This review has been reported in accordance with the PRISMA (Preferred Reporting Items for Systematic Reviews and Meta-Analyses) 2020 guidelines [27], and the protocol was developed in line with the PRISMA-P (Preferred Reporting Items for Systematic Reviews and Meta-Analyses Protocols) 2015 guidelines [28]. The protocol was prospectively registered in PROSPERO (CRD420251179324) and is publicly available in JMIR Research Protocols [29].

Eligibility criteria were predefined using the population, intervention, comparator, outcome, and study design framework and are summarized in Table 1. Studies were eligible if they reported full economic evaluations of algorithm-enabled remote monitoring linked to CIEDs in adult HF or at-risk populations.

An eligible intervention used CIED remote monitoring with continuous connectivity or intermittent data transmissions. It generated an alert for clinical review when the data indicated possible worsening HF or an event. A commercial algorithm name was not required.

A systematic search was conducted between November and December 2025 in MEDLINE (Ovid), Embase (Ovid), Scopus, Web of Science, and the Cochrane Library. Supplementary searches were undertaken in the National Health Service Economic Evaluation Database, the National Institute for Health and Care Excellence, the Canadian Agency for Drugs and Technologies in Health, the international HTA database, and the National Institute for Health and Care Research Journals Library. The search strategy was developed in consultation with a medical librarian and combined controlled vocabulary and free-text terms related to CIEDs, remote monitoring, HF algorithms, and economic evaluation. The full search strategy is provided in Multimedia Appendix 1 [15,17-21,26,30-32].

All records were exported to EndNote (version 21; Clarivate Analytics) for duplicate removal and then imported into Rayyan (Rayyan Systems Inc) [33] for screening. Study selection was undertaken in 2 stages: title and abstract screening followed by full-text screening. One reviewer screened all records against the prespecified eligibility criteria. A second reviewer independently checked the first 476 titles and abstracts and the first 27 full-text reports made available for verification in Rayyan. Reviewers’ decisions remained hidden until both had completed their assessments. Any disagreements were resolved through discussion. A third reviewer was available but was not required. For full-text exclusions, one primary reason was recorded. Partial verification was used because of time and resource constraints. The study selection process was summarized using a PRISMA 2020 flow diagram.

Data were extracted using a structured Microsoft Excel–based form, tested on an initial sample of included studies, and refined iteratively through discussion with the review team before full data extraction. A second reviewer checked the extracted data against the source documents for 3 reports. The review team selected these reports without a formal sampling procedure. The 2 reviewers resolved any differences through discussion. A third reviewer was available but was not needed. Partial verification was used because of time and resource constraints.

The methods and reporting of the included economic evaluations were appraised using the CHEQUE (Criteria for Health Economic Quality Evaluation) tool [34]. Reporting completeness was also assessed at the item level using the CHEERS (Consolidated Health Economic Evaluation Reporting Standards) 2022 statement [35].

This approach differed from the registered protocol, which proposed several appraisal tools based on study design. A single tool was used to support consistent descriptive appraisal across the different types of economic evaluation included in the review.

The first author (HD) completed both assessments. CHEQUE items were rated as “yes,” “somewhat,” “no,” or “not applicable.” “Yes,” “somewhat,” and “no” received scores of 1, 0.5, and 0, respectively. Methods and reporting percentages were calculated separately by dividing the sum of item scores by the number of applicable items and multiplying by 100. Not applicable items were excluded, and applicable items were weighted equally. CHEERS was used to assess reporting completeness. It was not used to assess methodological quality or risk of bias.

Data were synthesized narratively because of substantial heterogeneity in study design, intervention type, algorithm specification, analytical perspective, model structure, time horizon, and reported outcomes. Structured summary tables and descriptive comparisons were used to synthesize study characteristics, intervention and alert workflow features, economic evaluation characteristics, model structure, clinical outcomes, economic outcomes, and main cost-effectiveness findings. No quantitative meta-analysis was undertaken due to substantial heterogeneity in design, setting, outcome measurement, perspective, and modeling assumptions.

