for exclusion after full-text screening was because STS was a part of the intervention (19/70, 27%), followed by a lack of measuring and transmitting physiological parameters in the program (17/70, 24%). Most studies (11/29, 38%) originated in the United States.

Table 1 presents an overview of study and program characteristics. A detailed description of the studies and their characteristics are presented in Table 2. Telemonitoring programs showed a high degree of heterogeneity in terms of physiological variables measured as shown in Multimedia Appendix 2 [4,20-47]. All studies included weight as a parameter, whereas only 4 included electrocardiography measures as a physiological parameter [20-23]. The studies used 10 unique combinations of physiological parameters. In addition, 2 studies included other measures such as lung fluid [24] and body composition [25], which were acquired by noninvasive means. Most programs (n=16) used telemonitoring in addition to usual care; regular check-up visits were still scheduled [4,22,24,26-38]. One study [20] stated that telemonitoring was used as a substitute for these visits, and 12 studies did not elaborate on this matter. The follow-up period of the studies ranged between 1 and 89 months, with most studies having a follow-up period of 6 months. Follow-up was consistent with the duration of the telemonitoring program itself, with the exception of 2 studies that also included a follow-up after the telemonitoring program was completed [24,39].

Patient inclusion criteria for previous hospitalizations differed among the studies. If a hospital admission was required for inclusion, this was most often (n=10) within 7 days of discharge [4,23,24,29,35,37,39-42]. Inclusion was started within 1 month of index hospitalization for 2 studies [20,25], and 5 studies [21,32,43-45] included patients within 6 months. Most studies (n=12) [22,26-28,30,31,33,34,36,38,46,47] did not specify requirements regarding previous admissions for CHF or included patients from outpatient clinics. New York Heart Association (NYHA) Class II was most often reported (n=9) as the NYHA Classification with most patients [21,22,25,26,28,32,38,45,46]. NYHA Class III was reported in 8 studies as the dominant class [20,23,29,31,35,43,44,47].

We found that most RCTs (12/16, 75%) showed at least some bias concerns, with 2 RCTs having a high risk of bias [21,27]. The main reasons for some concerns in the RCTs were related to the randomization process. In addition, 75% (9/12) of nonrandomized studies were assessed as having a serious or critical risk of bias. No nonrandomized studies were assessed as having a low risk of bias, which indicates that the strength of evidence from none of the included nonrandomized studies is equivalent to a well-performed RCT [19]. Bias due to confounding was the domain that most often resulted in studies being assessed with a serious or critical risk for bias. Table 3 shows the consolidated overall bias assessments. A detailed overview of the risk of bias assessment of all studies can be found in Multimedia Appendix 3 [20-23,25-27,29,31-33,39,43,44,46,47] and Multimedia Appendix 4 [4,24,28,30,34-38,40-42,45].

The goal of this review was to identify the broad impact and various aspects of telemonitoring programs for patients with CHF and their effects on health care utilization. Most studies showed no statistically significant effects of telemonitoring on overall hospitalizations (14/21, 67%), ED visits (7/8, 88%), length of stay (4/4, 100%), and days of hospitalization (6/8, 75%). However, the remaining studies showed reductions in health care utilization for these measures. Overall, non-ED outpatient visits and health care costs were increased in 67% (6/9) of studies and 30% (3/10) of studies, respectively. The most ambiguously reported health care utilization measure was health care costs, which showed increases in 30% (3/10) of the studies, decreases in 30% (3/10) of the studies, and no differences in 40% (4/10) of the studies. Heart failure–specific health care utilization measures showed similar results as the overall health care utilization measures.

We found a high degree of clinical diversity among the interventions, in terms of physiological parameters as well as the targeted populations. Although most telemonitoring programs were targeted at patients in NYHA Classes II-III, a variety of additional inclusion criteria were used. In addition to clinical diversity, methodological heterogeneity was also high, as can be observed by the varying risk of bias assessments.

Some of the studies included in this review suggested that telemonitoring may result in lower hospital admission rates, but most studies did not report such reductions. This mixed effect is consistent with the current literature [10,48]. In addition, we found no clear associations between the extracted study characteristics and the effect of telemonitoring and a reduction in hospitalizations. However, we found that 67% (6/9) of studies reported an increase in non-ED outpatient visits when telemonitoring was used.

Thus far, the effect of telemonitoring on non-ED visits has received little attention. However, the early warning designs of telemonitoring might also attract additional use of frontline services and false-positive alarms. An overview of systematic reviews [11] showed that 2 of 19 systematic reviews included non-ED visits in their outcomes [49,50]. These systematic reviews included 5 studies, 1 [51] of which can be classified as telemonitoring based on our definition. This RCT showed a substantial increase in non-ED visits and phone calls for telemonitoring compared with usual care. Our results suggest that this may be the case for most telemonitoring programs, as we found that 67% (6/9) of studies showed an increase in non-ED visits. One study found a decrease in heart failure–related non-ED visits. This study [20] explicitly stated that “patients allocated to the intervention group were only allowed to visit the cardiologist or HF-nurse in case of an absolute need for intervention.” In addition, 83% (5/6) of studies, which found an increase in visits, had telemonitoring as an additional component in addition to usual care [22,26,37,47], and 17% (1/6) did not state this explicitly [33]. This stresses the need to treat telemonitoring as a substitute for regular care, rather than an addition to regular care, that is, if a reduction in health care utilization is the primary aim. Five studies reported an increase in non-ED visits and hospitalizations as outcome measures. Villani et al [23] found an increase in non-ED visits simultaneously with a reduction in hospitalizations. The other 4 studies [22,26,33,37] reported an increase in non-ED visits and found no effect on hospitalization. This indicates that additional visits can be a side effect of telemonitoring rather than a prerequisite for effective telemonitoring programs in terms of reduced health care utilization. Therefore, telemonitoring programs may become more cost-effective if they pay more attention to decreasing these visits.

