Telemonitoring modalities in heart failure: comparative effectiveness across the heart failure population—a meta-analysis

Abstract

Remote monitoring (RM) is effective for managing heart failure (HF), but it remains unclear which patients benefit most from which RM modality. We conducted a meta-analysis of 79 randomised trials including 31,669 patients comparing RM with standard care for total and first HF hospitalisations and all-cause mortality. Subgroup analyses evaluated effects by geographic region and HF status, and meta-regression assessed the influence of age, left ventricular ejection fraction, New York Heart Association class, sex, and publication year. Network meta-analysis ranked RM modalities using Surface Under the Cumulative Ranking scores. Overall, RM reduced total HF hospitalisations (incidence rate ratio 0.81, 95% confidence interval [CI] 0.72–0.91), first HF hospitalisations (risk ratio 0.82, 95% CI: 0.76–0.88), and all-cause mortality (risk ratio 0.90, 95% CI: 0.84–0.95). Subgroup and meta-regression analyses showed consistent benefits across patient and study characteristics without significant interaction effects. In network meta-analysis, invasive hemodynamic monitoring ranked highest for reducing total HF hospitalisations, while structured telephone support ranked highest for reducing first HF hospitalisation and all-cause mortality. RM consistently improves HF outcomes across a range of patient and study characteristics, supporting its broad use. However, based on these characteristics, current evidence does not allow targeted RM implementation for specific patients most likely to benefit.

Introduction

Over the past decades, growing evidence has highlighted the potential of remote monitoring (RM) to reduce the burden of heart failure (HF) on patients and healthcare systems¹. Several meta-analyses have evaluated the effectiveness of both non-invasive and invasive RM modalities^2–7, however, variability among studies limits the understanding of which modalities are most effective in specific patient subpopulations. This variability likely arises from differences in RM modalities, study design, and patient characteristics, including age, sex, left ventricular ejection fraction (LVEF), New York Heart Association (NYHA) class, and HF stability, as well as regional differences in healthcare infrastructure. Patients in wealthier countries, with advanced reimbursement systems and greater resources for RM integration, tend to benefit more than those living in lower-resource settings^8–10. Furthermore, the existing evidence spans more than two decades, beginning in 1999, a period marked by significant advancements in technology, HF treatment, and healthcare systems¹¹. Understanding how these factors influence RM effectiveness is important for identifying which patients benefit most from specific modalities and for implementing patient-specific strategies. In this context, we conducted subgroup and meta-regression analyses of clinical trial data, along with a network meta-analysis (NMA), to study the clinical effects of various RM modalities in relation to relevant patient- and healthcare characteristics. We ultimately aim to identify the most effective patient-specific RM strategies for reducing HF hospitalisations and mortality.

Results

Study and patient characteristics

The search identified 7968 articles after removing duplicates, of which 79 were included in the NMA^12–90 and 78 in the subgroup and meta-regression analyses^{12–80,82–90}. Among the included non-invasive RM studies, 36 focused on TM, 20 on STS, and 7 on CTM. Among the invasive RM studies, 14 investigated CIED monitoring and five examined IHM. Of the 79 studies, three included direct comparisons between RM interventions. One study compared TM and STS⁸¹, while two others had a three-arm design: one comparing STS, TM, and standard care³⁰, and another comparing CTM, STS, and standard care⁶⁴. Compared to our previous meta-analysis⁵ five new RCTs were published, one focused on CIED⁹⁰, two on TM^13,62, two on IHM^25,89, and one provided a head-to-head comparison between TM and STS, which was only included in the NMA⁸¹. The PRISMA flow diagram is available in Supplementary Fig. S1.

In total, 31,669 patients were included across all studies, with a mean follow-up of 12.0 months (range: 1–45 months). An overview of study and patient characteristics, stratified by RM modality, is presented in Tables 1 and 2. The overall mean ± standard deviation age of the patients was 67.3 ± 4.9 years, with a majority being male (62.9%), consistent across all RM modalities. A more detailed overview of patient and study characteristics can be found in Supplementary Table S2–S7.

Table 1.

