This review was performed according to the Cochrane Collaboration methodological guidelines [26]. We conducted a literature search in the PubMed, EMBASE, CINAHL, and Cochrane Library databases, covering the publication period from January 1990 to February 2020. The publications were required to be (1) relevant to telehealth, telemedicine, telemonitoring, telecare, internet, mobile, smartphone, remote monitoring, or home monitoring; (2) involving patients with CHF; and (3) in the English language. An information specialist officer at the Commonwealth Scientific and Industrial Research Organization (Brisbane, Australia) and an expert librarian at Curtin University, Western Australia, Australia helped develop the bibliographies and conduct the database search (for a more complete description of our search strategy, see Multimedia Appendix 1).

Two investigators (HD and SC) independently reviewed the articles obtained. Disagreements between the two investigators were resolved by a third reviewer (AM or IE).

In this review, we employed a hierarchical structure considering that telehealth encompasses telemonitoring, as well as eHealth care processes and communication, telemedicine, and mHealth [27]. We then defined the scope of telemonitoring as “the use of information technology to monitor patients at a distance,” as described by Meystre [28]. Finally, we included a telemonitoring intervention in the analysis if it involved “the transfer of physiological data such as blood pressure, weight, electrocardiographic signals, or oxygen saturation through technology such as telephone lines, broadband, satellite, or wireless networks” [27].

This review focused on noninvasive telemonitoring interventions evaluated through an RCT. The inclusion criteria were: (1) studies evaluating telemonitoring for CHF for at least 3 months, (2) prospective RCTs comparing telemonitoring-based care with usual care, and (3) full peer-reviewed journal articles reporting outcomes of all-cause mortality or all-cause hospitalization. The exclusion criteria were: (1) articles reporting preliminary analysis outcomes; (2) studies with a sample size less than 50 (Multimedia Appendix 2), because, compared with large studies, small studies are often associated with a lower level of reporting quality [29], are more likely to be heterogeneous [30], and overestimate outcome effects [31]; and (3) telemonitoring via implantable devices, as these interventions often involve a different care paradigm to noninvasive devices and have been the subject of dedicated reviews [32,33].

We extracted 18 telemonitoring strategies according to three categories: technology applications (6 strategies), care objectives (7 strategies), and care support methods (5 strategies) (Table 1).

The risk ratio (RR) of all-cause mortality and the incidence rate ratio (IRR) of all-cause hospitalization in the RCTs were analyzed. The RR and IRR values in each RCT were calculated from the event counts of mortality and hospitalization. For each strategy, we divided the RCTs into two subgroups: RCTs that applied the strategy in the telemonitoring intervention (subgroup 1) and RCTs that did not apply the strategy (subgroup 2). We then compared the two subgroups (subgroup 1 vs subgroup 2), and examined whether the difference between the two groups in the RR and IRR outcomes was statistically significant.

In the meta-analysis, we used a random-effects model with the DerSimonian–Laird estimator [34,35], and report the RR, IRR, and 95% CI for each group. For RCTs with no events in one arm, we applied a continuity correction of 0.5. The heterogeneity of RCTs in each subgroup was examined by the Q test and I² statistic [36,37]. The statistical significance of heterogeneity was determined by a relaxed P value of .10 (P_H<.10) [38]. The I² values of 25%, 50%, and 75% were used to reflect a low, moderate, and high level of heterogeneity, respectively [37]. To evaluate the risk of bias, a regression test was used to analyze the asymmetry of a funnel plot of the RR or IRR results in a subgroup [36]. The regression test was used to examine whether the outcomes of individual RCTs were related to the corresponding sampling variances [39]. A significant regression (P_F<.05) indicated a significant risk of bias. The difference between the two groups (subgroup 1 vs subgroup 2) was evaluated by a Wald-type test [36], and statistical significance was determined if the corresponding two-sided P value was less than .05 (P_C<.05). A mixed-effects model [36] was also used to evaluate the effects of potential confounders, including sex, age, or the severity measure of left ventricular ejection fraction (LVEF). The meta-analysis methods and tests were performed using RStudio Version 1.1.383 [40] associated with the “metafor” meta-analysis package (version 2.0) [36].

A summary of the methodological risk of bias of the included studies was conducted in accordance with the Cochrane Handbook for Systematic Reviews of Interventions [26] by two investigators (HD, SC) using the risk of bias tool in the Cochrane Collaboration’s review-writing software RevMan 5.3. This involved reporting the following individual elements for the included RCTs: random sequence generation, allocation sequence concealment, blinding of participants and personnel, blinding of outcome assessment, completeness of outcome data, and selective outcome reporting. Each item was judged as being at a high, low, or unclear risk of bias. Studies were deemed to be at the highest risk of bias if they were scored at a high or unclear risk of bias for either the sequence generation or allocation concealment domains [26].

