Digital biomarkers in cardiovascular medicine: a scientific statement of the ESC Working Group on e-Cardiology, the ESC Working Group on Cardiovascular Pharmacotherapy, the European Heart Rhythm Association (EHRA) of the ESC, the Association for Acute Cardiovascular Care (ACVC) of the ESC, the Heart Failure Association (HFA) of the ESC, the European Association of Preventive Cardiology (EAPC) of the ESC, the European Association of Cardiovascular Imaging (EACVI) of the ESC, in collaboration with the Digital Cardiology and Artificial Intelligence Committee of the ESC, and the ESC Regulatory Affairs Committee

Abstract

Cardiovascular disease is the leading cause of morbidity and mortality globally, imposing a substantial patient burden. Digital biomarkers derived from digital health technologies enable quantifiable, near-continuous, real-world capture of pathophysiological signals and are increasingly used across cardiovascular medicine. Their growth is often supported by big data analytics, artificial intelligence and connected sensors within the Internet of Things. This scientific statement systematically reviews digital biomarkers, their clinical utility across the cardiovascular care continuum, and summarizes major methodological and implementation challenges. We searched PubMed/MEDLINE, Embase and Scopus for phase III–IV cardiovascular trials published between 1 January 2019 and 1 August 2024 that collected digital biomarkers. We identified 541 records and included 32 trials (40 reports) enrolling 32 246 participants (44.71% women; mean age 66.2 ± 11.0 years). Heart failure was the most frequent target condition (16 trials, 50%), followed by cardiac implantable electronic devices trials (12 trials, 38%). Digital biomarkers included cardiac rhythm and heart rate, blood pressure, body weight, impedance-based measures, haemodynamic pressures, and physical activity and sleep patterns captured using diverse sensor technologies. Methodological quality was good (modal quality score 5; range 3–7), although most trials were unblinded. Current studies suggest potential clinical utility in selected cardiovascular conditions. Implementation remains challenging because of measurement variability, privacy and regulatory requirements, affordability and reimbursement barriers and inequities in access. Robust validation and outcome- and cost-effectiveness evidence are still needed. Registration number: CRD42024573879.

Graphical Abstract

Graphical Abstract.

Visual summary of digital biomarkers in cardiovascular medicine and key barriers to implementation.

Introduction

Cardiovascular disease (CVD) remains the leading cause of morbidity and mortality worldwide, burdening patients with multiple debilitating symptoms. CVD diminishes workforce productivity and imposes substantial economic burdens due to high treatment costs and hospitalizations. Digital health technologies (DHTs) have the potential to solve the challenges within global healthcare systems. They may enhance early detection, optimize treatment, reduce disparities in access and lower overall healthcare costs, while improving quality of life and outcomes.^1,2

Biomarkers are frequently used in clinical practice, both in hospital and after discharge, as well as in medical research. Both the European Medicines Agency (EMA) and the US Food and Drug Administration–National Institutes of Health (FDA–NIH) Biomarker Working Group define a biomarker as an objective, quantifiable characteristic that reflects normal biological processes, pathogenic processes or responses to exposures or interventions, excluding measures of how an individual feels or functions.^3,4 Biomarkers can be classified based on their function into overlapping categories: Risk/susceptibility, diagnostic, monitoring, prognostic, predictive, pharmacodynamic/response, and safety biomarkers.^4,5 According to the EMA, the clinical relevance of biomarkers requires a reliable association with validated clinical endpoints.⁶

Digital biomarkers (DBs) are biomarkers derived from data captured by DHTs, processed, and stored using digital tools.⁷ However, definitions of DBs remain heterogeneous in the biomedical literature. Despite this variability, DBs typically share common characteristics. They are often acquired non-invasively, collected longitudinally in near real time, and are characterized by high sampling rates, high data volume and algorithm-based processing.⁸ Moreover, multiple signals can be captured concurrently and integrated into a composite DB.⁹ Nevertheless, there can be overlap between traditional biomarkers and DBs, particularly when measurements are digitally captured and processed (Figure 1).

Figure 1.

Traditional vs. digital biomarkers. Traditional biomarkers are typically obtained in clinical settings and provide episodic measurements (e.g. molecular, histological and imaging findings). Digital biomarkers are derived from digital health technologies and algorithms and are often captured longitudinally in near real time, generating high-frequency, high-volume data (e.g. physiological and haemodynamic signals, imaging-derived metrics and behavioural measures). Imaging may be considered ‘digital’ when derived through automated/AI-based quantification or algorithmic post-processing. Abbreviations: DNA, deoxyribonucleic acid; RNA, ribonucleic acid; CT, computed tomography; MRI, magnetic resonance imaging; PET, positron emission tomography; LVEF, left ventricular ejection fraction; GLS, global longitudinal strain; FFR-CT, fractional flow reserve computed tomography; ECG, electrocardiography; HRV, heart rate variability.

Mobile health (mHealth, smartphone, and wearable health applications) and the Internet of Things (connected sensors and devices) enable early, extensive remote capture and transmission of signals to clinical dashboards for evaluation and management. Big data analytics and artificial intelligence (AI) can derive DBs from these signals to support disease phenotyping, risk prediction and personalized medicine. The widespread use of consumer DHTs facilitates the collection of DBs in real-world, naturalistic environments and may promote more equitable access to care. Patient empowerment through DHTs may facilitate a shift towards more proactive care and self-management, in line with the goals of value-based healthcare. In parallel, DBs are increasingly being incorporated into clinical trials (CTs) with decentralized trial designs and continuous endpoint capture, improving trial efficiency and accelerating the evaluation of novel therapies.^10,11

Although the term ‘digital biomarker’ is not yet widely used, measurements obtained using DHTs are already applied in cardiology across prevention, prediction, diagnosis, treatment, and prognosis. For example, DBs range from blood pressure trends to metrics derived from remote monitoring of cardiac implantable electronic devices (CIEDs) (see Central illustration).

Central illustration.

Digital biomarkers (DBs) in cardiovascular medicine include measurements such as electrocardiography, cardiac rhythm and heart rate, blood pressure, body weight and impedance-based measures, cardiac implantable electronic device/implantable cardiac monitor telemetry, haemodynamics including pulmonary artery pressure and left atrial pressure, and imaging-derived DBs from echocardiography (Echo), cardiac magnetic resonance, and coronary computed tomography angiography. DBs are captured using invasive and non-invasive digital health technologies and are used across the clinical continuum from prevention and risk prediction to diagnosis, treatment monitoring, and prognostication. DBs can support disease phenotyping and personalized care, helping enable earlier detection and intervention and thereby contributing to value-based, cost-effective, and more equitable healthcare. Their role is expected to expand further as AI-enabled analytics continue to evolve.

However, integrating DBs into clinical practice and research remains challenging. Despite major technological advances in cardiovascular medicine, there is still substantial variability in measurement accuracy across devices and algorithms. Additionally, the legal and regulatory framework is still evolving, particularly regarding the standardization of validation processes and compliance with data protection and privacy requirements across clinical and consumer settings, which complicates widespread adoption of DBs. Moreover, heterogeneous access to DHTs and the failure to establish clear reimbursement programmes remain substantial barriers, even in well-resourced healthcare systems, and may be further exacerbated by socioeconomic and cultural factors.^12,13

Given the rapid advances in DHTs and the exponential growth of available DBs, this paper aims to summarize the current clinical applications and associated technologies of DBs across cardiology subspecialties. It also outlines key considerations in DB development and validation, including methodological, regulatory, and implementation barriers.