Characteristics of the included studies

		Non-invasive remote monitoring			Invasive remote monitoring
	All studies	Telemonitoring	Structured telephone support	Complex telemonitoring	Cardiac implantable electronic devices	Invasive hemodynamic monitoring
Study characteristics, n (%)	n = 79	n = 36	n = 20	n = 7	n = 14	n = 5
Participants per study, mean [range]	405 [20–1653]	285 [20–1653]	409 [40–1518]	660 [57–1538]	549 [80–1650]	470 [180–1000]
Outcomes
Total heart failure hospitalisations	26 (32.9)	12 (33.3)	3 (15.0)	2 (28.6)	6 (42.9)	4 (80.0)
First heart failure hospitalisation	49 (62.0)	22 (61.1)	13 (65.0)	3 (42.9)	10 (71.4)	3 (60.0)
All-cause mortality	73 (92.4)	32 (88.9)	18 (90.0)	7 (100.0)	14 (100.0)	5 (100.0)
Follow-up time, mean [range]	12.0 [1–45]	11.9 [1–45]	9.2 [1–18]	14.5 [6–26.3]	17 [12–36]	8 [6–12]
Unstable or stable HF
Unstable	23 (29.1)	15 (41.7)	7 (35.0)	2 (28.6)	0 (0)	0 (0)
Stable	8 (10.1)	3 (8.3)	2 (10.0)	2 (28.6)	1 (7.1)	0 (0)
Both	48 (60.8)	18 (50.0)	11 (55.0)	3 (42.9)	12 (92.8)	5 (100.0)
Global region
Asia and Oceania	6 (7.9)	3 (8.3)	3 (15.0)	0 (0)	0 (0)	0 (0)
Europe	37 (48.7)	16 (44.4)	6 (30.0)	5 (71.4)	10 (91.7)	1 (25.0)
Middle and Southern America	5 (6.6)	3 (8.3)	2 (10.0)	0 (0)	0 (0)	0 (0)
Middle Eastern	1 (1.3)	1 (2.8)	0 (0)	0 (0)	0 (0)	0 (0)
Northern America	27 (35.5)	13 (36.1)	9 (45.0)	2 (28.6)	1 (8.3)	3 (75.0)

HF heart failure, NYHA class New York Heart Association class, LVEF left ventricular ejection fraction, SD standard deviation.

Table 2.

Patient characteristics per study arm

	No. patients with available data (no. studies)	Standard of care	Telemonitoring	Structured telephone support	Complex telemonitoring	Cardiac implantable devices	Invasive hemodynamic monitoring
		n = 15214	n = 5092	n = 4092	n = 2291	n = 3815	n = 1165
Age, mean ± SD	30,878 (73)	67.3 ± 4.9	67.6 ± 5.9	67.1 ± 4.8	69.7 ± 3.9	66.3 ± 3.5	66.1 ± 4.6
Sex, n (%)	30,982 (74)
Women		4213 (30.9)	1630 (33.4)	594 (18.9)	741 (43.3)	692 (18.4)	379 (35.2)
Men		9406 (69.1)	3247 (66.6)	2543 (81.1)	969 (56.7)	3062 (81.6)	698 (64.8)
NYHA class III–IV, n (%)	28567 (62)	7683 (55.7)	2430 (58.9)	1793 (52.3)	1326 (57.9)	1705 (45.2)	1019 (87.5)
LVEF, mean ± SD	21430 (52)	33.2 ± 7.4	33.0 ± 6.4	40.2 ± 11.1	37.2 ± 6.5	27.6 ± 2.1	37.2 ± 3.7
Ischaemic aetiology, n (%)	18484 (49)	4567 (51.4)	1657 (53.1)	759 (58.2)	751 (47.7)	1320 (51.4)	491 (47.6)

SD standard deviation, NYHA class New York Heart Association class, LVEF left ventricular ejection fraction.

Conventional meta-analysis

The conventional meta-analysis of all RM modalities combined resulted in an IRR for total HF hospitalization of 0.81 (95% CI: 0.72–0.91), an RR for first HF hospitalization of 0.82 (95% CI: 0.76–0.88), and an RR of all-cause mortality of 0.90 (95% CI: 0.84–0.95) (Fig. 1 and Supplementary Figs. S2–S5). For the outcomes total HF hospitalisation and all-cause mortality, there was no indication for publication bias (Supplementary Figs. S6 and S8), whereas the outcome first HF hospitalisation revealed significant publication bias (Supplementary Fig. S7). The majority of the studies were classified as having ‘some concerns’ regarding risk of bias, while 20% were classified as ‘high risk’ (Supplementary Figs. S9 and S10). The results of the sensitivity analysis (Supplementary Fig. S14) were consistent with the primary analysis.

