2025 Clinical Research on ECG Detection Accuracy of Medical-Grade Wearable Devices

2025 Clinical Research on ECG Detection Accuracy of Medical-Grade Wearable Devices

By Vositone Team 2025-08-20

Abstract

Based on the PRISMA 2020 guidelines, this white paper conducts a systematic review of clinical studies on medical-grade wearable electrocardiogram (ECG) devices published between 2019 and 2025, covering 152 eligible research articles and 23 industry technical reports.

The study focuses on the detection efficacy of wearable ECG devices for cardiovascular diseases such as atrial fibrillation (AF) and aortic stenosis. Evaluation using the QUADAS-2 tool shows that the sensitivity of current ECG smart chest patches and photoplethysmography (PPG) smartwatches ranges from 76.4% to 94.49%, with specificity between 71.0% and 96.4%, indicating significant technical heterogeneity.

The white paper analyzes the clinical limitations of single-lead devices and breakthrough directions of multi-lead technology, proposes the “scenario-based accuracy threshold” theory and a three-level regulatory framework, and provides a clinical translation path for the USD 70 billion wearable health device market.

The research finds that multi-lead flexible electronic technology can reduce motion artifacts by 5 dB, while the generalization ability of AI algorithms in diverse populations still needs improvement. It is recommended to promote the accurate transformation of wearable devices from health monitoring to disease diagnosis through a “technology-clinic-ecosystem” three-dimensional development model.

1. Research Background and Significance

1.1 Global Demand for Cardiovascular Health Management

Cardiovascular diseases (CVDs) remain the leading cause of death worldwide, resulting in approximately 17.9 million deaths annually. Among these, there are 330 million patients with atrial fibrillation (AF) and over 120 million patients with aortic stenosis. Although the traditional 12-lead ECG is the gold standard for diagnosis, limited by uneven distribution of medical resources, only 38% of AF patients globally receive standardized monitoring, and the missed diagnosis rate of aortic stenosis is as high as 62%, mainly due to the insidious nature of its early symptoms. The emergence of wearable ECG devices has broken the temporal and spatial constraints, enabling long-term cardiac monitoring. However, the uncertainty of clinical diagnostic accuracy has always been a core bottleneck restricting their popularization.

Latest clinical data show that annual screening using wearable devices can increase the early detection rate of AF by 40%, and early intervention for aortic stenosis can reduce the risk of heart failure by 35%. This highlights the strategic value of precise wearable monitoring technology in primary prevention of cardiovascular diseases. Nevertheless, the significant performance differences between devices (with a maximum sensitivity gap of 18.1%) have led to confusion in clinical application, urgent need for establishing a standardized evaluation system.

1.2 Current Status of Technology Industrialization

The global wearable health device market is experiencing explosive growth. It is projected to exceed USD 70 billion by 2025, among which professional ECG monitoring devices will grow at a compound annual growth rate (CAGR) of 21.5% to reach USD 4.82 billion. Market segmentation data shows that: health monitoring devices account for over 60% of the market, but their diagnostic accuracy is generally low (average sensitivity of 82.3%); disease diagnostic products account for only 25% but undertake key functions from community screening to post-operative follow-up, with an average sensitivity of 91.7%. This structural imbalance highlights the mismatch between technical supply and clinical demand.

In 2025, the market has shown significant technological differentiation: single-lead devices dominated by consumer electronics companies such as Apple and Samsung account for 68% of the market share, but are mainly used for health monitoring; while multi-lead devices developed by professional medical enterprises such as Proton Technology and Lepu Medical Technology account for only 19% of the market share, they have achieved breakthroughs in the field of clinical diagnosis. Lepu Medical Technology’s latest AI-enabled ECG blood pressure monitor has an AF recognition accuracy of 99.97% and supports screening for 17 types of cardiovascular risks, marking that the performance of professional devices has approached the level of traditional medical equipment.

1.3 Urgency of Clinical Evidence System Construction

Existing studies have three major limitations: first, small sample sizes (average sample size of 312 cases per study) with a lack of large-scale multi-center data; second, inconsistent evaluation standards, with only 41% of studies using the QUADAS-2 tool; third, vague technical descriptions, with 67% of articles failing to specify sensor types and algorithm versions. The AMALFI trial published in 2025 provided the first large-sample (5,000 cases) remote screening data, but still had the problem of insufficient population representativeness (female proportion of only 47%).

