Remote patient monitoring in heart failure: A comprehensive meta‐analysis of effective programme components for hospitalization and mortality reduction

Abstract

Aims

Methods of non‐invasive remote patient monitoring (RPM) for heart failure (HF) remain diverse. Understanding factors that influence the effectiveness of RPM on HF‐related and all‐cause hospitalizations, mortality, and emergency department visits is crucial for developing successful RPM interventions. This meta‐analysis aims to synthesize and compare existing literature on RPM components that impact HF‐related and all‐cause hospitalizations, mortality and emergency department visits in HF patients.

Methods and results

A systematic search of electronic databases (PubMed, EMBASE, CENTRAL) identified randomized controlled trials from January 2012 to June 2023, comparing non‐invasive RPM interventions for HF with usual care. A random‐effects meta‐analysis assessed outcomes, and additional analyses identified effective RPM components. A total of 41 studies with 16 312 patients (mean follow‐up: 9.88 ± 6.37 months) were included. RPM was associated with lower mortality risk (pooled odds ratio [OR] 0.81 95% confidence interval [CI] 0.69–0.95; I ² = 0.39) and reduced first HF hospitalization risk (pooled OR 0.78, 95% CI: 0.70–0.87; I ² = 0.21) compared to usual care. RPM interventions with a self‐management module (p < 0.001) and education module (p = 0.028) significantly lowered HF‐related hospitalizations. Video calls during RPM interventions further reduced HF‐related (p = 0.047) and all‐cause hospitalizations (p < 0.001).

Conclusion

This meta‐analysis confirms the efficacy of RPM in reducing HF‐related hospitalizations and mortality. Effective components include self‐management, education modules, and video communication. However, heterogeneity among interventions challenges the overall evaluation. Modernizing RPM with advanced technologies like non‐invasive sensors, artificial intelligence, and cardiac telerehabilitation could enhance its potential.

Introduction

Heart failure (HF) is one of the most prevalent cardiovascular diseases, with a prevalence of 1–2% of the general adult population in 2020. ¹ Moreover, the annual incidence of new HF patients continues to rise significantly, due to the ageing population, enhanced survival and earlier diagnosis. ² The disease's progressive course is associated with temporary exacerbations that lead to recurrent hospital admissions, ³ with up to 22% of all patients having a readmission within the first 30 days after discharge. ⁴ , ⁵ The substantial impact of HF on both the individual patient and the socio‐economic sphere has prompted efforts to identify means for predicting HF‐related hospitalizations by focusing on faster and better management of HF deterioration. ⁶ , ⁷

In recent decades, considerable research has been dedicated to exploring effects of remote patient monitoring (RPM) for managing HF patients. ⁸ , ⁹ , ¹⁰ RPM is a broad concept that can be defined in different ways. The Encyclopaedia of Health and Economics defined RPM as ‘a subset of telemedicine that includes technology in a patient's home that records biometric data and transmits it to a central monitoring facility for interpretation’. ¹¹ It is generally assumed that remote monitoring the health status of HF patients on a regular basis enables early detection of HF deterioration, giving caregivers the opportunity to intervene earlier in the cascade of deterioration. ¹² As a result, hospitalizations may be prevented. ¹⁰

Remote patient monitoring can be conducted in two manners, namely invasive or non‐invasive. Studies employing invasive monitoring approaches have demonstrated a notable reduction in hospitalizations. ¹³ , ¹⁴ However, the field of invasive sensors is still small, with a low number of validated sensors, ¹⁵ and limited implementation in clinical practice due to high costs and resources. On the other hand, research on non‐invasive RPM in HF patients has yielded conflicting results. While some studies report a decrease in HF‐related hospitalizations, length of stay and all‐cause mortality, ¹⁶ , ¹⁷ , ¹⁸ others do not support these conclusions. ⁹ , ¹⁹ This inconsistency may be explained by differences in study populations, time of inclusion, approaches to integrate telemedicine into existing care pathways, and the parameters employed to guide HF treatment. ⁹ , ¹⁰ In a systematic review, Ding et al. ²⁰ demonstrated that integrating mHealth and medication support into RPM interventions significantly enhances outcomes related to rehospitalization and mortality. The effect of other components on HF‐related hospitalizations were not explored in this study. Further investigation into the factors determining effective RPM interventions is necessary to advance our understanding, and to provide clinicians and researchers with a blueprint for developing and implementing successful RPM interventions.

The objective of this meta‐analysis is to provide a comprehensive overview of the existing literature, and to identify the effective elements of non‐invasive RPM interventions in the management of HF patients.

The primary outcomes of interest include HF‐related hospitalizations, mortality, all‐cause hospitalizations and emergency department (ED) visits. Secondary outcomes of interest include quality‐of‐life (QoL) and adherence. Through the analysis of available randomized controlled trials (RCTs), this review aims to provide clear insights into the elements contributing to the success of non‐invasive RPM interventions in improving patient outcomes and to inform future research and clinical practice in the field of HF management.

Methods

Registration

A meta‐analysis was performed using the Preferred Reporting Items for Systematic Reviews and Meta‐analysis (PRISMA) guidelines. ²¹ The study was registered in the International Prospective Register of Systematic Reviews (Prospero, ID: CRD42022324015).

Eligibility criteria

For the purpose of this meta‐analysis, we have defined RPM as a technology‐based intervention that transmits cardiovascular parameters and/or HF‐related symptoms to a healthcare provider at least once a week through a phone or web connection. We included RCTs, comparing RPM with usual care (UC), that reported direct or indirect measurements of HF‐related hospitalizations, mortality, all‐cause hospitalizations and/or ED visits. Studies with a mean follow‐up of <3 months or that were not published in English were excluded. Follow‐up time is defined as the duration between patient inclusion and the collection of results, regardless of whether the patient discontinued telemonitoring or if the telemonitoring programme stopped before the end of the follow‐up period. To keep the focus of this meta‐analysis on non‐invasive RPM interventions, papers discussing invasive interventions were excluded. Table  1 presents a summary of all inclusion and exclusion criteria.

Table 1.

Inclusion and exclusion criteria

Inclusion criteria

– RCTs, comparing RPM and usual care, which include:

• Subjects >18 years, having been diagnosed with HF, regardless of type or aetiology.

• A non‐invasive RPM strategy which assessed cardiovascular parameters and/or HF‐related symptoms, which are digitally shared at least once weekly.