Figure 1 summarizes the study selection process. The database searches identified 3108 records. In the full-text assessment, 132 full texts were screened for eligibility. Of these, 122 (92.4%) were excluded. The review included 10 (7.6%) reports: 8 (80%) primary study reports and 2 (20%) HTA reports. The broad search retrieved records on remote care and economic outcomes beyond the scope of this review. Some broad search terms returned large numbers of records before the search sets were combined (Multimedia Appendix 1 [15,17-21,26,30-32]).

Table 2 shows the 10 studies. Most were set in Europe or North America. Study designs included model-based economic evaluations, randomized controlled trials, prospective nonrandomized clinical studies, population-based cohort studies, real-world multicenter comparative studies, systematic reviews, meta-analyses, and HTAs.

Studies reported sample characteristics in varying ways (Table 2). Sample sizes ranged from 200 participants to 15,254 [17,18], while another study synthesized evidence from 17 randomized controlled trials comprising 10,229 participants [21]. Where reported, study populations generally consisted of older adults, and several studies included predominantly male participants [15,19,20,30,31]. Some studies focused specifically on patients with established HF [17,32], whereas others included broader CIED populations with either mixed HF status or HF risk [19,26,30]. Baseline clinical characteristics and comorbidities were not consistently described across studies. HF prevalence was explicitly reported in 2 studies [15,18]. In contrast, other studies primarily described HF severity indicators such as New York Heart Association (NYHA) class or left ventricular ejection fraction [20,21,31,32] or eligibility based on monitoring pathway and implant status [19,26] rather than detailed comorbidity profiles. Funding sources were also mixed (Table 2).

The following are the main clinical evidence sources identified from the included evaluations. Zanaboni et al [17] evaluated the Evolution of Management Strategies of Heart Failure Patients With Implantable Defibrillators (EVOLVO) trial [36]. Sequeira et al [21] used EVOLVO utility values [36] and evidence from Effectiveness and Cost of ICD Follow-up Schedule With Telecardiology (ECOST) [37]. Health Quality Ontario [31] also used EVOLVO data [36]. EVOLVO was therefore a shared clinical evidence source across these 3 economic evaluations.

The included studies showed substantial heterogeneity in remote monitoring system specifications, algorithm types, monitored parameters, and clinical response pathways (Table 3).

Table 4 summarizes the economic evaluation characteristics of the included studies, including economic analysis type, analytical perspective, time horizon, discounting, currency, price year, and sensitivity or uncertainty analysis.

Table 5 summarizes the structural characteristics of the model-based evaluations, including modeling approach, model type, cycle length, and health state specification.

This section summarizes the clinical and economic outcomes reported across the included studies (Table 6).

The included studies were evaluated via the CHEQUE assessment, as shown in Figures 2 and 3. As summarized in Table 8, methodology scores ranged from 56.3% to 87.5%, and reporting scores ranged from 70.8% to 83.3%. Overall, reporting scores were generally slightly higher than methodology scores. Full item-level CHEQUE ratings are provided in Table S1 in Multimedia Appendix 1 [15,17-21,26,30-32].

The CHEERS-based assessment showed that core elements were generally reported consistently, whereas reporting was more variable for health economic analysis plans, heterogeneity or distributional effects, and patient or stakeholder engagement. Full CHEERS ratings are provided in Table S2 in Multimedia Appendix 1 [15,17-21,26,30-32].

This review identified 10 economic evaluations of algorithm-enabled CIED remote monitoring in populations with established HF or considered at risk of HF. The included evaluations differed in population definitions, intervention designs, model structures, time horizons, and reported outcomes, limiting comparability and generalizability. HF status was also defined inconsistently: some evaluations included clearly characterized populations using indicators such as NYHA class or left ventricular ejection fraction, whereas others used less clearly operationalized definitions of populations at risk of HF. Economic findings should therefore be interpreted as context specific rather than directly transferable across populations, algorithms, model structures, and health care settings.