Methods

Search strategy

A structured literature search was performed in PubMed/MEDLINE, Embase and Scopus, covering the period from 1 January 2019 to 1 August 2024 (inclusive). We focused on studies published from 2019 onward because the volume of research on DBs in cardiovascular medicine increased substantially during this period, with peak publication volume observed in 2022. Search terms incorporated controlled terms [Medical Subject Headings (MeSH)/Emtree] and free-text keywords related to DBs and DHTs (e.g. digital biomarker*, digital technolog*, digital tool*, wearable sensor*, remote monitor*) together with cardiovascular terms (e.g. heart, cardiovascular system, heart diseases, cardiology, heart failure, hypertension, coronary artery disease, cardiovascular diseases). Boolean operators were applied, and truncation (*) was used to maximize sensitivity. Trial filters and publication date limits were applied. The full search strings for each database are provided in the Supplementary material online, Appendix  S1. The review protocol was registered in the International Prospective Register of Systematic Reviews (PROSPERO; CRD42024573879). The review was conducted and reported in accordance with the Cochrane Collaboration guidance and the Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) statement.¹⁴

Eligibility criteria

The inclusion and exclusion criteria were as follows.

Inclusion criteria

• Studies were phase III or IV CTs.

• DBs were derived from data captured by DHTs.

• The clinical focus was cardiology/cardiovascular medicine.

• Studies were published between 1 January 2019 and 1 August 2024 (inclusive).

• Full-text articles were available in English.

Exclusion criteria

• Phase I–II CTs; preclinical studies; and studies using non-human samples.

• CT protocols and design/rationale/implementation articles; observational studies; systematic reviews/meta-analyses; letters; editorials; expert opinion pieces; and conference abstracts/presentations.

• Studies primarily focused on stroke/cerebrovascular disease, diabetes mellitus, obstetric disorders, or congenital cardiovascular disorders.

• Studies with insufficient reporting to evaluate the DB, intervention or outcomes (e.g. unclear DB definition, missing key trial details, or incomplete outcome data).

• CTs scoring <5 on the seven-item modified Jadad scale (0–7), unless clinically relevant and non-substitutable.

Study selection

Two authors (P.M.C. and R.C.A.) independently screened titles and abstracts and subsequently assessed full texts to identify studies meeting the eligibility criteria. Eligibility was confirmed at full-text review, including restriction to phase III–IV CTs. Discrepancies were resolved by discussion, with reference to the original articles, and, when necessary, through adjudication by a third reviewer.

Data extraction

Data were extracted independently by two authors (P.M.C. and R.C.A.) using a predefined data-extraction form. Extracted variables included first author, year of publication, CT registration number and study acronym (when available), country of enrolment, participant age (mean ± standard deviation [SD] or median [interquartile range (IQR)]), sample size (screened and randomized), sex, population characteristics, DHTs and corresponding DBs, sensor technology, trial objectives/endpoints, key results, and additional comments.

Data availability

No new data were generated or analysed in support of this article. All data underlying this article are included in the main manuscript, its Supplementary material and the referenced literature.

Risk of bias and quality assessment

Methodological quality of included CTs was assessed using the seven-item modified Jadad scale. This tool was selected because it provides a pragmatic and reproducible appraisal of key trial features (randomization, blinding, and participant flow). We used the modified version because the original adverse effects item was not applicable, as most included CTs were measurement/monitoring studies and did not systematically report safety outcomes. Quality appraisal was performed by one reviewer and verified by a second. Disagreements were resolved by consensus or, when required, by discussion with a third reviewer. CTs scoring ≥5 (0–7) were considered high quality and were included. In line with the eligibility criteria, the exception for trials scoring <5 was applied. Details of the quality assessment and scoring are reported in the Supplementary material online, Appendix  S2.

Results

A total of 541 records were identified from three electronic databases. Duplicates (n = 187), non-English records (n = 2), and one record retracted by the publisher (n = 1) were removed prior to screening. Subsequently, 351 records were screened by title and abstract, and 191 records were excluded at this stage. After title/abstract screening, 160 reports were sought for retrieval, 12 could not be retrieved. Overall, 148 full-text articles were assessed for eligibility. Of these, 108 were excluded for the following reasons: study designs other than phase III or IV CTs (n = 49), outside the scope of cardiology (n = 37), use of DHTs primarily for behavioural change (digital therapeutics) rather than DB capture (n = 9), not related to DBs (n = 9) or not using DHTs (n = 4). Among the studies excluded as outside the scope of cardiology, stroke/cerebrovascular disease (n = 19), diabetes mellitus/diabetic foot (n = 6), pregnancy-related or congenital cardiovascular conditions (n = 6), and other reasons (n = 6) were the main topics related to the cardiovascular system but not focused on cardiology. Finally, 32 CTs (40 reports) met the inclusion criteria and were included in the systematic review. The PRISMA flow diagram is presented in Figure 2.

Figure 2.

PRISMA flow diagram in four levels.

Study and population characteristics are summarized in Table 1 (40 reports corresponding to 32 CTs). Across the 32 CTs, 30 trials were randomized and 2 were single-arm. In total, 32 246 participants were enrolled, including 14 417 women (44.71%). No trial enrolled healthy volunteers. Among studies reporting mean ± SD, the sample size–weighted mean age was 66.2 years (pooled SD, 11.0 years). Across studies and trial arms, median age ranged from 59 to 74 years with the first and third quartiles ranging from 53–67 years and 66–80 years, respectively. Geographically, nine CTs were conducted in the USA, with the remaining trials undertaken across 10 other countries, including six multinational studies. Most participants [29 666 (92%)] were enrolled in studies conducted in the USA, China, the UK, Poland, Japan, or multinational settings (see Supplementary material online, Figure S3).

Table 1.

Characteristics of 40 reports corresponding to 32 phase III–IV clinical trials that evaluated cardiovascular digital biomarkers derived from digital health technologies. Reports are listed in reverse chronological order