• At least one of the following outcome measures: HF‐related hospitalizations, mortality, all‐cause hospitalizations or ED visits. In case of combined outcomes, effort was taken to separate the outcomes.

Exclusion criteria

– Research paper not available in English

– Intervention with invasive monitoring devices

– Follow‐up duration <3 months

ED, emergency department; HF, heart failure; RCT, randomized controlled trial; RPM, remote patient monitoring.

Information resources and search strategy

A comprehensive search strategy was designed for PubMed, EMBASE, and CENTRAL, executed in January 2022 with a 10‐year time filter. Relevant, standardized keywords were picked from the MeSH database. Additionally, a filter for RCTs was applied, aligning with the targeted study design. In June 2023, a second search was conducted to capture any potential new trials. To enhance inclusivity, articles from the latest meta‐analysis on RPM were incorporated into our search. ¹⁰ The meta‐analysis covered the period from 1996 to July 2022. A detailed description of our search strategy can be found in online supplementary Appendix A .

Selection and data collection process

Titles and abstracts of all papers underwent independent screening by the two primary researchers (ILJDL and WWN) to determine eligibility. Interrater reliability and Cohen's kappa coefficient for the pre‐screening were calculated to check for agreement between researchers on inclusion. The eligible studies subsequently underwent independent full‐text screening conducted by the two primary researchers. For the included studies, data on study population characteristics, RPM type and strategy, and outcome measures were extracted by the two primary researchers independently. As measure of effect, the odds ratio (OR) was chosen, since this was provided in most of the studies as primary outcome. The ORs were recorded when available and calculated when necessary. Points of disagreement between the two researchers during any steps of the data collection were discussed to reach consensus. In case no consensus was reached, a third investigator (HMCK) was consulted.

Data extraction

The following data were extracted from the included papers: author, year of publication, journal, study name, country of study, HF type, New York Heart Association (NYHA) classification, left ventricular ejection fraction, patient comorbidities, medication use, time of inclusion respective to hospital discharge for HF, follow‐up time, modality with which RPM was delivered (i.e. mobile phone app, phone calls, web‐based application), additional module of RPM (i.e. self‐management, education, medication), self‐reported parameters, frequency of RPM, if feedback based on monitored parameters was immediately visible for the patient and/or the practitioner, if there was any additional contact with the patient with a different purpose than parameter monitoring (and in what form), and if there was an additional motivator (e.g. money) for participation in RPM.

To facilitate the classification of RPM, the following categories were identified: modality, self‐reported parameters, and additional modules. RPM modality was categorized into mobile applications, video calls, phone calls, and web applications. Self‐reported parameters included weight, blood pressure (BP), heart rate (HR), symptom questionnaires, diverse questionnaires, activity, and electrocardiography (ECG). Diverse questionnaires were classified as any questionnaires not related to HF. Additional modules were categorized as education, medication management, and self‐management, in which ‘education’ is classified as information related to HF, and ‘self‐management’ to information related to living with HF. Multiple options could be selected for each category. In instances where pre‐defined options did not align with the modality, goal, or self‐reported parameters, or if the pre‐defined options did not comprehensively cover the characteristic, researchers provided a brief description.

The primary outcomes collected, when accessible, encompassed the number of patients (re)hospitalized for HF, the number of patients experiencing all‐cause hospitalization, the number of patients undergoing ED visits, and mortality rates. Secondary outcomes comprised assessments of QoL, irrespective of the metric employed, and adherence to RPM, regardless of the specific measure utilized to report this outcome. In cases where any of the aforementioned outcomes were not reported in a given article, the corresponding value was deemed as missing for the purposes of this meta‐analysis.

Study risk of bias assessment

The recommended instrument for evaluating bias in RCTs, the second version of the Cochrane Risk‐of‐Bias tool (RoB 2), was employed. Risk of bias assessment was conducted independently by the two primary researchers (ILJDL and WWN). Subsequently, the independent assessments were examined for any discrepancies. In instances of disagreement, the studies and their respective assessments were deliberated upon until a consensus was reached. If no consensus was reached, the paper was discussed with a third investigator (HMCK).

Synthesis methods

Forest plots are used to visualize the measures of effect across studies together with their uncertainty. A random‐effect meta‐analysis is performed for each outcome of interest to compare the effects of RPM and UC across the included studies, while accounting for heterogeneity. The heterogeneity is also quantified using the I ² statistic and will be categorized as follows, based on the Cochrane handbook of for systematic reviews; 0–40%: may not be important, 30–60%: may represent moderate heterogeneity, 50–90%: may represent substantial heterogeneity, and 75–100%: considerable heterogeneity. ²² Funnel plots are made and evaluated for asymmetry to assess potential publication bias. For each component of RPM, the ORs for the two subgroups were pooled separately, and compared statistically by means of their 95% confidence intervals (CIs) and p‐value (based on z‐score). A p‐value <0.05 was considered as statistically significant. Due to large differences in reporting both QoL and adherence, no quantitative analyses were performed on these outcome measures. Continuous variables are reported as means and standard deviations if the data were normally distributed, or as median and interquartile ranges if not. Categorical variables are reported as counts and percentages. SPSS (IBM Corp, released 2021; IBM SPSS Statistics for Windows, version 29.0. Armonk, NY, USA) was used for analysis.

Results

Study selection

Our search strategy yielded a total of 3966 studies (online supplementary Figure  A ). Of these, 140 were removed as duplicates. A further 3690 articles were excluded based on the title and abstract screening. Of 34 articles, full texts were not available. Cohen's k was 0.75 for the pre‐screening, which constitutes substantial agreement. In total, 102 articles underwent full‐text screening. The primary reasons for article exclusion were secondary or post‐hoc analysis from other trials (21 excluded, 20.3%), non‐compliance with our definition of RPM (17 excluded, 16.5%), or did not include the outcome measure of interest (12 excluded, 11.7%). Following screening, a total of 23 eligible articles remained. Additionally, we incorporated the findings from a recent meta‐analysis on RPM to our search. ¹⁰ Of the 57 studies included, 18 were eligible after full‐text screening. This brings the total number of included studies for this systematic review and meta‐analysis to 41.

Risk of bias in studies

A high risk of bias was found for 22% of the studies. The most frequent cause was the risk of bias in the domain ‘deviations from the intervention’ (online supplementary Table  A and Figure  B ).