The interventions were conceptually and operationally heterogeneous and should not be treated as a single intervention when assessing cost-effectiveness. They included predefined alert rules and algorithms that combined several measures to classify HF risk [17,19,20,26,30,31]. In some reports, the alert function was described, but the underlying rules were not reported in detail. Monitored parameters ranged from device-related measures to physiological and broader clinical indicators [17,19,26,30-32], while clinical response pathways varied from structured clinician-led review and treatment adjustment to limited reporting of operational processes [17,19,20,30,31]. Economic evaluations reported inconsistently on key implementation and performance features, including alert burden, false-positive alerts, and how alerts were reviewed and acted upon. This limits interpretation of whether favorable economic findings were driven by the algorithm itself, the surrounding clinical pathway, or both.

Economic methods and study contexts varied across the included studies, including patient populations, intervention definitions, evaluation type, analytical perspective, model structure, time horizon, discounting, uncertainty analysis, and clinical input sources [15,17-21,26,30-32]. Favorable findings appeared in both trial-based and model-based evaluations, although results were not consistently cost saving or dominant. In several evaluations, favorable estimates were linked to reported or assumed reductions in hospitalizations, length of stay, or other types of health care use [15,17,18,20,26,30]. QALY gains were often modest; Ahmed et al [30], for example, reported an incremental gain of 0.01 QALYs per patient over the modeled lifetime horizon, while Chew et al [15] reported similar QALYs between groups.

Some evaluations used evidence from the same clinical trial, but they did not use identical economic analyses or reach identical conclusions. Zanaboni et al [17] evaluated the EVOLVO trial. Sequeira et al [21] used EVOLVO utility values alongside ECOST evidence, while Health Quality Ontario [31] used EVOLVO data on unscheduled health care use. The overlap concerns specific clinical inputs. These inputs should not be interpreted as fully independent evidence, but the 3 evaluations were counted and interpreted separately.

The economic results were sensitive to assumptions about baseline hospitalization risk, length of stay, unit costs, clinical response, and the duration of any benefit. This was particularly relevant in long-term models that extended beyond the period of observed clinical data. Comparative evidence on hospitalization and other clinical outcomes was limited for some technologies [26]. For de Graaf et al [32], conclusions also depended on the cost of monitoring and the country setting. These factors restrict the transfer of an estimate from one health care system to another.

Markov-type approaches predominated, ranging from simple two-state alive-or-dead models to more detailed multistate and multiphase frameworks. One study used a patient-level DICE simulation, which tracked mortality and rehospitalization events in continuous time rather than through fixed model cycles [18]. More detailed structures can represent differences in events and their timing, but they require additional data and assumptions. Model structure should therefore be judged against the decision problem and available evidence. Differences in structure also limit direct comparison of costs and QALYs across evaluations.

The included monitoring systems also differed in their function and value mechanism. Where predefined alerts were described, remote monitoring could support earlier review and reduce scheduled in-person visits or other service use [17,19,20,31]. For multiparameter risk systems, economic estimates were particularly sensitive to the assumed or observed effects on hospitalizations and length of stay [26,30]. This suggests that economic value should not be attributed to algorithm-enabled remote monitoring as a single intervention. It depends on system function, alert management, clinical response, evidence strength, implementation cost, and patient population.

Interpretation is further complicated by variation in the populations studied. Included studies covered patients with established HF and those at risk of HF across device groups, including ICDs, CRT-Ds, CRT-Ps, and pacemakers. Because clinical trajectories, monitoring needs, baseline event rates, and downstream costs may differ across these groups, findings should not be assumed to apply equally across all CIED populations. Health Quality Ontario [31], for example, estimated higher costs and higher QALYs for ICD or CRT-D recipients but lower costs and higher QALYs for pacemaker recipients. A result for one device group should not be assumed to apply to all people with HF.