First author (year)	Clinical trial registration number (study abbreviation)	Country of enrolment	Age, years [mean ± SD or median (IQR)]	Screened sample	Randomized sample	Women, %	Population characteristics	Digital health technologies (digital biomarkers)	Sensor technology	Objective	Results	Comments
Victoria-Castro (2024)15	NCT04394754	USA	61 [53–69]	182	182	37.9	HFpEF or HFrEF	Bodyport cardiac scale (HR; body weight; bioimpedance)	ECG; ballistocardiogram; impedance plethysmogram signals	Evaluate the 90-day change in quality of life (KCCQ-OSS)	No significant change in KCCQ-OSS, and no improvement vs. usual care	The intervention used Conversa (conversational platform) and Noom (smartphone app); at 90 days, symptoms and physical function improved significantly in the Noom arm
Jøns (2024)16	NCT02341534 (BIO-GUARD-MI)	Multinational (USA, Australia, Austria, Belgium, Czech Republic, Denmark, France, Germany, Hungary, Latvia, Netherlands, Poland, Slovakia, Spain)	71.2 ± 8.1	802	790	28	ACS and a CHA2DS2-VASc score ≥4 (men) or ≥5 (women), LVEF > 35%	BioMonitor 2-AF (ICM) (cardiac rhythm; arrhythmia detection; HR); BIOMONITOR III (ICM) (cardiac rhythm; arrhythmia detection; HR)	ECG	Evaluate whether arrhythmia monitoring with an ICM improves treatment and outcomes	Asymptomatic but actionable arrhythmias were common post-infarction; however, ICM monitoring did not improve outcomes in the overall cohort	Post-hoc analyses suggested potential benefit in non-ST-segment elevation myocardial infarction patients and other high-risk subgroups
Gomez (2024)17	NCT05366803 (WHISH STAR)a	USA	78.8 ± 4.3	4791	1257	100	Postmenopausal women with 5-year predicted risk of incident AF ≥ 5% by CHARGE-AF score	Single-lead ECG patch (cardiac rhythm; arrhythmia detection; HR)	ECG	Determine the prevalence of frequent premature ventricular contractions and non-sustained ventricular tachycardia	Premature ventricular contractions were relatively common (4.3%) and were associated with higher CHARGE-AF scores	Seven-day ECG patch screening was performed at baseline, 6 months and 12 months
Alshahrani (2024)18	NCT05015634 (TELE-ACS)	UK	Intervention 57.6 ± 8.7; Control 58.6 ± 9.7	372	337	13.9	ACS with percutaneous coronary intervention and at least one cardiovascular risk factor	SmartHeart ECG belt (12-lead) (cardiac rhythm; arrhythmia detection; HR)	ECG	Evaluate time to first readmission at 6 months	Readmissions were reduced over 6 months (HR 0.24; 95% CI 0.13–0.44; P < 0.001)	Telemedicine-based post-ACS care reduced readmissions, emergency department visits and unplanned revascularization and improved symptoms
Chin-Sheng Lin (2024)19	NCT05118035	China	61.2 ± 18.3	16 335	16 335	47.3	Emergency and inpatient patients	AI-enabled ECG (cardiac rhythm; arrhythmia detection; HR)	ECG	Identify hospitalized patients at high risk of mortality	The AI-ECG alert was associated with reduced all-cause mortality (HR 0.83; 95% CI 0.70–0.99)	The mechanisms by which the AI-ECG alert reduced mortality remain to be fully elucidated
Stolen (2023)20	NCT04732728 (LUX-Dx PERFORM)	USA	65.8 ± 14.6	369	—h	49.9	ICM implantation	LUX-Dx™ ICM (cardiac rhythm; arrhythmia detection; HR)	ECG	Characterise ICM remote monitoring rates	High data transmission rates, detection of atrial flutter and feasibility of remote programming
Nassif (2023)21	NCT03030222 (EMBRACE-HF)	USA	66.2 ± 12.9	93	65	37	HFpEF or HFrEF and an elevated PA diastolic pressure	CardioMEMS™ (PA pressure)	MEMS	Evaluate change in NT-proBNP from baseline	Short-term reductions in ambulatory PA diastolic pressure were associated with reductions in NT-proBNP
Kronborg (2023)22	(DANPACE II trial)	Denmark	DDR-60 73 [67–79]; DDD-40 74 [67–80]	540	539	DDR-60 48; DDD-40 52	Sinus node dysfunction with a first-time pacemaker implantation	RM platforms (Biotronik; Boston Scientific; Medtronic; Abbott) (cardiac rhythm; device diagnostics; HR)	ECG	Evaluate whether atrial pacing minimization in sinus node dysfunction reduces the incidence of AF	Atrial pacing minimization did not reduce AF incidence; a base rate of 40 bpm without rate adaptation increased syncope/presyncope risk
Ginder (2023)23	NCT00559988 (IMPACT)	Multinational (USA, Australia, Canada, Denmark, Germany, UK)	64.4 ± 11.5	4226	2718	26.3	ICD/CRT-D, CHADS2 risk score ≥1 and ability to tolerate anticoagulation	RM platform (cardiac rhythm; device diagnostics; HR)	ECG	Evaluate the composite endpoint of stroke, systemic embolism and major bleeding	Anticoagulation initiation/interruption based on remotely detected atrial tachyarrhythmias did not prevent thromboembolism or bleeding
Desai (2023)24	NCT03387813 (GUIDE-HF)	Multinational (USA, Canada)	HF hospitalization in Previous Year 67.6 ± 11.6, Elevated natriuretic peptides Only 71.2 ± 9.9	1000	999	HF hospitalization in Previous Year 39.7, Elevated natriuretic peptides Only 34.6	NYHA II-IV HF and either previous HF hospitalization or elevated natriuretic peptides levels	CardioMEMS™ (PA pressure)	MEMS	Evaluate the efficacy and safety of haemodynamic-guided HF management in patients with elevated natriuretic peptides but no recent HF hospitalization	Haemodynamic-guided HF management supports consideration in chronic HF with elevated natriuretic peptides and no recent HF hospitalization
D’Amario (2023)25	NCT03775161 (VECTOR-HF)	Multinational (Germany, Italy)	71.50 [63.75–75.75]	30	—h	23.3	HFpEF or HFrEF with optimal medical and device therapy in NYHA Functional Class III	V-LAP™ system (left atrial pressure)	MEMS	Evaluate device deployment, feasibility of measurements and freedom from major adverse cardiac and neurological events through 3 months post-procedure	Generally safe, with good correlation vs. invasive pulmonary capillary wedge pressure; early evidence suggested possible symptom improvement
Wita (2022)26	NR	Poland	66.1 ± 10.5	NR	60	21.7	HFrEF with CRT-P/CRT-D	BP monitor (BP); 3-lead ECG recorder (cardiac rhythm; HR); weight scale (body weight)	NR	Evaluate whether telemedicine devices reduce the composite endpoint of death or first emergency hospitalization	Telemonitoring in CRT recipients improved 2-year prognosis and reduced HF hospitalization
Spindler (2022)27	NCT03388918	Denmark	Intervention 61.73 ± 10.75; Control 61.36 ± 11.46	353	137	Intervention 24; Control 23	HF with NYHA I-IV, of whom a maximum of 20% of the patients were of NYHA class I, ≥18 years of age	UA-767PBT BP monitor (BP); UC-321PBT weight scale (body weight); Fitbit Zip (physical activity); Fitbit Charge (physical activity; sleep metrics); Beddit 3 (sleep metrics)	Oscillometry (UA-767PBT BP monitor); strain-gauge load cell (UC-321PBT weight scale); accelerometer (Fitbit Zip; Fitbit Charge); PPG (Fitbit Charge); piezoelectric sensor (Beddit 3)	Evaluate changes over time in eHealth Literacy Questionnaire in HF patients in a telerehabilitation programme vs. a traditional rehabilitation programme	A digital toolbox to support health information processing in HF may improve eHealth Literacy Questionnaire scores, motivation and engagement with digital services
Piotrowicz (2022)28	NCT02523560 (TELEREH-HF)b	Poland	Intervention 62.7 ± 10.5; Control 62.3 ± 10.1	2333	850	Intervention 11; Control 11.8	HFrEF with NYHA I-III after a cardiovascular hospitalization within the 6 months before randomization	EHO mini device-ECG (cardiac rhythm; HR); BP monitor (BP); weight scale (body weight)	ECG; NR (BP monitor; weight scale)	Evaluate the influence of HCTR on depressive symptoms and physical capacity in HF	HCTR produced similar reductions in depressive symptoms to usual care and improved physical capacity in patients with and without depression
Treskes (2022)29	NCT02976376 (TheBox)	Netherlands	59 [53–66]	NR	200	32	ACS with more than 90% occlusion on coronary angiography (≤48 h after onset of symptoms)	Withings ecosystem: scale (body weight), BP monitor (BP), Pulse Ox (oxygen saturation); KardiaMobile (single-lead) (cardiac rhythm; arrhythmia detection; HR)	Strain-gauge load cell (Withings scale); oscillometry (Withings BP monitor); PPG + accelerometer (Withings Pulse Ox); ECG (KardiaMobile)	Describe a cost-utility analysis of an eHealth intervention in patients with ACS	Outpatient eHealth follow-up was likely to be cost-effective vs. usual follow-up
Masterson Creber (2022)30	NCT02731326 (iHEART)	USA	62 (26–87)g	NR	105	25	AF and at least one AF-related risk factor with direct cardioversion or radiofrequency ablation	KardiaMobile (single-lead) (cardiac rhythm; arrhythmia detection; HR)	ECG	Categorise AliveCor use as infrequent (≤5 times/week), moderate (>5 to ≤11 times/week) or frequent (>11 times/week)	More frequent KardiaMobile use was associated with AF symptoms and possibly symptomatic cardiac events
Liu (2022)31	NCT04652648	USA	Intervention 49.6 (22–78)g; Control 42.9 (29–84)g	483	51	Intervention 53; Control 74	Hospitalized patients with positive COVID-19 nasal PCR tests	KardiaMobile (6-lead) (cardiac rhythm; arrhythmia detection; HR)	ECG	Assess ECG evidence of cardiotoxicity in the hydroxychloroquine group vs. the observation group	Remote monitoring of 407 ECGs showed no corrected QT interval prolongation or other ECG changes in either group	Daily remote 6-lead ECG monitoring to exclude cardiac toxicity was feasible and operationalized
Lin (2022)32	NCT05366803 (WHISH STAR)a	USA	78.8 ± 4.3	4791	1257	100	Postmenopausal women with 5-year predicted risk of incident AF ≥ 5% by CHARGE-AF score	Single-lead ECG patch (cardiac rhythm; arrhythmia detection; HR)	ECG	Determine the frequency of AF detected by serial 7-day ECG patch screening in older women at elevated AF risk	AF detection increased from 2.5% at baseline to 3.7% at 6 months and 4.9% at 12 months; yield was higher with CHARGE-AF ≥10% (4.2%, 5.9% and 7.2%, respectively)	Most participants with patch-detected AF had no clinical AF diagnosis (36/46; 78%)
De Graaf (2022)33	NCT01691586 (REMOTE-CIED)c	Multinational (Netherlands, Germany, France, Spain, Switzerland)	Netherlands: RM 67 [60–72], in-clinic 66 [59–72]; Germany RM 66 [58–77], in-clinic 67 [59–74]; RM 66 [58–73], in-clinic 64 [59–73]	≈ 900	600	Netherlands: RM 25.5, in-clinic 24.2; Germany RM 19.7, in-clinic 13.4; RM 18, in-clinic 6.5	HFrEF with first-time ICD and NYHA II-III	LATITUDE™ platform (cardiac rhythm; device diagnostics); weight scale (body weight); BP monitor (BP)	ECG (LATITUDE™ platform); NR (weight scale; BP monitor)	Assess the impact of RM in HF patients with an ICD on medical resource use, direct medical costs, quality-adjusted life-years and travel time	RM reduced medical resource use and travel time; whether it is cost-saving or cost-effective depends strongly on RM costs