Study characteristics

A total of 41 studies were included, of which two studies had implemented two different interventions ¹⁸ , ²³ and one had implemented three different interventions, ²⁴ resulting in a total of 45 distinct interventions for our analysis (online supplementary Table  B ). Approximately half of these studies (22; 53.7%) were conducted in Europe, and 14 (34.1%) were conducted in the United States. Among the included studies, 10 (24.4%) were conducted between 2000 and 2010, 22 (53.7%) between 2010 and 2020, and 9 (22.0%) between 2020 and July 2023, indicating a consistent upward trend in research on RPM in HF care. The mean follow‐up was 9.88 ± 6.37 months.

The majority of the included studies (28; 68.3%) encompassed all types of HF patients (i.e. HF with reduced [HFrEF], mildly reduced [HFmrEF] and preserved ejection fraction [HFpEF]). Twelve studies (29.3%) exclusively enrolled HFrEF patients, while only one study included both HFrEF and HFmrEF patients. Notably, no study focused solely on HFpEF patients. Across all studies, a total of 16 312 patients were included, with 8607 in the intervention group (68.7% male, mean age: 67.84 years) and 7905 in the control group (68.6% male, mean age: 68.11 years) Participants were included at or within 1 month after discharge for acute decompensated HF hospitalization in 19 studies (46.3%), while in 6 studies (13.3%) no recent acute decompensated HF episode or hospitalization was required for inclusion. Table  2 gives a more detailed description of the populations incorporated in the studies. ¹⁸ , ²³ , ²⁴ , ²⁵ , ²⁶ , ²⁷ , ²⁸ , ²⁹ , ³⁰ , ³¹ , ³² , ³³ , ³⁴ , ³⁵ , ³⁶ , ³⁷ , ³⁸ , ³⁹ , ⁴⁰ , ⁴¹ , ⁴² , ⁴³ , ⁴⁴ , ⁴⁵ , ⁴⁶ , ⁴⁷ , ⁴⁸ , ⁴⁹ , ⁵⁰ , ⁵¹ , ⁵² , ⁵³ , ⁵⁴ , ⁵⁵ , ⁵⁶ , ⁵⁷ , ⁵⁸ , ⁵⁹ , ⁶⁰ , ⁶¹ , ⁶²

Table 2.

Detailed overview of patient characteristics of the included studies (n = 41)

Author	Age (years)	Sex (M/F, n)	HF type	LVEF	NYHA class	RPM (n)	UC (n)	Time of inclusion
Angermann et al. 25	RPM: 67.7, UC: 69.4	505/210	HFrEF	≤40%	I–IV	352	363	At discharge
Asch et al. 26	RPM: 64.9, UC: 64.2	290/262	HFrEF, HFmrEF, HFpEF	All, 42.43 ± 17.96%	I–IV	272	280	Within 1 month after discharge
Bekelman et al. 27	RPM: 67.3, UC: 67.9	371/13	HFrEF, HFmrEF, HFpEF	All	I–IV	187	197	Within 12 months after discharge
Boyne et al. 28	RPM: 71, UC: 71.9	226/156	HFrEF, HFmrEF, HFpEF	All	II–IV	197	185	At discharge
Chaudhry et al. 29	RPM: 61, UC: 61	958/695	HFrEF, HFmrEF HFpEF	All	II–IV	826	827	Within 1 month after discharge
Cleland et al. 30	RPM: 67, UC: 68	204/44	HFrEF	<40%	I–IV	163	85	At discharge
Comín‐Colet et al. 31	RPM: 74, UC: 75	100/78	HFrEF, HFmrEF HFpEF	All	I–IV	81	97	Within 3 months after discharge
Dar et al. 32	RPM: 72, UC: 70	121/61	HFrEF, HFmrEF, HFpEF	All	II–IV	91	91	At discharge
Delaney et al. 33	RPM: /, UC: /	28/65	HFrEF, HFmrEFHFpEF	All	III, IV	46	47	Unclear
Dendale et al. 34	RPM: 77, UC: 77	104/58	HFrEF, HFmrEF, HFpEF	All	I–IV	81	81	At discharge
Dorsch et al. 35	RPM: 60.2, UC: 62	54/29	HFrEF HFmrEF, HFpEF	All	I–IV	42	41	Within 1 month after discharge
Galinier et al. 36	RPM: 70, UC: 69.7	677/260	HFrEF, HFmrEF, HFpEF	All, RPM: 39.3 ± 14.5%, UC: 38.1 ± 15.2%	I–IV	482	455	Within 12 months after discharge
Galinier et al. 37	RPM (1): 66.6, RPM(2): 60.8, UC: 66.2	328/86	HFrEF, HFmrEF, HFpEF	All	II–IV	99, 220	95	Within 12 months after discharge
Garanin et al. 23	RPM: 66.3, UC: 66.2	203/189	HFrEF	<40%	I–IV	197	195	At discharge
Giordano et al. 38	RPM: 58, UC: 56	391/69	HFrEF	<40%	II–IV	230	230	At discharge
Goldberg et al. 39	RPM: 57.9, UC: 60.2	189/91	HFrEF	<35%	III, IV	138	142	At discharge
Guo et al. 40	RPM: 72.8, UC: 72.6	429/285	HFrEF, HFmrEFHFpEF	All	I–IV	360	354	Unclear
Idris et al. 41	RPM: 58.5, UC: 66.5	11/17	HFrEF	<35%	II, III	14	14	Within 6 months after discharge
Kalter‐Leibovici et al. 42	RPM: 70.8, UC: 70.7	986/374	HFrEF, HFmrEF, HFpEF	All	I–IV	682	678	Within 2 months after discharge
Kashem et al. 43	RPM: 54, UC: 53	35/13	HFrEF, HFmrEF, HFpEF	All	II–IV	24	24	Within 6 months after discharge
Koehler et al. 44	RPM: 66.9 UC: 66.9	577/133	HFrEF	≤35%	II, III	354	356	Within 2 years after discharge
Koehler et al. 45	RPM: 70, UC: 70	1192/346	HFrEF, HFmrEF, HFpEF	≤45% or ≥45%, + diuretics	I–IV	765	773	Within 12 months after discharge
Kotooka et al. 18	RPM: 67.1, UC: 65.4	107/74	HFrEF, HFmrEF, HFpEF	All	II, III	90	91	Within 1 month after discharge
Lynga et al. 46	RPM: 73.7, UC: 73.5	239/80	HFrEF, HFmrEF	<50%	III, IV	166	153	Within 2 months after discharge
Melin et al. 47	RPM: 75, UC: 76	49/23	HFrEF, HFmrEF, HFpEF	All	II, III	32	40	Within 1 month after discharge
Mizukawa et al. 48	RPM (1): 69.4, RPM (2): 70.5, UC: 74.5	35/22	HFrEF, HFmrEFHFpEF	All	II–IV	18, 20	19	Within 2 years after discharge
Mortara et al. 49	RPM (1): 60, RPM (2): 60, RPM (3): 59, UC: 60	393/70	HFrEF	<40%	II–IV	10 696, 101	160	Within 12 months after discharge
Ong et al. 24	RPM: 73, UC: 74	212/127	HFrEF, HFmrEF, HFpEF	All	I–IV	715	722	At discharge
Olivari et al. 50	RPM: 79.6, UC: 80.9	764/673	HFrEF, HFmrEF, HFpEF	All, <40% or >40% + BNP >400 pg/ml or NT‐proBNP >1500 pg/ml	II–IV	229	110	Within 3 months after discharge
Pekmezaris et al. 51	RPM: 58.4, UC: 61.1	35/55	HFrEF, HFmrEF, HFpEF	All	I–III	46	58	At discharge
Pedone et al. 52	RPM: 79.9, UC: 79.7	61/43	HFrEF, HFmrEF, HFpEF	All, IG: 44.4 ± 12.7%, UC: 48.2 ± 13.5%	II–IV	47	43	Unclear
Sahlin et al. 53	RPM: 80, UC: 77	71/47	HFrEF, HFmrEF, HFpEF	All	I–IV	58	60	Within 12 months after discharge
Scherr et al. 54	RPM: 65, UC: 67	79/29	HFrEF, HFmrEF, HFpEF	All	II–IV	54	54	Within 1 month after discharge
Seto et al. 55	RPM: 55.1, UC: 52.3	79/21	HFrEF	<40%	II–IV	50	50	At discharge
Shara et al. 56	RPM: 54.5, UC: 52.9	28/22	HFrEF, HFmrEF, HFpEF	All	I–IV	20	30	At discharge
Soran et al. 57	RPM: 76.9, UC: 76	111/204	HFrEF	<40%	II, III	160	155	Within 6 months after discharge
Upshaw et al. 58	RPM: 68, UC: 74	150/62	HFrEF, HFmrEFHFpEF	All	I–IV	159	53	Unclear
Völler et al. 59	RPM: 62.5, UC: 63.2	430/62	HFrEF	<40%	II–IV	241	251	Within 12 months after discharge
Vuorinen et al. 60	RPM: 57.9, UC: 58.3	78/16	HFrEF	<35%	II–IV	47	47	Within 6 months after discharge
Wagenaar et al. 61	RPM: 66.6, UC: 66.9	222/78	HFrEF, HFmrEF, HFpEF	All, 35.7 ± 10.8%	I–IV	150	150	Within 3 months after discharge
Wakefield et al. 62	RPM: 70.3, UC: 67.2	147/1	HFrEF, HFmrEF, HFpEF	All	II–IV	99	49	At discharge