Fig. 1. Updated meta-analysis.

RM remote monitoring, SoC standard of care, HF heart failure, IRR incidence rate ratio, RR risk ratio, CI confidence interval.

Subgroup and meta-regression analyses

The results of the subgroup analysis for all clinical endpoints are presented in Fig. 2. There was consistency across all subgroups across all outcomes for all RM modalities, without any significant unadjusted or adjusted p-values for interaction. The results of the meta-regression are shown in Fig. 3. Only within the modality IHM, an interaction was observed within the percentage NYHA class III/IV (p-interaction: 0.03) and sex (p-interaction: 0.03). However, both disappeared after adjusting for multiple testing.

Fig. 2. Subgroup analysis.

^*Only the geographical regions in which studies are carried out. HF heart failure, IRR incidence rate ratio, p-int. p-interaction, RR risk ratio, CI confidence interval, SoC standard care.

Fig. 3. Meta-regression analysis.

HF heart failure, IRR incidence rate ratio, p-int. p-interaction, OR odds ratio, CI confidence interval, LVEF left ventricular ejection fraction, NYHA class New York Heart Association class, SoC standard care.

NMA

Figure 4A presents the results of the NMA for total HF hospitalisations. IHM (IRR: 0.67; 95% CI: 0.51–0.87; SUCRA: 0.90) was associated with the greatest reduction in total HF hospitalizations compared with standard care, followed by TM (IRR: 0.73; 95% CI: 0.62–0.88; SUCRA: 0.77) and STS (IRR, 0.81; 95% CI: 0.62–0.99; SUCRA, 0.61). The SUCRA values indicate that IHM had the highest probability of being the most effective RM modality, with a SUCRA score of 0.90. CTM (IRR: 0.93; 95% CI: 0.64–1.35; SUCRA: 0.37) and CIEDs (IRR: 1.07; 95% CI: 0.84–1.35; SUCRA: 0.14) did not demonstrate an effect on total HF hospitalisations compared with standard care. Detailed results for each RM comparison are provided in a league table (Supplementary Table S8). The certainty of evidence was classified as moderate for most comparisons (Supplementary Table S9). The asymmetrical distribution within the funnel plot and results of the Egger regression test suggested potential publication bias for this outcome (Supplementary Fig. S10).

Fig. 4. Network meta-analysis.

HF heart failure, IRR incidence rate ratio, SUCRA surface under the cumulative ranking curve, RR risk ratio, CI confidence interval, SoC standard care.

The results of the NMA for first HF hospitalization are presented in Fig. 4B. STS (RR: 0.66; 95% CI: 0.55–0.79; SUCRA: 0.83) and IHM (RR: 0.70; 95% CI: 0.49–0.99; SUCRA: 0.71) were associated with the greatest reduction in the rate of first HF hospitalizations compared to standard care, followed by CTM (RR: 0.71; 95% CI: 0.51–1.00; SUCRA: 0.69) and TM (RR: 0.78; 95% CI: 0.67–0.91; SUCRA: 0.51). The SUCRA values indicate that STS had the highest probability of being the most effective RM modality, with a SUCRA score of 0.83. CIEDs (RR: 0.94; 95% CI: 0.77–1.15; SUCRA: 0.19) did not demonstrate an effect on first HF hospitalisations compared with standard care. Detailed results for each RM comparison are provided in a league table (Supplementary Table S5). The certainty of evidence was classified as moderate for most comparisons (Supplementary Table S10). The asymmetrical distribution within the funnel plot and results of the Egger regression test suggested potential publication bias for this outcome (Supplementary Fig. S11).