The effect of telemonitoring on health care costs has been inconsistent across studies. Health care costs were severely affected by the costs of telemonitoring programs, which showed large differences. These differences were attributable to both assumptions regarding the development costs per patient and the actual cost differences. These cost differences can have a detrimental effect on the cost-effectiveness and financial viability of the program. No studies included indirect cost savings or expenses from a patient perspective, such as a reduction or increase in travel costs.

A major strength of our study is the inclusion of various study designs and outcome measures for health care utilization. Therefore, this review offers a broader scope than other reviews [11]. The inherent weakness of including nonrandomized studies was the introduction of biases because of case mixes. This was demonstrated by the results of the Risk of Bias in Non-randomized Studies of Interventions tool, which showed a high degree of bias in the domain of baseline confounding. The exclusion of studies also using STS within their telemedicine program resulted in the exclusion of a significant number of studies. However, the exclusion of these studies also increases the relevance of our study for those interested in the stand-alone effect of telemonitoring on health care utilization.

The main limitation of this review is the lack of a meta-analysis and thus a limited ability to draw strong conclusions regarding the effect of telemonitoring on health care utilization. However, this review clearly shows why meta-analyses are difficult to perform on this subject and, if performed, should be interpreted with caution. We observed high heterogeneity within the study populations, telemonitoring programs, and outcomes. Therefore, a meta-analysis was not appropriate for this study. In addition, the results of meta-analyses of other studies may only be applicable to small subsets of populations, interventions, and outcomes and thus may not represent the true effect of telemonitoring on health care utilization because of limited external validity.

This study has several practical implications. The finding that telemonitoring often increases non-ED visits has consequences for the workload of outpatient clinics. These additional non-ED visits may occur because of a variety of reasons such as false positives, true positives, equipment malfunction, and whether telemonitoring is used as additional or substitute care. Detecting true positives is the primary aim, as it prevents more expensive health care utilization and improves the quality of life. Although false-positive alerts and equipment malfunctions may be reduced by improving technology and algorithms [52], our results suggest that addition instead of substitution is likely to remain, resulting in additional non-ED visits. Health care providers must be aware of this increase and adopt organizational structure of the outpatient clinic or find other ways to mitigate this increase, such as using telemonitoring as a substitute rather than additional care [20] or outsourcing technical difficulties experienced by patients to medical service centers [34].

This review showed that telemonitoring might shift health care utilization to outpatient settings, as opposed to only reducing inpatient admissions. This may complicate adoption, as the benefits may not be attributable to the same stakeholder as the costs. This is especially the case for health care systems where outpatient care is delivered by organizations other than inpatient care, such as that of Germany. In such cases, conflicting interests are to be expected, and health care payers have to come to an agreement with health care providers to overcome these issues [53]. Health care regulators can facilitate this process by creating and supporting new payment models such as shared savings and lump-sum payment models.

Finally, the high costs of developing telemonitoring programs for a selected population can diminish and even cancel future monetary gains of reduced health care consumption [23]. One way of suppressing the costs of telemonitoring can be achieved through the use and development on a larger scale. Villani et al [23] showed that developing a telemonitoring system for 40 patients was not financially viable. In contrast, Vestergaard et al [46] reduced the costs of the program per patient by targeting a larger population, namely a whole region of North Denmark constituting 6700 patients. Health care providers should either use the scale realized by a third party or codevelop telemonitoring systems with other health care providers to realize possible scale advantages. In addition, in general, substitutive programs are expected to achieve higher financial savings.

Future studies should consider clinical diversity by including subgroup meta-analyses or performing meta-regression, as opposed to pooled meta-analyses [54]. As mentioned previously, we found a high degree of clinical and methodological diversity [54]. Despite this, some meta-analyses [48,55] have found a low to moderate effect for certain health care utilization measures. The presence of high clinical diversity and low statistical heterogeneity may be due to minimal marginal effects of components in the telemonitoring program; a part of the intervention did not affect the outcome measure. If this is the case, certain (combinations of) telemonitoring components do not add value to the intervention. Yun et al [56] and Kotb et al [57] performed such analyses on all-cause mortality and hospitalization in patients with CHF. Certain program characteristics, such as having 3 or more physiological parameters [56] or including an electrocardiograph [57], were statistically associated with a reduction in all-cause mortality and hospitalization, respectively. Similarly, essential and effective parts of telemonitoring for reducing other health care utilization can be identified, thereby supporting the development of (cost-) effective telemonitoring programs. Clear descriptions of the intervention and context are needed to perform such analyses. This review can be used as guidance for forming subgroups or variables of interest for meta-regressions. As there are many more factors that may affect the impact of telemonitoring on health care utilization, qualitative research may be used to develop hypotheses and guide meta-regression protocols.

This review investigated the effects of telemonitoring programs on different aspects of health care utilization. Telemonitoring has the potential to reduce hospitalization rates. However, this was not achieved in most studies, as the number of non-ED visits increased in the majority of studies. The effect of telemonitoring on health care costs is highly ambiguous and depends on the effectiveness of the intervention in reducing health care utilization as well as on the costs of the telemonitoring program itself. Health care providers and payers should be aware that the majority of current telemonitoring programs do not result in a reduction in health care utilization and may even increase health care utilization by increasing the number of non-ED visits. Possible payer strategies should be focused at increasing the scale to reduce program costs and implement telemonitoring as a substitute to reduce possible increases in outpatient visits. Nevertheless, more focus is needed to determine the essential factors of telemonitoring programs that reduce health care utilization.