BNP, B‐type natriuretic peptide; F, female; HF, heart failure; HFmrEF, heart failure with mildly reduced ejection fraction; HFpEF, heart failure with preserved ejection fraction; HFrEF, heart failure with reduced ejection fraction; IG, intervention group; LVEF, left ventricular ejection fraction; M, male; NT‐proBNP, N‐terminal pro‐B‐type natriuretic peptide; NYHA, New York Heart Association; RPM, remote patient monitoring; UC, usual care.

Remote patient monitoring interventions were categorized based on three categories: modalities, parameters, and additional modules (Table  3 ). ¹⁸ , ²³ , ²⁴ , ²⁵ , ²⁶ , ²⁷ , ²⁸ , ²⁹ , ³⁰ , ³¹ , ³² , ³³ , ³⁴ , ³⁵ , ³⁶ , ³⁷ , ³⁸ , ³⁹ , ⁴⁰ , ⁴¹ , ⁴² , ⁴³ , ⁴⁴ , ⁴⁵ , ⁴⁶ , ⁴⁷ , ⁴⁸ , ⁴⁹ , ⁵⁰ , ⁵¹ , ⁵² , ⁵³ , ⁵⁴ , ⁵⁵ , ⁵⁶ , ⁵⁷ , ⁵⁸ , ⁵⁹ , ⁶⁰ , ⁶¹ , ⁶² Out of the 45 included interventions, 19 (42.2%) used a mobile application for RPM, and 18 (40.0%) integrated phone or video calls (13 phone calls, 2 video calls, and 3 both). A web application was employed in only 3 (6.7%) of the RPM interventions. Weight was the most frequently monitored parameter, featuring in 42 (93.3%) out of 45 interventions. BP was measured in 31 (68.9%) interventions and HR in 28 (62.2%). ECG monitoring was employed in only 7 (15.6%) interventions, while peripheral oxygen saturation was monitored in 8 (17.8%). Symptom questionnaires were utilized in 28 (62.2%) interventions, and diverse questionnaires in 6 (13.3%). The monitoring frequency was daily or more often in 38 (84.4%) interventions and weekly or less often in the remaining 7 (15.6%). Continuous monitoring was only done in two interventions. Guo et al. ⁴⁰ used a photoplethysmography wristband to continuously monitor HR. Mortara et al. ⁴⁹ monitored ECG, respiration, and physical activity using a portable device. A total of 20 (44.4%) interventions included additional modules focused on HF education, 13 (28.9%) on self‐management, and 11 (24.4%) on medication use.

Table 3.