The literature search results are presented in Figure 1. We found 3870 records from the bibliographic search and 56 records from three existing systematic reviews [9,11,41] and a manual search, resulting in a total of 3926 records. After removing duplicates, we obtained 1632 articles for screening. In the screening process, we excluded 1553 articles because of absence of inclusion criteria and consequently obtained 79 articles for a full-text assessment. We then excluded 53 articles according to the inclusion and exclusion criteria, and one article because of its poor completion rate recognized by the authors [42]. Finally, this review included 26 RCTs. Among them, 25 RCTs provided all-cause hospitalization events and 21 RCTs provided mortality events.

Among the assessment elements of bias risk, the blinding of participants and personnel was the least used method in the RCTs included (Figure 2). There were 11 RCTs that did not blind participants and personnel (Figure 3). Nine RCTs did not report their blinding status and only six RCTs used a blinding approach. The blinding of outcome assessment was the least reported element, and 14 RCTs (54%) had “unclear risk of bias.”

The 26 RCTs included 11,450 participants. The participants’ characteristics are shown in Table 2. The median age was 67.4 years and the median rate of male participants was 73.15% (8376/11,450). The participants generally had a significantly reduced (<40%) LVEF, with a median LVEF of 29.6%, and they experienced mild to moderate levels of symptoms, with a median New York Heart Association functional class score of 2.6. The median trial size of the RCTs was 290 participants and the median follow-up duration was 12 months.

We extracted 18 telemonitoring strategies from the 26 RCTs, as shown in Table 3. Some strategies were commonly used, such as telemonitoring weight scales (26/26, 100% RCTs), call-center support (24/26, 92%), and daily weight monitoring (25/26, 96%). Strategies that were not commonly used included nurse support (2/26, 8%), intervention for depression and anxiety (3/26, 12%), and exercise (3/26, 12%). The telemonitoring programs in the RCTs generally contained multiple strategies, with a mean of 8.7 strategies per care program.

There were 21 RCTs (n=10,536) with event counts of all-cause mortality and 25 RCTs (n=9912) with event counts of all-cause hospitalization. The outcomes of mortality (RR) and hospitalization (IRR) with 95% CIs are shown in Figure 4 and Figure 5, respectively. Overall, telemonitoring interventions were found to be more effective than usual care on reducing all-cause mortality (RR=0.85, 95% CI 0.76-0.94, P=.01) and all-cause hospitalizations (IRR=0.90, 95% CI 0.81-0.99, P=.04). The outcomes of both RR and IRR were heterogeneous (P_H=.001), with a low-to-moderate level of heterogeneity (I²=35.3%) in the RR outcomes and a moderate-to-high level of heterogeneity (I²=73.2%) in the IRR outcomes. In the funnel plot–based test, the risk of bias was significant for both RR (z=1.89, P_F=.001 and IRR (z=3.33, P_F=.001) outcomes. We also used the mixed-effects model to adjust for sex, age, or LVEF, but did not find significant differences in these results.

The subgroup of RCTs that provided medication support (subgroup 1, 15 RCTs, n=4563, IRR=0.83, 95% CI 0.72-0.95) was found to be associated with a significantly (P=.01) lower IRR of all-cause hospitalization than the comparison subgroup of RCTs that did not apply this strategy (subgroup 2, 10 RCTs, n=5349, IRR=1.02, 95% CI 0.93-1.12), as shown in Tables 4-6 and Multimedia Appendix 3. Within the subgroup of RCTs that provided medication support, the telemonitoring interventions were found to be more effective than usual care on reducing hospitalizations (15 RCTs, n=4563, IRR=0.83, 95% CI 0.72-0.95, P=.01). The IRR outcomes in both subgroups were heterogeneous (Multimedia Appendix 3 and Multimedia Appendix 4). The outcomes in the comparison subgroup of RCTs that did not apply this strategy were associated with the risk of bias.

Similarly, the subgroup of mHealth (subgroup 1, 12 RCTs, n=2662, IRR=0.79, 95% CI 0.64-0.96) was associated with a significantly (P=.03) lower IRR of all-cause hospitalization than the comparison subgroup (subgroup 2, 13 RCTs, n=7250, IRR=1.00, 95% CI 0.94-1.06), as shown in Tables 4-6 and Multimedia Appendix 5. Within the mHealth subgroup, the telemonitoring interventions were found to be significantly more effective than usual care on reducing all-cause hospitalizations (subgroup 1, 12 RCTs, n=2662, IRR=0.79, 95% CI 0.64-0.96, P=.03). The IRR outcomes in the mHealth subgroup were heterogeneous and were associated with a risk of bias.

In analysis of all-cause mortality, the mHealth subgroup (subgroup 1, 10 RCTs, n=3711, RR=0.67, 95% CI 0.53-0.85) was also associated with a significantly (P=.01) lower RR than the comparison subgroup (subgroup 2, 11 RCTs, n=6852, RR=0.95, 95% CI 0.84-1.07), as shown in Tables 6-8 and Multimedia Appendix 6. Within the mHealth subgroup, the telemonitoring interventions were significantly more effective than usual care on reducing all-cause mortality (subgroup 1, 10 RCTs, n=3711, RR=0.67, 95% CI 0.53-0.85, P<.001). No significant heterogeneity was detected in both the mHealth subgroup and comparison subgroup. A significant risk of bias (P_F=.01) was found in the comparison subgroup.