The appraisal findings further limit the strength of a general economic conclusion. Burri et al [19] and de Graaf et al [32] had lower unweighted CHEQUE method percentages. The model by Burri et al [19] predicted cost neutrality under assumptions that did not include a reduction in HF events or mortality. The conclusions by de Graaf et al [32] depended in part on monitoring costs and country context. Kenny et al [26] had a higher method percentage, but comparative outcome evidence for the algorithms remained limited. The favorable cost-effectiveness estimates reported by Kenny et al [26] for HeartLogic and TriageHF Plus should be interpreted in that context. For CorVue and HeartInsight, sufficient comparative outcome evidence was unavailable to produce evidence-based cost-effectiveness estimates. A scenario assuming no clinical benefit produced higher costs without QALY gains [26]. Evaluations using clinical evidence from nonrandomized studies may also be affected by residual confounding. Several evaluations reported manufacturer funding. This warrants careful examination of assumptions and reporting but does not itself establish bias.

Decision-makers should interpret cost-effectiveness estimates in relation to the monitoring system and the local care pathway. Several evaluations reported favorable findings, while others were cost neutral, mixed, or dependent on costs and assumptions [19,26,31,32]. Local staffing, alert workload, hospitalization costs, and existing monitoring infrastructure may affect whether routine care can achieve a reported benefit. Estimates from one device group or health care setting should therefore not be applied automatically to another. This is consistent with methodological work showing that the value of digital health interventions depends on how they are implemented and integrated into clinical pathways [10,11].

The review also indicates that costs beyond the health care payer perspective may be relevant to decision-making. Some studies reported reductions in patient and caregiver costs, including travel, time, and attendance burden [17,20,32]. A broader analytical perspective that includes patient, caregiver, and societal costs may therefore better capture the value of remote monitoring, particularly where it reduces the need for in-person follow-up [10-12].

Future evaluations should describe the monitoring intervention and its clinical pathway more clearly. Reports should specify the data used, the conditions that generate alerts, the output shown to clinicians, and the actions taken after review. A commercial algorithm name is less informative than a clear description of what the system does. Consistent reporting of alert frequency, false positives, and staff workload would also help explain economic results across settings.

Greater attention is also needed to patient heterogeneity. Most studies reported average cost-effectiveness results and rarely examined whether costs, outcomes, or economic value differed across clinically relevant subgroups. This gap matters because the value of remote or algorithm-enabled monitoring may vary by device type, HF severity, prior hospitalization, baseline risk, and comorbidity burden. Future evaluations should therefore incorporate subgroup analyses where feasible and examine how structural uncertainty and long-term extrapolation affect conclusions across patient groups.

Models should provide clear reasons for their structural choices and test influential assumptions. Validation and uncertainty analysis should be reported transparently. Real-world data may improve relevance to routine care provided that selection bias and confounding are addressed.

This review focused on economic evaluations of algorithm-enabled CIED monitoring in HF-relevant populations. Multiple bibliographic databases and HTA sources were searched. The protocol was registered prospectively. This review describes intervention functions, monitored measures, clinical pathways, economic methods, and findings. It used the CHEQUE checklist for methodological appraisal and the CHEERS 2022 statement to assess completeness of reporting.

Several limitations should also be acknowledged. One reviewer conducted screening and data extraction, with verification rather than full independent duplicate review, which may have increased the risk of missed studies or extraction errors. The CHEQUE and CHEERS assessments were completed by one reviewer. CHEQUE percentages were calculated with equal weights and should be interpreted descriptively. The review was restricted to English-language publications and may therefore have excluded relevant evidence published in other languages.

The evaluations also differed substantially in interventions, populations, settings, perspectives, time horizons, and outcomes. These differences prevented a meaningful pooled economic estimate. Some evaluations used inputs from the same clinical trials, so parts of the underlying evidence were not independent. Incomplete intervention reporting and limited comparative clinical evidence constrained interpretation. No formal certainty of evidence assessment was conducted. These limitations should be considered when judging the overall strength and transferability of the findings.

Economic evidence on algorithm-enabled CIED remote monitoring in HF-relevant populations appears promising. Most studies reported cost-effective, cost-saving, or dominant findings, with reductions in hospitalization and health care use acting as important drivers in several evaluations. However, the evidence remains heterogeneous and difficult to compare directly across studies. Reported findings were sensitive to differences in intervention design, model structure, analytical assumptions, and clinical input data, which limits certainty and transferability. More robust and standardized evidence is needed before economic conclusions can be applied confidently across different devices, algorithms, and health care settings.