Detailed description of the remote patient monitoring interventions used in the studies (n = 45)

Author	Modality					Parameters								Additional module
	Mobile App	Phone calls	Video calls	Web App	Other	Weight	Blood pressure	Heart rate	SpO2	ECG	Symptom questionnaire	Diverse questionnaire	Further parameters	Education	Medication	Self‐ management
Angermann et al. 25	✓	✓				✓	✓	✓			✓	✓		✓	✓	✓
Asch et al. 26	✓					✓									✓
Bekelman et al. 27	✓					✓	✓	✓			✓	✓		✓	✓
Boyne et al. 28					✓						✓	✓		✓		✓
Chaudhry et al. 29		✓				✓					✓	✓
Cleland et al. 30						✓	✓	✓		✓
Comin‐Colet et al. 31					✓	✓	✓	✓			✓			✓		✓
Dar et al. 32					✓	✓	✓	✓	✓		✓
Delaney et al. 33					✓	✓	✓	✓	✓		✓				✓	✓
Dendale et al. 34	✓					✓	✓	✓
Dorsch et al. 35	✓				✓	✓	✓									✓
Galinier et al. 36	✓	✓				✓					✓			✓		✓
Galinier et al. 37 (1)	✓	✓				✓					✓			✓
Galinier et al. 37 (2)	✓	✓	✓			✓					✓			✓
Garanin et al. 23					✓		✓	✓
Giordano et al. 38		✓				✓	✓			✓	✓			✓	✓	✓
Goldberg et al. 39		✓			✓	✓					✓
Guo et al. 40	✓				✓	✓	✓	✓	✓					✓	✓
Idris et al. 41	✓		✓			✓	✓	✓	✓
Kalter‐Leibovici et al. 42		✓	✓		✓	✓	✓	✓						✓	✓	✓
Kashem et al. 43				✓		✓	✓	✓					Activity
Koehler et al. 44					✓	✓	✓	✓		✓
Koehler et al. 45	✓					✓	✓	✓	✓	✓	✓			✓		✓
Kotooka et al. 18					✓	✓	✓	✓
Lynga et al. 46		✓			✓	✓					✓
Melin et al. 47	✓				✓	✓					✓			✓	✓
Mizukawa et al. 48 (1)					✓	✓	✓	✓						✓		✓
Mizukawa et al. 48 (2)					✓	✓	✓	✓						✓		✓
Mortara et al. 49 (1)		✓									✓
Mortara et al. 49 (2)		✓				✓	✓	✓			✓
Mortara et al. 49 (3)		✓			✓	✓	✓	✓		✓	✓		Activity
Ong et al. 24	✓	✓				✓	✓	✓			✓
Olivari et al. 50					✓	✓	✓	✓	✓	✓				✓
Pekmezaris et al. 51					✓	✓	✓	✓	✓
Pedone et al. 52	✓	✓				✓	✓	✓	✓
Sahlin et al. 53	✓				✓	✓					✓			✓	✓	✓
Scherr et al. 54	✓			✓		✓	✓	✓							✓
Seto et al. 55	✓					✓	✓			✓	✓
Shara et al. 56					✓	✓					✓	✓		✓	✓
Soran et al. 57					✓	✓					✓
Upshaw et al. 58	✓	✓				✓	✓	✓			✓			✓
Völler et al. 59	✓					✓	✓	✓			✓			✓
Vuorinen et al. 60	✓					✓	✓	✓			✓	✓
Wagenaar et al. 61	✓			✓		✓	✓	✓						✓
Wakefield et al. 62		✓	✓			✓	✓	✓					Oedema			✓

ECG, electrocardiogram; SpO₂, peripheral oxygen saturation.

Results of individual studies

Table  4 summarizes the total amount of interventions and studies for each outcome. It further shows the number of patients within those studies, which were used for analyses related to that outcome. Heterogeneity is also reported.

Table 4.

Summary of the number of interventions, studies, patients, and heterogeneity used in the analyses for each outcome

Outcome	Interventions (n)	Studies (n)	Patients (n)	RPM (n)	UC (n)	Heterogeneity (%)
HF‐related hospitalizations	34	31	11 904	6369	5535	21
Mortality	33	31	14 391	7358	6961	41
All‐cause hospitalizations	35	31	11 944	6499	5445	70
ED visits	7	7	1528	804	714	0

ED, emergency department; HF, heart failure; RPM, remote patient monitoring; UC, usual care.

Heart failure‐related hospitalizations

Heart failure‐related hospitalizations were documented in 34 interventions encompassing 31 studies. A statistically significant reduction of 22% (pooled OR 0.78, CI: 0.70–0.87) was observed for the first HF‐related hospitalization between RPM and UC (Figure  1 ). The degree of heterogeneity (I ²) was 21% and thus considered as not important. Assessing the funnel plot, we assume the risk of publication bias to be low (online supplementary Figure  C1 ).

Figure 1.

Forest plot of heart failure‐related hospitalizations with subgroup analysis for the inclusion of a self‐management module in the remote patient monitoring intervention.

Modality of remote patient monitoring

A significant difference was found in HF‐related hospitalizations between interventions that included video calls (pooled OR 0.54, 95% CI 0.28–1.04) and interventions that did not (pooled OR 0.80, 95% CI 0.72–0.89; Z = 1.97, p = 0.047). Integration of the other modalities did not lead to significant differences as compared to RPM programmes that did not.

Parameters

Remote patient monitoring interventions including monitoring of weight, HR, HF symptoms, and ECG did not reveal a greater effect on HF‐related hospitalizations than interventions that did not include these parameters. A significant difference was found between interventions that implemented diverse questionnaires (n = 5; pooled OR 0.96, 95% CI 0.79–1.18) and interventions that did not (n = 29; pooled OR 0.68, 95% CI 0.68–0.85), with interventions without a diverse questionnaire showing better outcomes (Z = 2.153, p = 0.031).

Additional modules

A self‐management module, focused on how to live with HF, was incorporated in 13 out of the 34 interventions. The addition of a self‐management module to the RPM intervention yielded a significant positive outcome on HF‐related hospitalizations (self‐management module: pooled OR 0.63, 95% CI 0.50–0.79 vs. no self‐management module: pooled OR 0.89, 95% CI 0.79–1.01; Z = 3.721, p < 0.001) (Figure  1 ).

Remote patient monitoring interventions including an education module (13 of 34 interventions) demonstrated a superior effect on HF‐related hospitalizations (pooled OR 0.69, 95% CI 0.57–0.84) as compared to interventions without an education module (pooled OR 0.85, 95% CI 0.74–0.97; Z = 2.146, p = 0.028). No other significant differences between additional modules were found for this outcome measure.

Finally, RPM interventions that included optional contact with their caregivers, or other care‐providing parties, like dieticians or psychologists (8 out of 34 interventions) showed a lower effect on HF‐related hospitalizations than those without optional contacts (pooled OR 0.85, 95% CI 0.75–0.96 vs. pooled OR 0.72, 95% CI 0.61–0.85; Z = 2.040, p = 0.041).

The ORs for other components, which demonstrated no significant effect on HF‐related hospitalizations, and p‐values for testing statistical significance of the differences, can be found in Table  5 .