The results of the NMA for all-cause mortality are presented in Fig. 4C. Only STS (RR: 0.76, 95% CI: 0.65–0.88; SUCRA: 0.96) was associated with a reduction in all-cause mortality compared to standard care, while TM (RR: 0.89, 95% CI: 0.78–1.01; SUCRA: 0.59), CTM (RR: 0.91, 95% CI: 0.76–1.08; SUCRA: 0.50), CIEDs (RR: 0.94, 95% CI: 0.79–1.11; SUCRA: 0.43), and IHM (RR: 0.94, 95% CI: 0.74–1.19; SUCRA: 0.38) did not show an effect on all-cause mortality. Detailed results for each type of RM comparison are included in a league table (Supplementary Table S5). The certainty of evidence was classified as moderate for most comparisons (Supplementary Table S11). The asymmetrical distribution within the funnel plot and results of the Egger regression test suggested potential publication bias for this outcome (Supplementary Fig. S12).

The sensitivity analysis results after excluding studies with a high risk of bias (Supplementary Fig. S15) showed that TM was associated with the greatest reduction in total HF hospitalizations (IRR: 0.62; 95% CI: 0.49–0.80; SUCRA: 0.89), closely followed by IHM (IRR: 0.67; 95% CI: 0.51–0.88; SUCRA: 0.80). TM was also associated with the greatest reduction in all-cause mortality (RR: 0.80; 95% CI: 0.67–0.96; SUCRA: 0.86). All other results aligned with the primary analysis.

Discussion

The aim of this review was to determine whether patient and study characteristics have an impact on RM effectiveness and can be used to identify the most effective RM modality for reducing HF hospitalisations and all-cause mortality, with the aim to ultimately providing further guidance to select the right patient for the right RM modality.

Our review is the first large-scale (network) meta-analysis to evaluate whether patient and study characteristics influence RM effectiveness while simultaneously providing comparisons of the effectiveness of different RM modalities against standard care. We demonstrate several key insights. First, the conventional meta-analysis of all combined modalities confirms RM to be effective in reducing clinical outcomes in HF patients. The evidence base for RM is growing, and these data are an advocate for its general use in HF patients. Second, the subgroup and meta-regression analyses demonstrated no clinically relevant interaction effects across patient and clinical study characteristics, suggesting that RM provides consistent benefits regardless of these factors. However, this lack of differentiation highlights the need for alternative approaches to better guide patient selection for specific RM modalities, which, based upon the current dat, is not achieved. The NMA therefore provided additional relevant information when selecting RM strategies by directly comparing different RM modalities and identified IHM as having the highest probability of being the most effective in reducing total HF hospitalisations. In contrast, STS was ranked highest for reducing first HF hospitalisation and was the only RM modality with a reduction in all-cause mortality. These findings provide a clue into the appropriate use of RM, as there appears to be robust potential advantages for IHM regarding recurrent HF hospitalisations in the longer term and STS for first HF hospitalisation and all-cause mortality. In practice, if patients deteriorate, more advanced RM modalities may have additional value, as neither guidelines nor clinicians will opt for upfront implantation of IHM devices. But in patients who are re-hospitalised, despite GDMT and despite simple RM, RM may be escalated to IHM devices, which have demonstrated efficacy, especially in patients with an increased risk of HF hospitalisation to prevent all HFH.

The consistent effects of RM across all RM modalities are in line with the subgroup results observed in individual landmark RM RCTs, as no interaction effects of specific subgroups were observed when reported. The RCTs that performed subgroup analyses reported consistent effects across age, sex, and NYHA class.^{25,30,53,54,68,91,92}. In addition to patient and clinical study characteristics, it is important to examine the specific components of RM modalities. These typically encompass various clinical measurements, different communication methods (e.g., video calls, phone calls, text messages, or apps), and additional modules (e.g., self-management, medication, or education). De Lathauwer et al. investigated these factors in the context of non-invasive RM and reported that self-management and educational modules positively influence its effectiveness⁹³. However, it remains unclear which combination of clinical parameters is most informative in an RM setting and therefore most effective in practice. The effectiveness of RM may depend not only on the type of data collected but also on how it is processed, interpreted, and integrated into clinical decision-making. An important step is to gain a clearer understanding of the workflow of RM systems, from the clinical parameters that are collected to how they are processed and acted upon by the physician. Mapping out the entire process transparently is essential to identifying which components ultimately influence clinical outcomes⁹⁴.