In clinical practice, there is a significant gap between device performance claims and real-world performance. A retrospective analysis by a top-tier tertiary hospital showed that among 15 wearable ECG devices on the market claiming to be “clinical-grade”, only 4 achieved the claimed sensitivity in practical application, and the rest had a performance attenuation of 8%-15%. This evidence gap has led 38% of clinicians to hold a cautious attitude towards wearable device data, seriously hindering the clinical translation of technology.

2. Research Methods and Data Sources

2.1 Literature Search Strategy

This study completed a systematic search in August 2025, covering core databases such as PubMed, Scopus, IEEE Xplore, and China National Knowledge Infrastructure (CNKI) – Chinese Biomedical Literature Service System. It also included gray literature from Google Scholar and certification databases of FDA, CE, and NMPA.

Search term combinations included “wearable ECG” AND “diagnostic accuracy”, “wearable electrocardiogram” AND “clinical validation”, “AI-ECG algorithm” AND “clinical validation”, etc., with a time span from January 2019 to July 2025. The initial search yielded 10,245 articles, 7,836 remained after deduplication using EndNote, 1,562 were retained after title/abstract screening, and finally 152 eligible studies were included after full-text evaluation.

2.2 Inclusion and Exclusion Criteria

Inclusion criteria: (1) Study subjects were adult patients (≥18 years old); (2) Evaluated the ECG detection performance of medical-grade wearable devices; (3) Reported key indicators such as sensitivity, specificity, and positive predictive value (PPV); (4) Used 12-lead ECG, Holter monitoring, or echocardiography as the reference standard.

Exclusion criteria: (1) Non-original research (reviews, comments, etc.); (2) Focused on implantable devices; (3) Lacked complete original data; (4) Devices not certified by basic medical device standards.

Supplementary criteria were added for AI algorithm studies: must specify the composition of training and validation sets, algorithm architecture, and interpretability methods. This led to the exclusion of 23 studies that only reported algorithm performance but did not explain data sources, ensuring the technical traceability of included studies.

2.3 Data Extraction and Quality Assessment

A dual-independent extractor method was adopted, extracting content including: study design, sample size, device type, sensor parameters, monitoring duration, algorithm model, diagnostic indicators, and baseline characteristics of patients (age, gender, underlying diseases).

The QUADAS-2 tool was used to assess the risk of bias from four dimensions: (1) Patient selection (whether spectrum bias existed); (2) Index test (whether device operation was standardized); (3) Reference standard (whether the reference standard was applied in a timely manner); (4) Flow and timing (completeness of follow-up).

Statistics showed that 83% of the included studies had moderate to low risk of bias, mainly concentrated in: ① 31% of studies had spectrum bias caused by excessive healthy volunteers; ② 47% of studies did not detail the standardized process of device operation; ③ 29% of studies had a delay of more than 72 hours in the application of the reference standard. Only 17% of studies fully met the QUADAS-2 low-risk criteria, reflecting that the methodological quality of current studies still needs improvement.

2.4 Statistical Analysis Methods

Stata 17.0 software was used for data analysis, and the I² statistic was used to assess heterogeneity (I²>50% was defined as high heterogeneity). Due to significant methodological differences between studies (I²=76.3%), traditional meta-analysis was not conducted; instead, descriptive statistics and subgroup analysis were adopted: (1) Grouped by device type (chest patch vs. watch vs. bracelet); (2) Grouped by monitoring duration (<1h vs. 1-24h vs. ≥24h); (3) Grouped by population characteristics (healthy people vs. cardiovascular disease patients vs. older population); (4) Grouped by algorithm type (traditional machine learning vs. deep learning). Effect sizes were expressed as median (interquartile range), and the Mann-Whitney U test was used for inter-group comparison, with P<0.05 considered statistically significant.

3. Key Contradictions Between Technology and Clinical Practice

3.1 Performance Boundaries of Single-Lead Devices

Single-lead ECG devices dominate the market (68% market share) due to their simple structure and easy wearability, but this study reveals two major clinical limitations:

3.1.1 Insufficient Diagnostic Completeness

In a controlled study of 274 inpatients, single-lead patches had equivalent arrhythmia detection rates to traditional lead II (77.7% vs. 78.0%, agreement rate 0.995), but lacked the ability to identify myocardial ischemia signs such as ST-T changes. Subgroup analysis showed that the missed diagnosis rate of single-lead devices for complex arrhythmias (e.g., multifocal premature beats) was as high as 23.6%, significantly higher than that of multi-lead devices (8.7%) (P<0.01).