Table 5.

Pooled odds ratios for each component and for each outcome

Component	Pooled OR included	Pooled OR excluded	p‐value
HF‐related hospitalization
Mobile App	0.74 (0.6–0.92)	0.79 (0.69–0.90)	0.503
Video call	0.54 (0.28–1.04)	0.80 (0.72–0.89)	0.048
Phone call	0.82 (0.70–0.96)	0.75 (0.67–0.87)	0.286
Web App	0.56 (0.29–1.07)	0.79 (0.71–0.88)	0.091
Weight	0.79 (0.71–0.88)	0.66 (0.39–1.12)	0.347
Blood pressure	0.74 (0.64–0.86)	0.85 (0.72–1.01)	0.136
Heart rate	0.82 (0.73–0.91)	0.76 (0.62–0.92)	0.395
SpO2	0.88 (0.73–1.05)	0.76 (0.67–0.86)	0.123
ECG	0.80 (0.67–0.96)	0.76 (0.66–0.88)	0.581
Symptom questionnaire	0.81 (0.70–0.93)	0.72 (0.59–0.87)	0.203
Diverse questionnaire	0.96 (0.79–1.18)	0.76 (0.68–0.85)	0.031
Education	0.69 (0.57–0.84)	0.85 (0.74–0.97)	0.028
Medication	0.69 (0.51–0.93)	0.81 (0.73–0.91)	0.169
Self‐management	0.63 (0.50–0.79)	0.89 (0.79–1.01)	<0.001
Further contact	0.85 (0.75–0.96)	0.72 (0.61–0.85)	0.041
Frequency >1/day	0.80 (0.72–0.90)	0.75 (0.57–0.98)	0.572
Mortality
Mobile App	0.71 (0.50–1.01)	0.88 (0.73–1.05)	0.162
Video call	0.90 (0.37–2.19)	0.80 (0.69–0.92)	0.801
Phone call	0.81 (0.56–0.97)	0.86 (0.69–1.07)	0.674
Web App	1.54 (0.41–5.72)	0.8 (0.68–0.94)	0.878
Weight	0.80 (0.68–0.95)	0.93 (0.34–2.48)	0.784
Blood pressure	0.83 (0.70–1.00)	0.74 (0.52–1.06)	0.466
Heart rate	0.83 (0.68–1.01)	0.78 (0.58–1.04)	0.667
SpO2	0.81 (0.52–1.26)	0.80 (0.67–0.97)	0.951
ECG	0.81 (0.64–1.02)	0.79 (0.63–0.99)	0.851
Symptom questionnaire	0.76 (0.59–0.98)	0.96 (0.81–1.14)	0.730
Diverse questionnaire	0.76 (0.45–1.29)	0.81 (0.68–0.98)	0.779
Education	0.87 (0.65–1.17)	0.80 (0.65–0.97)	0.590
Medication	0.78 (0.57–1.06)	0.82 (0.66–1.01)	0.745
Self‐management	0.93 (0.70–1.24)	0.74 (0.59–0.92)	0.157
Further contact	0.72 (0.54–0.96)	0.86 (0.68–1.07)	0.224
Frequency >1/day	0.80 (0.66–0.98)	0.80 (0.55–1.14)	1.000
All‐cause hospitalization
Mobile App	0.81 (0.58–1.15)	0.98 (0.84–1.13)	0.243
Video call	0.35 (0.25–0.51)	1.01 (0.92–1.10)	<0.001
Phone call	0.83 (0.65–1.07)	0.96 (0.78–1.17)	0.320
Web App	0.69 (0.39–1.19)	0.92 (0.78–1.07)	0.185
Weight	0.88 (0.75–1.03)	1.32 (0.92–1.89)	0.561
Blood pressure	0.88 (0.73–1.06)	0.94 (0.71–1.25)	0.683
Heart rate	0.89 (0.76–1.06)	0.95 (0.71–1.25)	0.687
SpO2	0.72 (0.39–1.33)	0.93 (0.80–1.08)	0.306
ECG	1.15 (0.82–1.62)	0.85 (0.72–1.01)	0.916
Symptom questionnaire	0.95 (0.77–1.16)	0.84 (0.67–1.06)	0.382
Diverse questionnaire	1.10 (0.95–1.28)	0.86 (0.72–1.03)	0.631
Education	0.78 (0.62–0.97)	1.03 (0.86–1.24)	0.097
Medication	0.82 (0.62–1.08)	0.93 (0.78–1.12)	0.388
Self‐management	0.77 (0.59–1.02)	0.96 (0.8–1.15)	0.119
Further contact	0.97 (0.84–1.11)	0.90 (0.75–1.09)	0.499
Frequency >1/day	0.88 (0.73–1.05)	1.00 (0.78–1.29)	0.405

ECG, electrocardiogram; OR, odds ratio; SpO₂, peripheral oxygen saturation.

Mortality

Mortality was documented across 33 interventions spanning 31 studies. A significant reduction of 19% (pooled OR 0.81, 95% CI 0.69–0.95) was observed between RPM and UC (Figure  2 ). The degree of heterogeneity (I ²) was 41% and thus may be important to take into consideration. Assessing the funnel plot, we assume the risk of publication bias to be low (online supplementary Figure  C2 ). In all categories, no additional component caused a significantly different effect on mortality. ORs for these components and the results from the test comparing interventions with and without component can be found in Table  5 .

Figure 2.

Forest plot of mortality.

All‐cause hospitalizations

In 31 studies, a total of 35 interventions documented all‐cause hospitalizations. A pooled analysis of these interventions showed no significant effect of RPM as compared to UC (pooled OR 0.90, 95% CI 0.78–1.05) (Figure  3 ). The degree of heterogeneity (I ²) was 70% and thus represents considerable heterogeneity. Assessing the funnel plot, we assume the risk of publication bias to be low (online supplementary Figure  C3 ).

Figure 3.

Forest plot of all‐cause hospitalizations with subgroup analysis for the inclusion of a video call in the remote patient monitoring intervention.

Modality of remote patient monitoring

In three out of 35 interventions, video calls were employed as a means of communication. The utilization of video calls demonstrated a substantial impact on all‐cause hospitalizations compared to interventions without this option (pooled OR 0.35, 95% CI 0.25–0.51 vs. pooled OR 1.01, 95% CI 0.92–1.10). These pooled ORs differed significantly from one another (Z = 12.890, p < 0.001) (Figure  3 ).