In the current study, the reliance on aggregated data from multiple studies offers limited insight into subgroup-specific effects, a challenge further compounded by heterogeneity across studies. In addition to relative effects, absolute risk reductions should be assessed to guide the selection of RM modalities. However, due to limited data on patient characteristics, this analysis was unable to determine the absolute effects of RM within different subgroups or clinical contexts. To achieve this, access to individual patient-level data would be ideal. Alternatively, predictive models such as the Meta-Analysis Global Group in Chronic Heart Failure (MAGGIC) model and the Seattle Heart Failure Model could estimate absolute effects by applying relative risk estimates to the absolute risks calculated for specific patient subgroups. These models, however, primarily focus on all-cause mortality, and there is a lack of similar, validated models for (recurrent) HF hospitalisations. Despite the challenges, the current comprehensive analysis provides valuable insights into the comparative effectiveness of RM modalities and shows potential advantages for IHM regarding total HF hospitalisations and for STS for first HF hospitalisation and all-cause mortality. When selecting a monitoring modality, factors such as cost-effectiveness, hospital resources, healthcare staff availability, as well as healthcare system capacity, and the preferences of both patients and providers should therefore be considered alongside patient characteristics.

The head-to-head comparisons of RM modalities may provide additional direction in selecting RM modalities. Over a decade ago, a smaller NMA of recently discharged HF patients suggested possible benefits of STS with human contact and TM with medical support compared to standard care, though the results were inconclusive⁹⁵. Subsequent research, including another NMA of non-invasive RM modalities, disease management clinics, and pharmacist interventions, did not show reductions in all-cause readmissions and mortality⁹⁶. Both studies highlighted the need for further investigation into different RM types, which remains an area for future research. The findings in the current NMA suggest that IHM has the highest probability of being the most effective modality in reducing total HF hospitalisations as compared to other RM modalities. The SUCRA score of 0.90 reflects a strong level of evidence, reinforcing IHM’s superiority in managing HF hospitalisations effectively. The physiological benefits of IHM are well-understood; left ventricular filling pressures can increase days or weeks before clinical symptoms and an HF exacerbation become apparent. Timely decongestion through IHM leads to early intervention, thereby preventing hospitalisations. As mentioned earlier, the current analysis did not demonstrate a significant effect of IHM on all-cause mortality, this effect may be explained by the limited follow-up duration and the smaller number of studies performed with insufficient statistical power. Increases in pulmonary artery or left atrium pressures, or the distension of these cardiac structures, are known to trigger sympathetic activation, elevate heart rate, intensify myocardial wall stress, impair renal function, and provoke arrhythmias, all factors that negatively impact HF outcomes⁹⁷. Consequently, interventions that effectively reduce PA and LA pressures might enhance survival rates by alleviating one or more of these harmful conditions. While this effect was not shown in a meta-analysis of PA pressure sensor trials⁹⁸, a patient-level meta-analysis by Lindenfeld et al. found a 25% reduction in all-cause mortality (106).

Building on this reasoning, CIEDs, with or without impedance measurement techniques, may be less effective, as multiple meta-analyses, as well as the current analysis shows that CIEDs did not significantly reduce clinical outcomes. However, they remain valuable for monitoring device integrity and detecting arrhythmias, and ongoing studies may provide further insights into their effectiveness for clinical endpoints. Additionally, these measurements can help stratify the risk of impending HF events⁹⁹. A well-known challenge with CIEDs, however, is their high rate of false positive alarms.

While STS ranked as the most effective intervention for reducing all-cause mortality in the NMA, this finding did not persist in the sensitivity analysis excluding high-risk-of-bias studies, where TM ranked highest. This suggests that comparative performance may be influenced by study quality. Compared with the earlier meta-analysis by Inglis et al.⁶, STS demonstrated a larger effect in our review. This likely reflects methodological rather than clinical differences: conventional pairwise syntheses rely solely on direct comparisons, whereas our NMA integrates direct and indirect evidence and separates first from total HF hospitalisations. Temporal changes in guideline-directed therapy uptake across recruitment eras and healthcare settings may also have contributed, particularly given that the analysis by Inglis et al. was performed more than a decade ago.