Chiu (2022)34	NCT01691586 (REMOTE-CIED)c	Multinational (Netherlands, Germany, France, Spain, Switzerland)	65 [59–73]	≈ 900	600	21	HFrEF with first-time ICD and NYHA II-III	LATITUDE™ platform (cardiac rhythm; device diagnostics); weight scale (body weight); BP monitor (BP)	ECG (LATITUDE™ platform); NR (weight scale; BP monitor)	Evaluate the effect of partially substituting in-clinic visits with RM on clinical outcomes in ICD patients	RM was non-inferior to conventional in-clinic visits for clinical outcomes and reduced unnecessary visits without compromising safety or efficiency
Asch (2022)35	NCT02708654 (EMPOWER)	USA	64.5 ± 11.8	8759	566	47.5	HFpEF or HFrEF with a previous hospitalized episode	Weight scale (body weight)	NR	Evaluate time to death or all-cause readmission within 12 months	No reduction in the combined endpoint of readmission or mortality in a 12-month intensive RM programme	Financial incentives were provided for completing the measurements
Varma (2021)36	NCT00531661 (CHAMPION)	USA	Intervention 64 ± 13; Control 63.7 ± 11.6	NR	190	Intervention 13; Control 11	NYHA III, CRT recipients and at least 1 HF hospitalization in the prior 12 months	CardioMEMS™ (PA pressure)	MEMS	Assess the impact of PA pressure-guided medical therapy in CRT recipients	Remote haemodynamic-guided adjustment of medical therapy reduced PA pressure and HF symptoms beyond the effect of CRT
Szalewska (2021)37	NCT02523560 (TELEREH-HF)b	Poland	Ischaemic: Intervention 65.5 ± 8.6; Control 64.1 ± 8.8. Non-ischaemic: Intervention 57.0 ± 12.4; Control 58.5 ± 11.6	2333	850	Ischaemic: Intervention 7.1; Control 8. Non-ischaemic: Intervention 19.4; Control 17.9	HFrEF with NYHA I-III after a cardiovascular hospitalization within the 6 months before randomization	EHO mini device-ECG (cardiac rhythm; HR); BP monitor (BP); weight scale (body weight)	ECG; NR (BP monitor; weight scale)	Assess whether HF aetiology affects HCTR outcomes based on functional and clinical parameters	HCTR did not increase days alive and out of hospital in either ischaemic or non-ischaemic aetiology
Leppert (2021)38	NCT02888028	Germany	Intervention 61.5 ± 14.2; Control 63.0 ± 15.3	321	180	Intervention 80.2; Control 84	New implantation or replacement of an ICD	RM platforms (CareLink; Home Monitoring; Merlin) (cardiac rhythm; device diagnostics; HR)	ECG	Evaluate the effect of RM, in addition to standard of care, on patient-reported outcomes in a mixed ICD cohort	Patient-reported outcomes were not improved by RM in addition to standard-of-care follow-up
Dorsch (2021)39	NCT03149510	USA	Intervention 60.2 ± 9.2; Control 62 ± 9.2	342	83	Intervention 33; Control 37	LVEF of ≤ 40% or >40% (with a left atrial size of >40 mm, brain natriuretic peptide of >200 pg/mL, or N-terminal pro–B-type natriuretic peptide of >800 pg/mL), and recent decompensated HF	Fitbit Aria/Aria 2 (body weight)	Strain-gauge load cell	Assess the effect of a mobile app self-management intervention on quality of life and HF readmissions	The intervention improved MLHFQ at 6 weeks but effects were not sustained at 12 weeks; no effect on self-reported HF self-management	Fitbit Charge 2 (wrist-worn physical activity monitor) was provided but not used for self-monitoring
Chen (2021)40	ACTRN12614000916640 (ITEC-CHF)d	Australia	69.8 ± 12.4	NR	67	27	HFrEF	FORA TN’G W550 (body weight)	Strain-gauge load cell	Evaluate patient feedback following participation in the ITEC-CHF study	The telemonitoring programme was rated highly for usability and was well accepted as part of routine self-management
Brouwers (2021)41	NL5001 (SmartCare-CAD)	Netherlands	60.7 ± 9.5	300	300	11.3	Stable coronary artery disease, acute coronary syndrome and patients who had received coronary revascularization	Mio Alpha (HR); ActiGraph wGT3x-BT (physical activity)	PPG (Mio Alpha); accelerometer (ActiGraph wGT3x-BT)	Assess the cost-effectiveness of cardiac telerehabilitation with relapse prevention	Cardiac telerehabilitation with relapse prevention was likely to be cost-effective vs. centre-based cardiac rehabilitation
Zakeri (2020)42	UKCRN 10383 (REM-HF)	UK	No AF 68.9 ± 10.3, Paroxysmal AF 69 ± 9.8, Persistent/permanent AF 72.1 ± 9.1	NR	1561	No AF 85 Paroxysmal AF 83.7 Persistent/permanent AF 90.7	Stable HF with HFrEF, NYHA II–IV and a CIED implanted at least 6 months	RM platform (cardiac rhythm; device diagnostics; HR)	ECG	Compare RM vs. usual care on clinical outcomes in patients with and without AF	RM increased clinical activity in AF without reducing mortality and was associated with higher cardiovascular hospitalization in persistent/permanent AF
Watanabe (2020)43	NCT01523704	Japan	77 ± 10	1327	1274	50.4	Single- or dual-chamber pacemaker	CardioMessenger™ (cardiac rhythm; device diagnostics; HR)	ECG	Evaluate the safety and resource use of exclusive RM in pacemaker patients over 2 years	Exclusive remote follow-up over 2 years in pacemaker recipients did not increase major cardiovascular events and reduced resource use
Tajstra (2020)44	NCT02409225 (RESULT)	Poland	Intervention 64 ± 13; Control 64 ± 12	918	600	19	Stable HFrEF with NYHA II–IV and de novo implant of ICD/CRT-D	RM platforms (CareLink; Home Monitoring; Merlin; LATITUDE) (cardiac rhythm; device diagnostics; HR)	ECG	Analyse the effect of RM on clinical outcomes in HF patients with ICD/CRT-D in real-world conditions	RM in HF patients with ICD/CRT-D reduced hospitalization rates in the RM arm
Piotrowicz (2020)45	NCT02523560 (TELEREH-HF)b	Poland	Intervention 62.6 ± 10.8; Control 62.2 ± 10.2	2333	850	Intervention 88.7; Control 88.5	HFrEF with NYHA I-III after a cardiovascular hospitalization within the 6 months before randomization	EHO mini device-ECG (cardiac rhythm; HR); BP monitor (BP); weight scale (body weight)	ECG; NR (BP monitor; weight scale)	Assess whether a 9-week HCTR intervention improves clinical outcomes over 12–24 months	HCTR did not increase days alive and out of hospital and did not reduce mortality or hospitalization
Miyoshi (2020)46	UMIN000003351 (MOMOTARO II)	Japan	69.3 ± 10.3	156	57	26.3	ICD/CRT-D, LVEF < 40% or a history of HF hospitalization within the past year	OptiVol™ (thoracic impedance)	Impedance-based sensing	Examine which lifestyle modifications or medications improve HF indicators in asymptomatic HF patients identified via thoracic impedance	Compared with lifestyle modification, 1-week diuretic and nitrate therapy may be more effective management for HF detected by reduced thoracic impedance
Lopez-Villegas (2020)47	NCT02237404 (NORDLAND)e	Norway	75 ± 12	76	50	48	Single or dual-chamber pacemaker	CardioMessenger™ (cardiac rhythm; device diagnostics; HR)	ECG	Perform an economic assessment of whether telemonitoring pacemakers is a cost-effective alternative	Cost–utility was inconclusive owing to wide CIs; incremental cost-effectiveness ratio/incremental net benefit ranged from savings to high cost per quality-adjusted life-year, with most incremental cost-effectiveness ratios above Norwegian healthcare system thresholds
Ding (2020)48	ACTRN12614000916640 (ITEC-CHF)d	Australia	Intervention 69.5 ± 12.3; Control 70.8 ± 12.4	6587	184	Intervention 27; Control 19	HFrEF	FORA TN’G W550 (body weight)	Strain-gauge load cell	Evaluate patient adherence to self-management recommendations in an innovative telemonitoring care programme	ITEC-CHF improved adherence to weight monitoring, although withdrawal rates were high
Versteeg (2019)49	NCT01691586 (REMOTE-CIED)c	Multinational (Netherlands, Germany, France, Spain, Switzerland)	65 [59–73]	≈ 900	600	21	HFrEF with first-time ICD and NYHA II-III	LATITUDE™ platform (cardiac rhythm; device diagnostics); weight scale (body weight); BP monitor (BP)	ECG (LATITUDE™ platform); NR (weight scale; BP monitor)	Evaluate the effect of RM on patient-reported outcomes during the first 2 years after ICD implantation	RM safely replaced most in-clinic ICD follow-ups; patient-reported health status and ICD acceptance did not differ between RM and in-clinic follow-up
Pluymaekers (2019)50	NCT02248753 (RACE 7 ACWAS)	Netherlands	65 ± 11	3706	437	40	< 36 h haemodynamically stable and symptomatic atrial fibrillation	MyDiagnostick (cardiac rhythm; AF detection)	ECG	Determine the presence of sinus rhythm at 4 weeks	A wait-and-see strategy was non-inferior to early cardioversion for return to sinus rhythm at 4 weeks
López-Liria (2019)51	NCT02237404 (NORDLAND)e	Norway	75 ± 12	76	50	48	Single or dual-chamber pacemaker	CardioMessenger™ (cardiac rhythm; device diagnostics; HR)	ECG	Evaluate clinical parameters and quality of life at 12-month follow-up in a random sample of pacemaker users of a telemonitoring system	Telemonitoring was as effective as conventional hospital follow-up, with similar outcomes for clinic/emergency department visits and unplanned re-hospitalizations
Husser (2019)52	NCT00538356 (IN-TIME)f	Multinational (Australia, Austria, Czech Republic, Denmark, Germany, Israel, Latvia)	65.5 ± 9.4	716	664	19	HFrEF with NYHA II-III, no permanent AF and an indication for dual-chamber ICD/CRT-D	Home Monitoring™ (cardiac rhythm; device diagnostics; HR)	ECG	Analyse information flow and workflow in IN-TIME and whether differences in message content, speed, completeness and workflow may contribute to heterogeneous results	Limited evidence from other workflow studies suggests that early review after clinical events may be less effective than in IN-TIME
Geller (2019)53	NCT00538356 (IN-TIME)f	Multinational (Australia, Austria, Czech Republic, Denmark, Germany, Israel, Latvia)	65.5 ± 9.4	716	664	19	HFrEF with NYHA II-III, no permanent AF and an indication for dual-chamber ICD/CRT-D	Home Monitoring™ (cardiac rhythm; device diagnostics; HR)	ECG	Compare IN-TIME outcomes in ICD vs. CRT-D subgroups	Daily telemonitoring may reduce clinical endpoints in HF in both ICD and CRT-D subgroups; absolute benefit appeared greater in higher-risk populations with worse prognosis
García-Fernández (2019)54	(RM-ALONE)	Spain	Intervention 68.9 ± 13.3; Control 68.8 ± 12.8	NR	445	31.1	CIED equipped with Home Monitoring	Home Monitoring™ (cardiac rhythm; device diagnostics; HR)	ECG	Evaluate the safety and efficiency of a simplified CIED RM protocol in pacemaker and ICD recipients with at least 24-month follow-up	Six-monthly RM was safe and efficient, reducing hospital visits and staff workload