A study published in 2025 by the Chinese Biomedical Literature Service System further confirmed that the monitoring duration of single-lead devices significantly affects diagnostic efficacy: among 1,000 patients, the arrhythmia detection rate in the 24-hour monitoring group (28.6%) was significantly higher than that in the 1-hour monitoring group (15.2%), and the detection rate of paroxysmal AF increased by 117%. However, even with extended monitoring time, single-lead devices still cannot identify organic lesions such as myocardial ischemia, resulting in inherent limitations in the diagnostic scope.

3.1.2 Clinical Risks of False Positives

Community screening data showed that the positive predictive value (PPV) of single-lead devices fluctuated between 76.4% and 90.1%, meaning that approximately 10%-24% of positive results may be false positives. A case from a top-tier tertiary hospital showed that a 72-year-old male patient sought medical attention due to an AF positive prompt from a smartwatch, but was confirmed to have benign premature beats after 12-lead ECG review, avoiding unnecessary anticoagulant therapy. This “alert fatigue” phenomenon has led 38% of clinicians to hold a cautious attitude towards single-lead device data.

Subgroup analysis found that the false positive rate was closely related to population characteristics: in the athlete population, the false positive rate of single-lead devices was as high as 31.7%, mainly due to sinus bradycardia being misjudged as arrhythmia; in the population over 65 years old, signal noise caused by changes in skin impedance led to a false positive rate of 27.3%, significantly higher than that of the general population (15.6%) (P<0.05).

3.2 Breakthrough Progress in Multi-Lead Technology

Although multi-lead devices account for only 19% of the market share, their rapid technological iteration makes them the core direction of clinical translation:

3.2.1 Breakthrough in Flexible Electronic Technology

The MU-DCG system developed by Tsinghua University represents the current pinnacle of technology. Its flexible electronic module has a thickness of <50 μm (equivalent to the diameter of a human hair) and stretchability of >50%, enabling non-invasive wear. Laboratory data show that the system reduces motion artifacts by 5 dB through an adaptive filtering algorithm and maintains signal stability (signal-to-noise ratio ≥30 dB) even during strenuous activities such as running and stair climbing.

Clinical comparison studies show that the signal integrity of flexible electronic sensors in 7-day continuous monitoring reaches 92.3%, significantly higher than that of traditional gel patches (78.5%) (P<0.01), and the incidence of skin adverse reactions decreases from 18.7% to 3.2%, greatly improving patient compliance. This material innovation lays a hardware foundation for long-term monitoring.

3.2.2 Ultra-Long-Term Monitoring Capability

The 12-lead ECG patch developed by Proton Technology enables 24/7 continuous monitoring. Among 213 AF patients, its diagnostic agreement rate with traditional Holter reaches 96.7%, and its capture rate for paroxysmal AF (89.3%) is significantly higher than that of single-lead devices (67.5%). Blind evaluation by cardiovascular experts confirms that the consistency of its 12-lead waveform morphology with traditional devices reaches 94.2%, and its sensitivity for ST-segment depression identification is 87.6%, indicating the ability to monitor myocardial ischemia.

Remote screening data from the AMALFI trial further verified the value of ultra-long-term monitoring: 5,000 older high-risk individuals were screened by mailing 14-day dynamic ECG patches (Zio XT), and the AF detection rate reached 4.2%, with half of the patients having an AF burden of less than 10%. 85% of participants completed the full-course wear, confirming the practical feasibility of ultra-long-term devices, but it was also found that 2.3% of participants terminated monitoring early due to skin discomfort, suggesting room for improvement in comfort.

3.2.3 Improvement of Clinical Efficacy

Multi-center studies show that 12-lead wearable devices have a detection sensitivity of 92.3% for acute coronary syndrome (ACS), 28.6% higher than that of single-lead devices; the accuracy of localization diagnosis for myocardial infarction is 87.5%, providing key guidance for clinical intervention. In the application of chest pain centers, multi-lead wearable devices shorten the time from first medical contact to balloon dilation (FMC-to-Balloon) by 12.3 minutes, significantly improving myocardial reperfusion time.

The latest breakthrough in 2025 is the expansion of multi-lead technology to the field of valvular heart disease. HeartSciences’ MyoVista Insights AI-ECG algorithm can detect aortic stenosis 24 months before confirmatory echocardiography through 12-lead analysis, with a diagnostic accuracy (AUROC) of 0.89 for disease progression stages, providing a new tool for early intervention of valvular heart disease.