Several mechanisms may explain the mortality benefit of STS. Many STS trials were performed in periods with less comprehensive guideline-directed therapy, making incremental improvements in care delivery more impactful. STS also provides structured and frequent patient–provider contact, independent of symptom-triggered alerts, enabling earlier detection of deterioration, timely treatment adjustments, and potentially more GDMT titration. The principles underlying STRONG-HF¹⁰⁰, namely intensive follow-up, repeated contact, and rapid optimisation, suggest that the intensity of clinical interaction itself may be an important contributor to benefit. Thus, the observed effects of STS likely reflect not only its monitoring component, but also behavioural and system-level influences associated with high-contact care that directly benefit patients. These observations highlight that RM effectiveness may depend as much on engagement structure as on the underlying technology.

As with many meta-analyses, heterogeneity remains a concern. To address this, we included only RCTs to provide more accurate comparisons between RM modalities and standard care, as well as between the RM approaches themselves. Consequently, our approach reduced heterogeneity compared to the previous meta-analysis, especially for first HF hospitalisation and all-cause mortality⁵. Nevertheless, variations in RM definitions and the limited number of studies for certain RM modalities contributed to residual heterogeneity. Furthermore, asymmetry observed in the funnel plots suggests the potential presence of publication bias. This is an important consideration because not reporting negative or neutral RM trials, could inflate the apparent effectiveness of some modalities.

For the subgroup and meta-regression analyses, not all RCTs provided the necessary data, leading to their exclusion from these analyses. In addition, the currently available data did not permit the calculation of absolute effects of RM across different subgroups or clinical contexts.

Temporal changes in both guideline-directed medical therapy and the telemonitoring interventions themselves across the 26-year study span (1999–2025) may also have introduced heterogeneity. Uptake of therapies such as mineralocorticoid receptor antagonists, ARNI, and SGLT2 inhibitors varied widely across eras and regions, making a fixed cut point difficult to justify. In parallel, TM evolved from relatively simple, analogue or telephone-based contacts in the late 1990s to more sophisticated, digitally integrated platforms using mobile phones, internet connectivity, and automated data transfer in more recent trials. Although we modelled publication year as a continuous moderator, unmeasured differences in background therapy and in the technical maturity and intensity of TM programmes may still have influenced absolute event rates, even though relative effects remain internally valid within each randomised trial.

Another limitation concerns the characterisation of remote monitoring interventions. Although we now provide detailed modality definitions in the main text, granular information on specific telemonitoring platforms, alert pathways, and provider response processes was inconsistently reported across trials. This reporting gap limits the ability to fully interpret between-modality differences. In addition, several mediators of effectiveness, such as patient adherence, responsiveness of clinicians to alerts, care coordination structures, and broader health system context, are hard to measure and not often described. These behavioural and system-level components may influence real-world effectiveness, but could not be evaluated in this analysis.

For the NMA, similar to the conventional meta-analysis, funnel plots indicated the potential presence of publication bias. Moreover, substantial variability was observed in study characteristics, including sample size, follow-up duration, and inclusion criteria. For instance, while some studies predominantly enroled patients with NYHA class III HF, others included a wider spectrum (NYHA II–IV). These differences introduce complexity in comparing RM modalities, partly due to varying absolute baseline risks among patient populations. However, NYHA class did not significantly influence outcomes in the meta-regression analysis. Finally, because few studies conducted direct, head-to-head comparisons of RM modalities, much of the comparative evidence relied on indirect data, introducing additional uncertainty into the estimates for these comparisons.

This meta-analysis confirms that RM overall significantly reduces all-cause mortality and HF-related hospitalisations. Notably, these benefits remained consistent across diverse patient and study characteristics, despite the presence of heterogeneity among studies. These findings support the broad implementation of RM in HF management; however, current evidence does not identify a specific patient subgroup that derives the greatest benefit from a chosen strategy.