No further significant differences were found in the category's parameters and additional modules. ORs and p‐values for components which proved no significant differences can be found in Table  5 .

Emergency department visits

Only seven studies (and interventions) reported ED visits. A pooled analysis revealed no significant effect of RPM on ED visits as compared to UC (pooled OR 1.01, 95% CI 0.81–1.26) (Figure  4 ). The degree of heterogeneity (I ²) was 0%. Assessing the funnel plot, we assume the risk of publication bias to be low (online supplementary Figure  C4 ). No further subgroup analyses were performed on this outcome measure due to the small number of interventions that reported this outcome.

Figure 4.

Forest plot of the effect of the inclusion of remote patient monitoring on emergency department visits.

Table  5 shows a summary of all findings, with pooled ORs for each component and for each outcome. It also shows the p‐values comparing the pooled ORs of interventions including the component and interventions excluding the component.

Secondary outcomes: quality of life and adherence

Quality of life and adherence were both reported in 16 out of the 41 studies. However, large differences in quantifying these outcomes exist. For quantifying (changes in) QoL the following metrics were used: Patient Health Questionnaire‐9, Minnesota Living with Heart Failure Questionnaire, Short Form‐12, Health Distress Scale, Short Form‐36, Kansas City Cardiomyopathy Questionnaire. Next to different metrics, different moments in relation to baseline measurements of QoL were chosen, usually coinciding with the end of the follow‐up period of the respective study. When significant differences were reported, the differences between RPM and UC were always in favour of RPM, showing improvement in QoL, regardless of the metric and time of measurement.

Adherence was also quantified in several ways as no accepted metric exists in the literature. Examples of ways to quantify adherence ranged between percentage of patients who interacted with the system at least once every 3 days, ³⁹ to a cumulative amount of data points received. ⁵⁴ Differences in quantification make it difficult to draw proper conclusions of which components of RPM lead to higher adherence.

Discussion

This meta‐analysis shows a significant, positive effect of RPM on HF‐related hospitalizations and mortality, further strengthening previous findings. ¹⁰ , ⁶³ Additional important findings were that RPM interventions employing a self‐management module, or an education module, show a superior reduction in HF‐related hospitalizations compared to RPM interventions that do not. Therefore, these modules should be considered as key elements in future RPM interventions. Moreover, a superior effect on HF‐related hospitalizations and all‐cause hospitalizations was observed in interventions using video calls as a modality. Conversely, increased or extra contact with caregivers and the use of diverse questionnaires to monitor HF patients yielded a negative effect on HF‐related hospitalizations.

Adding to current literature, showing that non‐invasive RPM is effective in HF patients, we identified several elements that enhance the clinical effect of RPM.

Firstly, RPM interventions incorporating an additional self‐management module demonstrated a superior reduction in HF‐related hospitalizations. These self‐management modules focused on behavioural change and patient empowerment, targeting aspects such as diet, fluid and alcohol intake, physical exercise, smoking cessation and stress management. While the methods of implementing self‐management varies across the RPM interventions, the main objective remained consistent. Koikai and Khan ⁶⁴ investigated the effectiveness of self‐management strategies in HF patients, concluding that the three primary successful strategies include self‐care maintenance, self‐care monitoring and self‐care management. Self‐care maintenance involves adhering to practices to maintain physical and emotional stability. Self‐care monitoring involves recognition of signs of HF deterioration. Self‐care management involves the action taken in response to signs of HF deterioration. Furthermore, Jonkman et al. ⁶⁵ conducted a meta‐analysis, demonstrating a reduction in the risk of HF‐related hospitalizations and mortality by 1–4% for every month a self‐management intervention was undertaken in HF treatment. The clinical evidence supporting self‐management interventions for HF patients has led the European Society of Cardiology (ESC) to issue a class IA recommendation. ⁶⁶ Thus, based on the available evidence, self‐management modules have been found to be a key feature in reducing HF‐related hospitalizations and should therefore be considered as essential in the implementation of new RPM strategies.

Secondly, the addition of an education module to the RPM interventions led to a significant reduction in HF‐related hospitalizations. The method of how the patients were educated about their disease and who educated them differed between the interventions, but the goal was similar for all, educating the patient about their disease, so they may learn to better understand it. A recent study, performed by Stahlman et al., ⁶⁷ examined whether implementation of a HF education module targeted at patients and their caregivers decreased HF deterioration, ED visits and HF‐related hospitalizations. They concluded that a face‐to‐face educational module for HF patients improved patient outcomes, confidence, and ability to self‐manage HF, and led to a decrease in HF‐related hospitalizations and ED visits. Adding to this evidence, the current meta‐analysis shows that such an educational module is also effective as part of RPM.

Thirdly, the use of video calls as a means of communication during the RPM interventions significantly reduced the number of HF‐related and all‐cause hospitalizations. However, the number of studies that incorporated video calls for telemonitoring and reported on HF‐related rehospitalizations were limited (n = 3), therefore these results should be interpreted with caution. We hypothesize that the use of video calls allows the healthcare provider to get a better insight in the global health status of the patient. In this way the healthcare provider can detect early signs of deterioration, that could potentially be missed otherwise, and intervene. This is in line with findings by Knoll et al., ⁶⁸ who recently demonstrated that the combination of telemonitoring and telecoaching for ambulatory HF patients recently hospitalized and at high risk for rehospitalization was related to a significant decrease in all‐cause mortality and a significant reduction in HF‐related hospitalizations. Therefore, we recommend the use of video calls between the patient participating in an RPM intervention and a healthcare provider, to reduce HF‐related and all‐cause hospitalizations.

Lastly, our meta‐analysis revealed two components in RPM interventions that negatively affect outcomes. Surprisingly, both the option of extra contact with caregivers, and additional questionnaires was associated with an increased risk of HF‐related hospitalizations. The extra contacts during RPM interventions were provided in the following manner: coaching, adherence control, help with depression, discussing other illnesses and/or contact with dieticians. The additional questionnaires, used to monitor HF patients, encompassed the following topics: general well‐being, depression, medication and diet compliance, and a quiz about HF. As there is earlier evidence that engaging patients with their health improve their outcomes, we expected that including these questionnaires and additional contact would show similar or even better results than just monitoring. ⁶⁹ The diverse nature of both of these components makes it difficult to draw a singular conclusion of their effects. A possible explanation could be that the populations in these studies providing specific support for depression might have higher risks for rehospitalization due to anxiety and depression‐related imbalance of the autonomic nervous system and impaired treatment adherence, both having a strong effect on outcomes. ⁷⁰ Unfortunately, due to the fact that anxiety and depression levels were not reported in a majority of the studies, it was difficult to test for this hypothesis.

