This systematic review and meta-analysis was performed following the PRISMA (Preferred Reporting Items for Systematic Reviews and Meta-Analyses) guidelines [27]. We searched the PubMed, EMBASE, CENTRAL (Cochrane Central Register of Controlled Trials), and CINAHL databases from January 1, 2011, to July 12, 2022, to identify randomized controlled trials (RCTs) that provide eHealth self-management interventions for HF patients. We used combined search terms such as (heart failure OR cardiac failure OR heart decompensation) AND (self-management OR self care OR self-administration OR self-medication) AND (Telemedicine OR mobile health OR eHealth OR m-health) AND (randomized controlled trial OR randomization OR randomly). Screening titles/abstracts and full-text evaluation of the articles were managed in a database created with Rayyan software [28]. The full protocol of this systematic review and meta-analysis has been registered in PROSPERO (CRD42021246973).

Inclusion and exclusion criteria strictly followed the Participants-Intervention-Comparison-Outcome (PICO) framework [29]: (1) participants were defined as adults (aged≥18 years) who had been diagnosed with HF (studies in children or adolescents were excluded); (2) interventions consisted of self-management tools including at least one eHealth component (eg, mobile apps), with traditional interventions that did not use any technical support excluded, such as face-to-face meetings; (3) the comparison was HF patients versus usual care; and the (4) outcomes included patient and process outcomes. Patient outcomes are measures directly related to the disease and were considered as the primary outcomes in our study, including all-cause readmission rate, all-cause mortality, HF readmission, and cardiovascular mortality. Process outcomes are measures related to patients’ behavior, which were considered as the secondary outcomes in our study, including quality of life, self-care behaviors, HF knowledge, and medication adherence. In addition, only RCTs or cluster RCTs reporting one or more selected outcomes were included. Reviews, editorials, protocols, and non-English papers were excluded.

The following characteristics were extracted from studies that met the inclusion criteria: name of the first author, publication year, country, type of eHealth technologies, target of eHealth intervention, duration of intervention, patient demographics, number of participants, recruitment setting, outcomes, descriptions of the control, and the intervention. All outcome data used for the meta-analysis were extracted by the same two independent reviewers (J Li and DW). The synthesis of the results and data charting were performed independently by two researchers (SL and J Li). Disagreements were discussed and consulted with a third reviewer (J Liu) to reach consensus.

We used the Cochrane Risk of Bias tool (RoB 2) to assess the risk of bias for each study [30]. This tool assesses 5 domains to address different types of bias: randomization process, deviations from the intended interventions, missing outcome data, measurement of the outcome, and selection of the reported result. Each component includes a low risk of bias, some concerns, or high risk of bias. Risk of bias assessment was performed independently by two reviewers with consultation of the third reviewer when necessary.

The Grading of Recommendations, Assessment, Development, and Evaluation (GRADE) approach was used to rate the certainty of evidence for each outcome of interest [31]. This approach rates the risk of bias, inconsistency, indirectness, imprecision, and other considerations (eg, publication bias) as “high,” “moderate,” “low,” or “very low.” The GRADE assessment was completed with the GRADEpro Guideline Development Tool.

We performed meta-analyses using Review Manager [32] (RevMan v.5.4; Cochrane Training) and R (v.4.1.0 x64). The Mantel-Haenszel odds ratio (OR) was used as the effect size for dichotomous outcomes and the standardized mean difference (SMD) was used to calculate the intervention effect for continuous outcomes. Heterogeneity was then analyzed using the Cochran Q test and the I² statistic. For the Q test, P<.10 was considered to indicate statistically significant heterogeneity [33]. According to Cochrane Handbook guidelines [34], the interpretation of the I² statistic was as follows: 0% to 40% may represent not important heterogeneity, 30% to 60% may represent moderate heterogeneity, 50% to 90% may represent substantial heterogeneity, and 75% to 100% may represent considerable heterogeneity. The fixed-effects model was used for outcomes with low heterogeneity and the random-effects model was used when the heterogeneity was significantly high. To measure the publication bias, we constructed funnel plots and performed the Egger test for all primary outcomes to determine the significance of potential asymmetry [35]. The threshold of a significant P value was .05. A sensitivity analysis was performed to check the robustness of the pooled results using the leave-one-out approach [36].

A total of 1884 articles were retrieved in the literature search. After eliminating duplicates, 1203 titles and abstracts were screened in relation to the inclusion/exclusion criteria. Of these, 1144 articles were excluded and a total of 59 articles were subject to full-text review. Finally, 24 articles were included in the systematic review and meta-analysis (Figure 1).

The results of risk of bias assessment are summarized in Figure 2. The majority of studies showed a low risk of bias for the randomization process, and only 5 studies were judged to have some concerns due to lack of detailed information about randomization. The studies could not have been blinded to participants considering the nature of the intervention and 11 studies were judged to have some concerns regarding deviations from the intended interventions. One study was determined as “high risk” for the category of measurement of the outcome because all questionnaire scores were patient-reported, which may have possibly affected the outcome [37]. In addition, one study did not report all of the outcomes according to the trial registry record, resulting in a judgment of “high risk” regarding selection of the reported result [38]. For studies included in the meta-analysis, the overall bias was low in 6 studies (25%), with some concerns in 15 studies (63%) and high bias in 3 studies (13%).