Methods

Study design and registration

This review adheres to the Preferred Reporting Items of Systematic Reviews and Meta-Analysis (PRISMA) guidelines¹⁰¹ and has been registered in the International Prospective Register of Systematic Reviews (PROSPERO) database under record number CRD42024504260. The design of this review builds upon our previous meta-analysis, using similar methodology for article inclusion and selection, data collection, risk of bias assessment, and statistical modelling for meta-analysis⁵. In addition, this analysis includes several key modifications. To ensure accurate head-to-head comparisons between RM modalities and standard care, only randomised clinical trials (RCTs) were included in this analysis, allowing for direct comparisons between trial arms rather than reliance on observational data. A certainty of evidence assessment was conducted specifically for the NMA to evaluate the robustness of the comparative findings. A subgroup and meta-regression analysis was performed to explore the effects of patient and clinical study characteristics. This review is thus structured in three parts: (1) an update of our previous meta-analysis incorporating recent evidence on RM effectiveness, (2) a subgroup and meta-regression analysis examining the impact of patient and clinical study characteristics on RM outcomes, and (3) an NMA to compare the effectiveness of RM modalities compared to standard care.

Search strategy and study selection

A systematic literature search covering the period from 1996 to January 10, 2025, was conducted in Embase, Medline (Ovid), Web of Science, and Cochrane CENTRAL using predefined keywords related to HF and RM. Only peer-reviewed studies published in English were included if they assessed any form of RM, defined as a system used in the home setting to collect biometric or health-status-related data and remotely transmit it to a health care provider for assessment, in patients aged 18 years or older with chronic HF. Detailed information on the search strategy is described in detail in our previous review and in Supplementary Note 1⁵. A comprehensive overview of the definitions of RM modalities is provided in Table 3. This review included five RM modalities: telemonitoring (TM), structured telephone support (STS), complex telemonitoring (CTM), cardiac implantable electronic devices (CIEDs), and invasive hemodynamic monitoring (IHM). Inclusion and exclusion criteria are provided in Supplementary Table S1.

Table 3.

Definitions of different remote monitoring modalities

		Definitions
Non-invasive RM
TM	TeleMonitoring	The modality in which biometric data and/or health-related questionnaires are collected and sent to an HF clinic.
STS	Structural telephone support	A modality in which HF patients are called by a HF nurse or cardiologist on a frequent basis.
CTM	CTM	Modality in which multiple TM is combined with STS and/or a 24 h call centre or a mix of other sub-modalities.
Invasive hTMS
CIED	CIED	Modality in which PM/ICD systems (optionally with impedance leads) are used to monitor the patient.
IHM	IHM	Modality in which invasive hemodynamic parameters are used, e.g., (pressure) sensors.

RM remote monitoring, TM telemonitoring, STS structural telephone support, CTM complex telemonitoring, CIED cardiac implantable electronic devices, PM pacemaker, ICD intrinsic cardiac defibrillator, IHM invasive hemodynamic monitoring.

Eligibility criteria, data extraction, and risk of bias assessment

For the updated conventional meta-analysis, as well as subgroup and meta-regression analyses, only RCTs comparing at least one RM modality with standard care were included. For the NMA, RCTs that compared multiple RM modalities or at least one RM modality against standard care were included. Article screening, selection, data extraction, and risk of bias assessment using the RoB2 tool were conducted by two independent reviewers (N.T.B.S. and P.C.). Disagreements at any stage were resolved through consensus or, if needed, by a third independent reviewer (R.M.A.v.d.B.).

Endpoints and definitions of study and patient characteristics

The endpoints for all analyses were total HF hospitalisations, first HF hospitalisation, and all-cause mortality. For subgroup analyses, geographical region and HF status (unstable vs stable HF) were assessed. The geographical region subgroup included North America, Europe, Mid/South America, the Middle East, and Asia/Australia. Unstable HF was defined as a hospitalisation for HF within 6 months before intervention initiation or as specified in the study’s inclusion criteria. Stable HF was defined as discharge more than 6 months prior to inclusion or as described in the study’s inclusion criteria.

For study characteristics, continuous variables are presented as means with ranges, and categorical variables as counts with percentages. For patient characteristics per study arm, continuous variables were presented as weighted means with weighted standard deviations.

Statistical analysis: conventional meta-analysis, subgroup, and meta-regression

For the conventional meta-analyses, a random effects model with the DerSimonian and Laird variance estimator was used¹⁰². For first HF hospitalisation and all-cause mortality, pooled treatment effects were presented as risk ratios (RRs) with 95% confidence intervals (CIs). The endpoint total HF hospitalisations was presented as an incidence rate ratio (IRR) with 95% CI, requiring calculation of person-years. Similar to our previous meta-analysis, person-years were calculated using the mean or median follow-up time⁵. If neither was available, the planned follow-up time was used, except for patients who withdrew or died, for whom half of the planned follow-up time was used.