a–f Superscripts indicate multiple reports from the same trial: WHISH STAR^a; TELEREH-HF^b; REMOTE-CIED^c; ITEC-CHF^d; NORDLAND^e; IN-TIME^f.

^gReported as median (range), not median [IQR].

^hSingle-arm, non-randomized trial (enrolled n = 30).

Abbreviations: ACS, acute coronary syndrome; AF, atrial fibrillation; BP, blood pressure; bpm, beats per minute; CHA₂DS₂-VASc, congestive heart failure, hypertension, age ≥75 years, diabetes mellitus, prior stroke/transient ischaemic attack, vascular disease, age 65–74 years, sex category (female); CHADS₂, congestive heart failure, hypertension, age ≥75 years, diabetes mellitus, prior stroke/transient ischaemic attack; CI, confidence interval; CIED, cardiac implantable electronic devices; CRT, cardiac resynchronization therapy; CRT-D, cardiac resynchronization therapy with defibrillator; CRT-P, cardiac resynchronization therapy with pacemaker; ECG, electrocardiogram; HCTR, hybrid comprehensive telerehabilitation; HF, heart failure; HFpEF, heart failure with preserved ejection fraction; HFrEF, heart failure with reduced ejection fraction; HR, heart rate; ICD, implantable cardioverter-defibrillator; ICM, implantable cardiac monitor; IQR, interquartile range; KCCQ-OSS, Kansas City Cardiomyopathy Questionnaire Overall Summary Score; LVEF, left ventricular ejection fraction; MEMS, microelectromechanical systems; MLHFQ, Minnesota Living with Heart Failure Questionnaire; NR, not reported; NT-proBNP, N-terminal pro-B-type natriuretic peptide; NYHA, New York Heart Association; PA, pulmonary artery; PPG, photoplethysmography; RM, remote monitoring; SD, standard deviation.