In the subgroup comparison of RR and IRR outcomes, we also used the mixed-effects model to adjust for sex, age, or LVEF, but did not find significant improvements in these RR and IRR analysis results.

In this systematic review and meta-analysis, we evaluated 18 telemonitoring strategies in 26 RCTs. In addition to a traditional meta-analysis for overall effectiveness, we used a subgroup comparison method to analyze the effects of different telemonitoring components on clinical outcomes. We found that the telemonitoring strategy of providing medication support was associated with reduced all-cause hospitalization, whereas mHealth systems were associated with both reduced all-cause hospitalization and reduced all-cause mortality. Therefore, our review provides unique insight into specific telemonitoring strategies associated with improved clinical outcomes, which will help inform future telemonitoring interventions.

The positive findings related to the medication support strategy underscore the importance of medication therapy in telemonitoring interventions for CHF care. Strong evidence supports the role of modern pharmacological therapy in CHF management for delaying CHF deterioration [66,67], and reducing mortality and hospitalizations [18,67]. However, the therapeutic benefits are often limited by suboptimal patient adherence [68] and this limitation is not addressed by traditional face-to-face consultations [25]. Our findings suggest that the use of telemonitoring improves the efficacy of medication therapy, possibly through frequent reinforcement of compliance, leading to reduced episodes of clinical deterioration requiring hospitalization. Further research on optimizing medication therapy and underlying care processes in telemonitoring interventions is warranted to improve clinical outcomes in CHF care.

Using the subgroup comparison method, we also found that the strategy of providing telemonitoring interventions through an mHealth system was associated with a significant improvement in both all-cause mortality and hospitalization (or corresponding RR and IRR) outcomes. These positive findings could be supported by several unique advantages of using mHealth for general chronic disease care, including ease of use, portability, and real-time communication [69-71]. These advantages have been shown to improve the underlying care processes of patients’ self-management [72], care engagement [73,74], and medication adherence in CHF [75]. Therefore, our positive findings support delivering telemonitoring interventions through mHealth platforms, consistent with the increasing trend in using smartphones and computer tablets for the primary and secondary prevention of chronic disease [76,77].

Three recent reviews of mHealth in CHF management have resulted in inconsistent outcomes and, consequently, were unable to conclude significant clinical benefits [21,23,78]. In contrast to these traditional reviews, each intervention program in our mHealth subgroup combined both telemonitoring and mHealth interventions. Our positive finding indicates that simple mHealth apps without telemonitoring (enabling care providers to provide timely clinical intervention), such as apps only focusing on self-management or education, were insufficient to improve clinical outcomes. Similarly, this finding suggests that telemonitoring programs focusing on clinical assessment and intervention, but not delivered through an mHealth environment, fail to engage patients with CHF in self-management to the same extent as those provided via mHealth. Therefore, our finding warrants future research on comprehensive care programs combining telemonitoring and mHealth to improve both timely clinical intervention and patient engagement in CHF care.

As a part of our evaluation, we also conducted a traditional meta-analysis to evaluate the overall effectiveness of all of the telemonitoring interventions in the RCTs included in this review. We found that telemonitoring interventions were more effective than usual care on reducing both all-cause mortality and all-cause hospitalizations. This finding adds evidence to support telemonitoring interventions for CHF care generally. In our review, invasive telemonitoring interventions and small RCTs were excluded. These exclusions may have refined the selection of telemonitoring studies, leading to the significant findings, in contrast to the three previous mHealth reviews with inconclusive findings [21,23,78].

It is also important to note that several strategies such as daily weight monitoring, call center support, and exercise contained limited numbers of RCTs in the subgroup or comparison group. The evaluation of these strategies was therefore limited by our subgroup comparison method. However, these strategies should not be overlooked, and further research on their contributions to CHF care, such as improving patient adherence to daily weight monitoring and level of exercise, remains essential to continuously improve telemonitoring outcomes in future studies.

Because the objective of our review was to evaluate different telemonitoring strategies, our meta-analysis did not rigorously exclude RCTs with risk of bias, although we did exclude studies with small sample sizes. In addition, this review was an exploratory study, and hence we did not adjust the P value in the multiple comparisons of the telemonitoring strategies.

The issues of mixed mortality and hospitalization outcomes have deterred the adoption of telemonitoring in CHF care. To address this issue, this review extensively investigated strategy-related factors associated with improvements in the outcomes, and found that the strategies of (1) providing medication support and (2) combining telemonitoring interventions through mHealth were associated with a significant improvement in all-cause mortality or hospitalizations. Importantly, these findings emphasize the importance of prioritizing medication therapy and patient engagement through mHealth apps in future telemonitoring interventions for CHF care.