For subgroup analyses (geographical region and HF status), the same methodology described in the updated meta-analysis section was applied to each subgroup. For the meta-regression analysis, the following clinical study characteristics were assessed in univariate models: mean age, mean LVEF, percentage of patients in NYHA class III/IV, percentage of male participants, and publication year. Differential effects were evaluated by calculating p-values for interaction within the subgroup and meta-regression analyses. Given the number of subgroups and outcomes, p-values for interaction were adjusted to account for inflated type I error using the Holm procedure¹⁰³. Both unadjusted and adjusted p-values for interaction were presented^103–106. The subgroup and meta-regression analyses were conducted using the Metafor package for R¹⁰⁷.

Statistical analysis: NMA and certainty of evidence

For the NMA, a random-effects model with the DerSimonian and Laird variance estimator was also used¹⁰². Forest plots were used to compare each RM modality against standard care, and league tables summarised all head-to-head comparisons. RM modality hierarchies were calculated and compared using the surface under the cumulative ranking curve (SUCRA) score, based on 1000 simulations¹⁰⁸. A SUCRA score of 1.0 indicates the highest probability of being the most effective treatment, while a score of 0.0 represents the lowest. Notably, SUCRA estimates reflect comparisons against standard care rather than direct comparisons between RM modalities. A certainty of evidence assessment was conducted using the Confidence in Network Meta-Analysis (CINeMA) framework, which evaluates 6 domains:(1) within-study bias, (2) reporting bias, (3) indirectness, (4) imprecision, (5) heterogeneity, and (6) incoherence¹⁰⁹. Within-study bias was based on the RoB2 assessments, and the other domains were assessed according to the CINeMA framework recommendations¹¹⁰. Overall certainty of evidence was rated as very low when one domain was scored as major concerns, low when three domains were scored as some concerns, moderate when two domains were scored as some concerns, and high when one or no domains were scored as some concerns. The transitivity assumption was assessed within the indirectness (relevance of included studies for research question) and incoherence (Separating Indirect From Direct Evidence approach) domains, while the consistency assumption was assessed within the heterogeneity (comparing 95% CI with 95% predictive interval approach) domain¹¹¹. Publication bias was assessed with funnel plots and Egger regression tests and incorporated in the reporting bias domain of CINeMA. For the within-study bias domain, additional sensitivity analyses were conducted by excluding studies with a high risk of bias, as determined by the RoB2 tool. If the primary analysis remained consistent with the sensitivity analysis, the bias domain was rated as having no concerns. The Netmeta package for R was used for the NMA¹¹².

Author contributions

N.T.B.S. and P.R.D.C.: conceptualisation, methodology, investigation, writing—original draft, and visualisation. E.B., E.R., L.F., M.T.G., and R.M.A.v.d.B.: writing—review and editing. R.M.A.v.d.B. and J.J.B.: conceptualisation and supervision. All authors have read and approved the manuscript.

Data availability

The data underlying this article can be shared upon reasonable request to the corresponding author.

Competing interests

The institution of the authors Boer has received research grants and/or fees from Alnylam, AstraZeneca, Abbott, Bristol-Myers Squibb, NovoNordisk, and Roche. R.A.d.B.: has had speaker engagements with and/or received fees from and/or served on an advisory board for Abbott, AstraZeneca, Bristol Myers Squibb, NovoNordisk, Roche, and Zoll and received travel support from Abbott and NovoNordisk. R.M.A.v.d.B.: reports an independent research grant paid to the Institute from Abbott for IIS and has had speaker engagement in the past 5 years with Abbott, AstraZeneca, and Novartis. J.J.B:. reports an independent research grant paid to the Institute from Abbott for IIS and has had speaker engagement or advisory boards in the past 5 years with: AstraZeneca, Abbott, Boehringer-Ingelheim, Bayer, Daiichi Sankyo, Novartis and Vifor. All other authors declared to have no conflict of interest.

Supplementary information

The online version contains supplementary material available at 10.1038/s41746-026-02415-w.