Table 1 summarizes the characteristics of the included studies. Heart failure (HF) was the primary condition studied in 16 CTs (50%). These studies evaluated remote monitoring and telemanagement in both heart failure with reduced ejection fraction (HFrEF) and heart failure with preserved ejection fraction (HFpEF) using a range of devices. Some approaches were non-invasive, capturing data such as body weight, bioimpedance measurements, blood pressure (BP), electrocardiography (ECG) and physical activity/exercise metrics. Others relied on invasive or implantable systems, including measurements obtained from CIEDs, pulmonary artery (PA) pressure waveforms, and left atrial pressure (LAP). Outcomes assessed included quality of life, changes in biomarker levels [e.g. N-terminal pro-B-type natriuretic peptide (NT-proBNP)], the efficacy and safety of haemodynamic-guided HF management with CardioMEMS™ or V-LAP™ monitoring, reductions in major adverse cardiovascular events (MACE)—particularly HF hospitalizations—and digital health literacy related to HF devices.

Additionally, CIEDs were used in 12 CTs (38%). Within CIED-based CTs, pacemakers/cardiac resynchronization therapy pacemakers (CRT-P) were included in 3 (25%) and implantable cardioverter-defibrillators (ICDs)/cardiac resynchronization therapy defibrillators (CRT-D) in 6 (50%); the remaining 3 (25%) involved mixed device cohorts or did not specify device type. Remote monitoring was delivered through manufacturer-specific platforms (e.g. Biotronik Home Monitoring/CardioMessenger, Boston Scientific LATITUDE, Abbott/St. Jude Merlin.net and Medtronic CareLink), sometimes within the same trial depending on the implanted device. Most reports specified device type and/or the remote monitoring platform but rarely reported the exact CIED model.

Excluding CIEDs, 38 device uses were reported across the included CTs, corresponding to 33 distinct device models (manufacturer and model). These DHTs were grouped into eight device categories, with the number of distinct device models in parentheses: weight scales (9), wearable/portable ECG monitors (8), implantable cardiac monitors (ICMs) (4), step counters/physical activity monitors/actigraphs (4), BP monitors (5), CardioMEMS™ (1), LAP monitors (1), and sleep sensors (1).

The DBs measured across the CTs included measurements from ECG [including cardiac rhythm and arrhythmia-related metrics and heart rate (HR)], BP, body weight, bioimpedance/thoracic impedance, PA pressure monitoring, LAP, step counts/physical activity and sleep-related metrics. The sensor technologies used to capture these DBs included ECG, photoplethysmography (PPG), ballistocardiography, impedance-based sensing, oscillometry, microelectromechanical systems (MEMS), strain-gauge load cells, piezoelectric sensors, accelerometers, and altimeters. Among reported sensor technologies, ECG-based sensing was the most common (21/41, 51.2%), whereas among reported DB variables, HR was the most frequently captured (25/50, 50%). Several devices incorporated multiple sensing modalities. For a small number of devices, model-level specifications were not reported.

Methodological quality was good overall (modal modified Jadad score, 5; range, 3–7). Two trials scored 3 on the modified Jadad scale due to their design as single-arm studies and the lack of randomization; however, they were included because they were relevant for evaluating the performance and safety of the V-LAP™ and LUX-Dx™ ICMs. Most of the CTs were unblinded due to the nature of the interventions. Only two studies were double-blind, while three were single-blind. Withdrawals/dropouts, eligibility criteria and statistical analysis were appropriately described in all trials.

Discussion

The use of DBs has increased exponentially in the past decade, accelerated by the COVID-19 pandemic and the need to expand access to broader and effective healthcare. Because DBs can be derived from multiple signals and devices, CTs in cardiovascular medicine are highly heterogeneous with differences in study design and endpoints.

Heart failure

HF is the most common target condition with DBs primarily used for remote monitoring and telemanagement. Overall, evidence is mixed across studies,^{26,33,35,42,44,45,53} but it suggests that selected interventions may reduce HF hospitalizations and improve outcomes,^26,44,53 with cost-effectiveness depending on implementation and programme costs.³³ In the included trials, DBs ranged from non-invasive monitoring (e.g. weight, blood pressure, ECG/heart rate and activity) to implantable systems, including device telemetry and haemodynamic sensors (pulmonary artery and left atrial pressure), as summarized in Table 1. Importantly, programmes differed in monitoring intensity and in how alerts were handled in practice (who reviewed the data, how often reviews occurred and whether alerts resulted in timely clinical decisions), which likely contributed to heterogeneous results. Through longitudinal assessment outside the clinic, DBs can support earlier identification of clinical deterioration and more timely treatment adjustments than traditional episodic visits, helping refine disease phenotyping and deliver more personalized care (see Central illustration).

Cardiovascular imaging

Imaging increasingly uses AI for image acquisition, quantitative measurement (including segmentation), and post-processing. Nevertheless, most evidence to date is based on agreement with conventional measurements or prediction of clinical events, whereas prospective trials remain limited.

In echocardiography, automation spans both software that helps reduce operator dependence and variability and device innovation, with a wearable prototype being developed for continuous left ventricular ejection fraction (LVEF) monitoring.⁵⁵ Convolutional neural networks have enabled autonomous cardiac segmentation and measurement,⁵⁶ with automatic annotation of two-dimensional and Doppler acquisitions.⁵⁷ A CT demonstrated that the quantification of LVEF by AI was non-inferior to assessment by sonographers,⁵⁸ and another trial tested an intelligent Doppler spectrum analysis software designed to estimate LAP using the equation method.⁵⁹

For cardiac magnetic resonance (CMR), trials have focused on benchmarking automated DB measurements against manual method, improving reproducibility and reducing post-processing burden. Among patients after acute MI, an AI-based automated strain analysis showed good concordance with manual analysis, and the automatically derived global longitudinal strain was the only independent predictor of 12-month MACE.⁶⁰ For ventricular arrhythmia substrates, CMR improves risk stratification through a DL contrast-free virtual native enhancement showing high agreement with late gadolinium enhancement (LGE)-CMR for myocardial scar quantification.⁶¹ CMR further extends into ventricular tachycardia management, the VOYAGE trial evaluated AI automation software to integrate LGE-CMR with electroanatomical mapping for ventricular tachycardia ablation.⁶²

In coronary computed tomography angiography (CCTA), DBs include automated coronary artery calcium scoring, quantitative plaque burden and compositional phenotyping and perivascular/adipose tissue signatures used for risk prediction.^63–66 Functional assessment can also be derived via fractional flow reserve-computed tomography (FFR-CT), with ML-derived FFR-CT showing equivalent diagnostic performance to computational fluid dynamics–based FFR-CT.⁶⁷ In this context, CCTA studies have reported prognostic associations, including AI-derived plaque volume and stenosis quantification are prognostic for incident MI,⁶³ and ML risk models improve 5-year mortality prediction beyond established clinical scores.⁶⁸ Similarly, in invasive imaging, deep-learning (DL) algorithms have been applied to coronary angiography to localize and grade stenoses,^69,70 identify plaque erosion,⁷¹ and detect reduced LVEF.⁷²

Cardiac electrophysiology and rhythm monitoring

In cardiac electrophysiology, DBs are mainly focused on rhythm monitoring and detection of cardiac arrhythmias using different technologies for recording. Different trials implemented single-lead ECG⁷³ and PPG^74,75 for atrial fibrillation (AF) screening, and have also evaluated post–acute coronary syndrome (ACS) monitoring with ICMs and portable ECG devices showing that such monitoring may be beneficial in selected high-risk groups [e.g. elevated CHA₂DS₂-VASc or Global Registry of Acute Coronary Events scores]. In these post-ACS trials, enhanced monitoring increased detection of incident AF as well as pauses/high-grade atrioventricular block and ventricular tachyarrhythmias, mainly prompting anticoagulation initiation and, when indicated, antiarrhythmic therapy or device implantation.^16,76,77