A total of 9634 participants with HF were included in the 26 (22+2×2) RCTs, with 4820 patients allocated to the eHealth group and 4814 to the usual-care group. Since two studies performed a three-group parallel-randomized trial and reported their results separately [38,39], we divided each of the three-group trials into two RCTs. These studies were performed in 20 different countries, with the largest number of studies performed in the United States (7 studies). Most (15 studies) were multicenter studies, while the other 11 RCTs recruited patients from a single center. The proportion of male HF patients ranged from 46.8% to 83% in the eHealth groups and from 30.2% to 93% in the usual-care groups. The New York Heart Association classification, which classifies patients into one of four categories based on their limitations during physical activity and has been used clinically to determine trial eligibility, was reported in 24 studies. The basic characteristics of the included studies are presented in Table 1, whereas the characteristics of participants and details of interventions are summarized in Multimedia Appendix 1.

These studies used various types of eHealth technologies for the intervention (Table 1). Of these, 14 studies used a designed telemedical system; 5 studies used the telephone; 3 studies used a mobile phone or tablet app; 2 studies used an internet website and text messages, respectively; and 2 studies used a combination of telephone with a designed telemedical system intervention. The duration of the intervention ranged from 1 month to 28 months (17 studies had interventions lasting≥6 months). The target of the intervention was classified into three types: education, monitoring, and reminders. The mixed targets were defined as the presence of the three types in one study. Educational interventions were defined as interventions that aimed at improving the HF knowledge available through educational programs and instructions. Reminder interventions were defined as interventions that prompted patients to do something (eg, take medication on time) and could be delivered through a designed telemedical system, mobile text messages, or telephone calls. Monitoring interventions were defined as interventions that transmitted vital signs or symptoms (eg, weight, blood pressure, or heart rate) to an external telemedicine center, which may lead to lasting improvements in behaviors. The eHealth intervention varied regarding intensity and the extent of the human component, although most trials provided the eHealth intervention every day (Multimedia Appendix 1).

Since the inclusion criteria of the trials were different, which contributed to the heterogeneity of studies, subgroup analyses of primary outcomes regarding the enrolled population setting (after discharge or ambulatory clinic) were performed and the results are summarized in Multimedia Appendix 2. Compared with patients in an ambulatory clinic setting, the eHealth intervention showed a larger effect in patients with HF after discharge on the reduction of all-cause mortality (OR 0.73, 95% CI 0.58-0.93, P=.01), cardiovascular mortality (OR 0.71, 95% CI 0.53-0.95, P=.02), all-cause readmission (OR 0.70, 95% CI 0.56-0.87, P=.001), and HF-related readmission (OR 0.75, 95% CI 0.56-1.00, P=.05) than the usual-care group. The results showed that the effects of an eHealth intervention may vary in certain population settings.

Sensitivity analysis was performed for the primary outcomes of meta-analyses, including all-cause mortality, cardiovascular mortality, all-cause readmission rate, and HF-related hospitalizations, using the leave-one-out approach. The direction and magnitude of the combined estimates did not vary markedly with the exclusion of individual studies, indicating the reliability of the findings (Multimedia Appendix 3). The results showed no publication bias in the primary outcomes. No publication bias was detected for any of the primary outcomes. The funnel plots and linear regression test results of funnel plot asymmetry are shown in Multimedia Appendix 4.

There is a reasonable amount of original research exploring the effects of eHealth interventions in HF. However, among the 1884 articles we identified, only 24 met the inclusion criteria for this systematic review. The pooled results suggest that eHealth self-management interventions can improve primary and secondary outcomes in patients with HF.

eHealth self-management is a rapidly growing area of HF management, and considerable research is still needed to promote the widespread use of eHealth self-management and enhance its clinical outcomes. Future research should investigate how to expand the content of eHealth self-management tools while adapting to different HF patients. There is a need for a cost-benefit analysis of eHealth self-management, and how to most effectively integrate eHealth self-management into workflows and deploy it in different settings. Further work is needed to understand how eHealth self-management can help with the transformation of care delivery models and how to combine eHealth self-management with a workflow-oriented quality improvement program.

Although this systematic review and associated meta-analyses showed that eHealth self-management interventions benefit patients with HF, this study has several limitations. First, the included studies varied in participant demographics, types of HF, sample size, assessment tools, period of interventions, and type of eHealth interventions, which can lead to heterogeneity and bias. Furthermore, reliance on self-reports and the use of questionnaires to score outcomes may lead to bias in self-reported data. Therefore, the results of this systematic review should be interpreted with caution. Second, participants were recruited at a single site or concentrated in a single population in some cases; thus, the resulting sample was not representative of the entire HF patient population, which would limit the generalizability of the findings. Finally, we only searched for literature in English, excluding other languages. This may limit the retrieval of non-English but related papers.

In this systematic review, we performed a literature search and provided a comprehensive overview of eHealth self-management in patients with HF. We observed that eHealth self-management can support the health of HF patients in many ways. However, the effectiveness of eHealth in the self-management of patients with HF is affected by multiple aspects and more clinical studies are needed to prove its effectiveness.