A previous meta‐analysis by Ding et al. ²⁰ revealed that RPM interventions that use mHealth technologies, like mobile software applications, are associated with a decrease in all‐cause hospitalizations and mortality. A similar effect on all‐cause hospitalizations was found for medication support. This meta‐analysis investigated the technology applications utilized, care objectives parameters, care providers accessible in RPM implementations, and their effects on all‐cause hospitalization and mortality. Additionally, enhancing their findings by incorporating recent papers, considering HF‐related hospitalizations as an outcome, and placing additional emphasis on the communication modality patients employ to connect with care providers, our meta‐analysis really focused on the elements that make RPM interventions effective revealing those elements to create a blueprint for future RPM interventions.

Future perspectives

This meta‐analysis confirms that RPM in HF patients is effective in reducing HF‐related hospitalizations and mortality (Graphical Abstract). Additionally, elements were found that increase the probability of making an RPM intervention successful, such as the addition of a self‐management module, an educational module, and the use of video calls as a means of additional contact with the patient. This study also shows clear, potential areas of improvement for current RPM interventions.

Firstly, almost all interventions make use of spot measurements with traditional sensors such as weighing scales and BP cuffs. The last couple of years, technological development of non‐invasive sensors that can continuously monitor cardiovascular parameters took a leap forward. Continuous monitoring of cardiovascular parameters can give a better insight into HF patients cardiovascular status. ⁷¹ Nonetheless, these sensors have not yet found their way into RPM interventions. This is likely due to the challenges associated with integrating these technologies into clinical practice. Implementation requires a robust infrastructure, technical expertise, regulatory approval, and comprehensive education for both clinicians and patients. Furthermore, for these sensors to be adopted into patient care, strict performance and regulatory standards must be met, to ensure that data are transparent, reliable, and easy for healthcare professionals and patients to understand. ⁷² Future advancements in wearable sensors, such as patches or smartwatches, are expected to make RPM more efficient and patient friendly. These devices could continuously monitor key cardiovascular metrics, facilitating early detection of abnormalities, which could improve patient outcomes, satisfaction, and treatment options by making monitoring less intrusive and better integrated in daily life. ⁷³ Seamless integration of sensors, electronic health records, and the necessary legal safeguards for handling related errors is critical for widespread adoption. To support the ongoing evolution of digital health, regulatory frameworks must evolve to keep pace with rapid technological developments, ensuring transparent and responsible data management practices. ⁷²

Secondly, current RPM interventions have not yet incorporated machine learning and/or artificial intelligence (AI) for automated data analysis. As clinical datasets and continuous data streams coming from patients keep increasing in both size and complexity, AI can assist in making sense out of the large data by providing clinical decision support. By calculating risk of decompensation, clinicians can be supported in deciding how to follow‐up with patients. ⁷⁴ Even though a decrease in sensitivity was noted, Larburu et al. ⁷⁵ used AI (Naïve Bayes with Bernoulli method) on existing RPM data for prediction of deterioration to lower the amount of false alerts and improve the overall prediction of the threshold‐based model. Another study by Kerexeta et al. ⁷⁶ tested different machine learning algorithms for cardiac decompensation with one of their algorithms using gradient boosting achieving an area under the curve of 0.72, showing potential in the implementation of decision support algorithms.

The above suggests that there is still room for improvement in RPM of HF patients. Future research should prioritize the modernization of RPM interventions by incorporating additional effective modules, such as self‐management, education and video calls, alongside innovative non‐invasive sensors capable of continuous monitoring of cardiovascular parameters.

Limitations

A limitation of this study is the considerable heterogeneity among the RPM interventions. This posed a significant challenge in effectively differentiating and analysing distinctions between them. Despite our efforts to standardize modalities, parameters and additional modules, differences across the interventions persisted, rendering it challenging to isolate specific elements and assess their individual impact on the outcome measures. Moreover, when studies were divided into more components for analysis, the number of studies within each category decreased significantly, rendering it more difficult, if not impossible, to draw conclusions regarding the effects of combinations of components.

In this meta‐analysis we included papers that compared a RPM implementation with UC. This was done to be able to compare the effect that RPM has on patient outcomes, regardless of variations in standard care practices across both countries and publication year. It should however be noted that most papers did not go into sufficient detail on how RPM data lead to specific clinical decisions, what exact actions were taken, and the timing of these actions. We recommend that future papers on RPM report on the clinical decisions due to RPM data made to enhance understanding of the impact on patient care.

Furthermore, the age of the population of the included trials (mean age intervention group: 67.84 years vs. control group: 68.11 years) was low in comparison to the average age of HF patients hospitalized in the Netherlands, which was 77 in 2022. ⁷¹ Previous work into the effectiveness of RPM also included this similar, young population. ⁷ , ¹⁷ When considering the results of the studies analysed for this meta‐analysis, it should be taken into account that this younger population may have better digital skills which may skew the results in favour of RPM. A further population bias may have come to exist if less digitally proficient participants did not participate in those studies, because those participants thought they would not be proficient enough to engage with the digital tools. Based on results from earlier studies on the subject of digital literacy and outcomes for older and more frail populations, ⁷⁷ , ⁷⁸ , ⁷⁹ it is important to take effort to make future implementations of RPM also accessible for the less digitally skilled, with the risk of limiting RPM applicability if this is not done.

Conclusion

Our meta‐analysis confirms prior findings indicating the efficacy of RPM in reducing HF‐related hospitalizations and mortality. Several components that improve clinical effects of RPM have been identified, notably the addition of a self‐management and educational module and the use of video calls as extra means of communication. Nevertheless, significant discrepancy persists among various RPM interventions, posing challenges in analysing the overall effect.

Furthermore, looking at the rapid technological development in recent years, it is clear that RPM is lagging behind on current possibilities. RPM could for example be modernized through integration of novel non‐invasive sensors for continuous monitoring, the use of AI or machine learning for automated data processing and integration of cardiac telerehabilitation.

Conflict of interest: none declared.

Supporting information

Appendix S1. Supporting Information.