Beyond simple ECG and PPG tracings, these DBs can represent a basis for predictive analyses using AI. Most evidence to date relates to retrospective development and validation, including prediction of future AF from single-lead ECG strips recorded in sinus rhythm⁷⁸ and standard 12-lead ECG.⁷⁹ AI models have also been used to detect hyperkalaemia from a 2- or 4-lead ECG⁸⁰ and to identify hypertrophic cardiomyopathy from a 12-lead ECG.⁸¹ ML models have also been developed to predict atrial tachyarrhythmia recurrence after AF catheter ablation.⁸²

Prospective evidence demonstrating clinical impact remains more limited. For sudden cardiac death risk stratification, improvements will likely require integrating multiple DBs across modalities rather than relying on single-signal models.⁸³ In this context, digital twins—computational representations that integrate multimodal patient data—are novel tools to support personalized arrhythmia risk stratification and therapy planning.⁸⁴ Early evidence suggests that ML can be applied to diverse ECG-derived DBs. Examples include models based on heartbeat intervals, which have shown high precision for predicting life-threatening ventricular arrhythmias,⁸⁵ QRS complex shape used to predict ventricular fibrillation.⁸⁶ In patients with ICD/CRT-D devices, daily remote-monitoring data have also been used in machine-learning models to predict impending appropriate device therapies, including ICD shocks.^23,87

Acute and intensive cardiovascular care

DBs captured by DHTs are increasingly incorporated into risk stratification scores and used to continuously monitor patients with high-risk acute cardiovascular conditions.⁸⁸ DBs may improve triage, diagnosis and outcomes in patients with acute CVD through earlier identification of myocardial injury and coronary occlusion before hospital arrival.⁸⁹ For instance, a wrist-worn transdermal infrared sensor predicted elevated cardiac troponins and was associated with regional wall motion abnormalities and significant coronary stenosis,⁹⁰ while three-lead ambulatory occlusion detection systems have been validated⁹¹ and AI models applied to 12-lead ECGs outperformed ST-segment elevation myocardial infarction criteria in identifying acute coronary occlusion MI.⁹²

In hospitalized patients, ML early warning models using routinely measured vital signs can outperform aggregate scores on many outcomes, although performance decreases as the prediction window lengthens.⁹³ Notably, ML-based phenotyping in cardiogenic shock identified a high-risk cardiometabolic subgroup with the greatest likelihood of progression to Society for Cardiovascular Angiography and Interventions shock (SCAI) stages D/E and in-hospital mortality.⁹⁴ Prospective outcome data are limited but an AI ECG alert for reduced LVEF is among the few trials suggesting that a DB can improve survival. The alert prompted earlier clinical review and was associated with lower 90-day mortality.¹⁹

In the most time-critical scenario, out-of-hospital cardiac arrest (OHCA) is a major application with reported improvements in ECG classification, early dispatch and prognostication after return of spontaneous circulation.⁹⁵ AI algorithms have been investigated to strengthen prediction, detection, and management, including resource allocation and cardiopulmonary resuscitation quality and success.^96,97 Using ECG, ML algorithms were effective in predicting cardiac arrest within 24 h,⁹⁸ differentiating pulseless electrical activity-OHCA from ventricular fibrillation-OHCA⁹⁹ and identifying OHCA with acute coronary occlusion.¹⁰⁰ ML has also been embedded into automated external defibrillators,¹⁰¹ optimizing shock/no-shock decisions during rhythm analysis¹⁰² without interrupting chest compressions.¹⁰³ Beyond algorithms, novel DHTs are emerging in this field. In the Jewel IDE study, a water-resistant patch wearable cardioverter-defibrillator was shown to be safe and effective, with a high number of successful defibrillations and few adverse events.¹⁰⁴

Prevention and cardiac rehabilitation

In primary prevention, DBs have enormous potential for early identification of individuals at risk of CVD before symptoms manifest. Derived from data collected through wearables, smartphones, and other DHTs, DBs offer continuous monitoring of parameters such as cardiac rhythm, HR variability, BP, physical activity, and sleep patterns. For instance, smartphone PPG screening identified more than twice as many cases of AF at an early stage compared with usual care.⁷⁴ Activity trackers can provide DB derived feedback and boost physical activity across various age groups and populations.¹⁰⁵ Additional applications include monitoring CVD risk factors like BP via cuffless sensors,^106,107 and assessing psychological stress or sleep quality through HR variability.^108–110 Sports cardiology is currently exploring the potential of wearable devices and AI for monitoring, risk prediction and shared decision-making in athletes with structural or electrical heart diseases. While this field shows promise, there is a lack of evidence supporting its effectiveness and important ethical and legal issues remain to be addressed.¹¹¹

Comprehensive cardiac rehabilitation (CR) is recommended for structured secondary CVD prevention.¹¹² Digitally enhanced home-based programmes (cardiac telerehabilitation) are viable alternatives to centre-based programmes and may enable more personalized delivery, with improved participation and adherence.^113,114 Biosensing wearables and mobile devices enable the remote delivery of core components of CR, including exercise training and activity monitoring, nutritional counselling, and risk factor management, while capturing DBs that largely mirror those used in primary prevention.¹¹³ Systematic reviews suggest modest benefits of mHealth on exercise capacity, physical activity, BP, and body weight, but evidence quality is moderate and interventions are heterogeneous.¹¹⁵ Early evidence indicates that selected DBs, such as daily step count and BP, can provide actionable biofeedback to guide goal directed therapy.^116,117 Nevertheless, major challenges in rehabilitation include usability of DHTs in older and multimorbid CVD populations, defining optimal analysis for DBs, and reaching agreement on meaningful outcomes.¹¹⁸

Development and validation of digital biomarkers: methodological considerations

DHTs that collect DBs rely on hardware, software, and analytic pipelines that must be developed, validated and, when applicable, certified in accordance with regulatory requirements. For systems that incorporate self-measurement, the reliability of data acquisition, transmission, storage, and processing is critical.¹¹⁹ Validating DBs for clinical practice and for use as surrogate endpoints in research requires selecting appropriate digital instruments. Factors to consider when validating novel digital endpoints in CTs include endpoint selection, technical validation, clinical validation, and candidate endpoint evaluation.^120–122 In technical validation, the new assessment is compared with a gold standard under controlled conditions to demonstrate internal validity for the target construct and minimize bias and confounding.^120,123 Clinical validation assesses accuracy, safety and clinical utility in the intended setting. Two issues are particularly relevant at this stage: results may not generalize beyond the study cohort because measurement accuracy can vary across populations (e.g. PPG measurements in elderly individuals), and differences in disease prevalence can affect predictive values. Accordingly, validation should consider populations often underrepresented in CTs (e.g. ethnic minorities, people with advanced chronic conditions, sex/gender differences, and learning difficulties) and be complemented by real-world evaluations.¹²⁴

Even after technical and clinical validation, biomarkers should not be assumed to function as surrogate endpoints.¹²⁵ AI has the potential to assist in the identification of ‘omics’ biomarkers,¹²⁶ and, by extension, may also be applicable to the development of DBs. Wearable technologies generate longitudinal, multi-signal data that ML methods can use to derive candidate DBs.¹²⁷ This is relevant for behavioural interventions, as behaviour reflects complex interactions among biological, environmental, and social determinants that are difficult to quantify¹²⁸ but also for routinely collected hospital data, where ML methods can generate algorithm-based DBs that predict clinically relevant outcomes.¹²⁹

Barriers to the implementation of digital biomarkers in cardiovascular care

Patients with CVD often require frequent monitoring as well as intensive and precise longitudinal management. The use of DHTs—DBs in particular—is highly desirable, but implementation in routine clinical practice remains challenging, as highlighted in the European Society of Cardiology (ESC) Working Group on e-Cardiology position paper on the use of commercially available wearable technology for heart rate and activity tracking in primary and secondary cardiovascular prevention.¹³⁰ This is exemplified by the lower uptake of wearables among individuals with established CVD compared with those at risk for CVD and the general population¹³¹ and by the variability across ESC member countries in the implementation of remote monitoring of CIEDs.¹³² Despite the widespread availability of electronic devices in many regions of the world, the utilization of DBs for patient management is not without difficulty for a variety of reasons:

• Interoperability and electronic medical records (EMRs): EMRs provide the clinical context needed to interpret DB signals, including clinical notes, laboratory tests, ECGs, echocardiography and other imaging modalities. However, a key gap is the limited ability to integrate DB data into EMRs. This separation restricts access to the clinical variables and outcomes needed to develop and validate DBs. At the same time, maintaining this separation may improve security by limiting unnecessary data aggregation.^133,134 Potential solutions include standardized data formats and improved interoperability frameworks to enable secure and permissioned data exchange.

• Regulatory barriers: Validation and post-market performance monitoring remain insufficiently standardized across many DHTs, particularly consumer-grade tools, contributing to within- and between-device measurement variability and hampering the development of novel DBs.^135,136 In parallel, Europe and the US have outlined the criteria for classifying clinical decision support software as software as a medical device, which can increase evidence requirements, raise development costs and slow the deployment of new tools.¹³⁷

• Privacy and security: Individual, personal health-related information on digital platforms is highly sensitive. Beyond data protection, DBs raise governance risks related to data ownership and secondary use, as large technology platforms increasingly mediate the collection, storage, and sharing of personal health data. Due to genuine concerns about these issues, patients may decide not to take part in therapeutic schemes involving digital technologies. Safeguards are in place to protect patient data and to reduce the risk of data leaks, hacking or breaches, but further efforts are necessary to fully reassure the public that personal data are truly anonymized and protected. Laws and regulations exist in Europe and the US that provide a reasonable degree of data protection, but additional regional and local regulatory tools are needed to further reassure patients. In this context, providing patients with clear information about their data rights could enhance confidence in the use of DBs in the near future.

• Training: Healthcare professionals as well as patients and their carers must be educated regarding the application and implementation of DHTs, as these often differ markedly from conventional models.¹¹⁹

• Affordability and equity: Efforts should focus on making consumer electronics more affordable and establishing effective reimbursement policies. A key unresolved issue is who will bear these costs (patients, insurers, healthcare systems, or government agencies), including not only the device itself but also data infrastructure and clinician time required to review and act on DBs.^12,13 The ESC-European Heart Rhythm Association (EHRA) Atlas reports marked cross-country disparities in access to guideline-recommended arrhythmia care and identifies dissatisfaction with national reimbursement systems as a recurrent barrier to implementation.¹³² Moreover, research that is vital for the implementation of DHTs in cardiovascular care has often failed to include ethnic minority populations, culturally diverse individuals and socioeconomically disadvantaged groups, undermining equity in access to DHTs.

• Digital literacy: Even in high-income countries, there is a clear ‘digital divide’ between people who have access to DHTs and possess the digital literacy necessary to use these technologies, and those who do not. Many people have low levels of technology literacy, i.e. older patients and people with disabilities, which may limit their ability to use smartphones and digital therapeutic apps effectively. Interestingly, however, data from the Strategy of Blood Pressure Intervention in the Elderly Hypertensive Patients study showed that most older patients enrolled were able to use the smartphone app to submit blood pressure readings.¹³⁸

• Motivation: Lack of motivation, limited awareness of or trust in medical strategies such as telemedicine, or resistance to adopting new treatment paradigms, are also barriers to the implementation of these useful methodologies for patients as well as physicians.¹¹⁹ Additionally, lack of patient engagement decreases the accuracy of measurements.

• Outcome evidence: DBs have been correlated with multiple clinical markers and secondary endpoints. However, their impact on hard clinical outcomes (e.g. mortality) is uncertain, and evidence for effects on healthcare utilization (e.g. hospitalizations and visits) is also limited. This remains a promising area for future outcome-driven trials.

• Cost-effectiveness evidence: Given rising costs and demand on health systems, cost-effectiveness evidence is increasingly important. Recent studies suggest that DHTs—particularly software designed to support lifestyle change—may enable cost-effective delivery of interventions at scale.¹³⁹ However, health-economic evidence for DBs remains relatively scarce and heterogeneous.^33,140,141

Future perspectives

In the coming years, DHTs are expected to dominate biomarker capture across the cardiovascular spectrum—from the general population to high-risk individuals and patients with established CVD. DBs will further extend beyond traditional care settings, allowing remote and near-continuous monitoring outside the clinic, occasionally supported by AI methods.¹⁴² DBs will also help evaluate treatment response, facilitate guideline implementation, and promote patient engagement.

A major underused opportunity lies in clinical research, particularly the integration of DBs as surrogate endpoints. The Clinical Trials Transformation Initiative has issued recommendations on incorporating DHT-derived endpoints, including DBs, within clinical outcome assessments.¹⁴³ This could allow earlier and more precise ascertainment of outcomes, while also supporting novel endpoints that are not feasible with episodic monitoring. By providing objective measures that reduce bias and random error, DBs complement patient- and clinician-reported outcomes and may improve endpoint specificity, particularly when the signal-to-noise ratio is high.¹⁴⁴

Several emerging DB modalities have been explored as proof-of-concept and require rigorous evaluation in adequately powered, late-phase CTs. For example, facial imaging and ocular photographs have been explored for hypertension screening, while voice features have been investigated as potential DBs of AF and CAD.^145–148 To translate these innovations into routine clinical practice, regulatory frameworks must be supported by robust validation standards and clear requirements for data privacy and governance. Addressing these challenges will be essential for their widespread adoption. Ultimately, digital health and DBs are likely to support a shift towards proactive 4P care (preventive, participatory, predictive and personalized) through clinician–patient collaboration.

Conclusions

Compared with traditional biomarkers, DBs facilitate care along the continuum from prevention and prediction to diagnosis, treatment monitoring and prognostication. By capturing real-world data through DHTs, DBs can reduce reliance on episodic clinic visits and support earlier detection of clinical deterioration and more timely treatment adjustment. Selected DBs have demonstrated clinical utility in specific cardiovascular conditions, although current trials remain heterogeneous in technologies and endpoints. To support broader adoption in routine care, robust validation, clear regulatory requirements covering data privacy and governance, interoperability with clinical systems, and stronger evidence on clinical outcomes and cost-effectiveness will be needed. Finally, continuous innovation in AI technologies is likely to amplify the impact of DHTs in both research and clinical practice, supporting patient self-management and improving day-to-day health.

Supplementary material

Supplementary material is available at European Heart Journal – Digital Health.

Author contributions

Pablo M. Corredoira [Conceptualization, Writing—original draft, Writing—review & editing (equal)], David Duncker [Conceptualization (supporting)], Marcus Dörr [Conceptualization (supporting)], Matthias Wilhelm [Conceptualization (supporting)], Marco Tubaro [Conceptualization (supporting)], Mark Schuuring [Conceptualization (supporting)], Mamas Mamas [Conceptualization (supporting)], Steffen E. Petersen [Conceptualization (supporting)], Rafael Vidal-Pérez [Conceptualization (supporting)], Hareld M. C. Kemps [Conceptualization (supporting)], Emanuela T. Locati [Conceptualization (supporting)], Christian Mueller [Conceptualization (supporting)], Magnus T. Jensen [Conceptualization (supporting)], Raffaele De Lucia [Conceptualization (supporting)], Julia Ramírez [Conceptualization (supporting)], Matthijs Cluitmans [Conceptualization (supporting)], Nurgül Keser [Conceptualization (supporting)], Axel Verstrael [Conceptualization (supporting)], Polychronis Dilaveris [Conceptualization (supporting)], Paul A. Friedman [Conceptualization (supporting)], Rubén Casado-Arroyo [Conceptualization, Methodology, Writing—original draft, Writing—review & editing (lead)], Juan Carlos Kaski [Conceptualization (supporting)], and Loreena Hill [Conceptualization (supporting)].

Funding

This work received no specific funding from any public, commercial, or not-for-profit agency.