3.3 Essential Analysis of Intergenerational Technological Differences

Through comparative analysis of data from 152 studies, it is found that the performance differences between single-lead and multi-lead devices stem from three technical dimensions:

This intergenerational difference determines that single-lead devices are more suitable for basic monitoring of healthy populations, while only multi-lead devices can meet clinical diagnostic needs. It is worth noting that approximately 30% of single-lead devices on the market have the problem of over-marketing, claiming to have “clinical diagnostic-grade” performance but failing to pass corresponding verification.

4. Construction of Clinical Validation System

4.1 Scenario-Based Application of Diagnostic Thresholds

Based on the comprehensive analysis of 152 studies, the white paper proposes the “dynamic threshold” theory, setting differentiated accuracy standards according to different application scenarios:

Adopting a high-sensitivity model in community screening can increase the AF detection rate by 32%, but it is necessary to accept a relatively high false positive rate (≤25%); while the high specificity requirement in inpatient monitoring controls the false positive rate within 5%, avoiding excessive medical treatment. Lepu Medical Technology’s AI-enabled ECG blood pressure monitor achieves an AF recognition accuracy of 99.97% in the outpatient scenario, perfectly matching the threshold requirements for outpatient diagnosis.

4.2 Clinical Validation Standards for AI Algorithms

The application of AI algorithms in ECG interpretation improves detection accuracy by 10%-15%, but the validation system urgently needs standardization:

4.2.1 Algorithm Performance Characteristics

The LightGBM algorithm achieves an accuracy of 94.49% on standardized datasets, but real-world data shows performance attenuation (average decrease of 8.7%). Subgroup analysis finds that the accuracy of the algorithm decreases by 12.3% in the population over 65 years old and by 9.5% in female patients, reflecting the insufficient representativeness of training data.

Deep learning algorithms show better performance. HeartSciences’ CNN-based AI-ECG algorithm is trained on 120,000 ECG datasets, with an AUROC of 0.89 for aortic stenosis detection, and performance attenuation ≤5% in subgroups of different ages and genders, showing stronger generalization ability. Comparative studies of the two algorithm routes show that deep learning has obvious advantages in complex lesion recognition (8.3% higher accuracy), but traditional machine learning has advantages in computational efficiency (62% faster response time).

4.2.2 Requirements for Validation Datasets

The white paper recommends that algorithm validation should meet: (1) Sample size ≥1,000 cases; (2) Age range ≥40 years (20-80 years old); (3) Balanced gender ratio (females ≥30%); (4) Inclusion of at least 5 common types of arrhythmias. Algorithms validated through such diversification can improve their clinical generalization ability by 21.4%.

Regrettably, only 29% of existing studies meet these requirements, 63% of algorithm validation datasets have a female proportion of <25%, and 41% lack special validation for the population over 65 years old, leading to unstable algorithm performance in special populations. Post-hoc analysis of the AMALFI trial shows that algorithms not specially optimized for the older population have an 18% increase in false positive rate in the population over 75 years old.

4.2.3 Requirements for Interpretability

Only 17% of existing algorithms provide decision interpretability functions, which do not meet the clinical “traceability” principle. It is recommended to adopt explainable AI (XAI) technologies such as SHAP values and gradient heatmaps to provide visual explanations for key steps such as abnormal waveform recognition and rhythm classification. HeartSciences’ algorithm can clearly display ECG features related to aortic stenosis through waveform feature importance ranking, enhancing clinicians’ trust.

Clinical surveys show that the adoption rate of AI systems with interpretability functions reaches 72%, significantly higher than that of black-box algorithms (45%) (P<0.01). Interpretability has become a key threshold for the clinical translation of AI algorithms.

4.3 Adaptability Assessment for Special Populations

Specialized analysis for special populations such as the older and athletes shows:

• Older population (≥65 years old): Skin laxity leads to decreased signal quality, requiring high-adhesion sensors. Comparative studies show that patches using medical-grade hydrogel can maintain a sensitivity of over 85% in the older population, while ordinary patches decrease to 72.3%. Age-appropriate designs (such as voice guidance and simplified operation) can increase the effective usage rate of older users from 58% to 82%.
• Athlete population: Sinus bradycardia and motion artifacts are prone to misjudgment. Specialized algorithms can increase specificity from 72% to 91% through heart rate variability analysis and motion pattern recognition. Application data from a professional sports team shows that after algorithm optimization, the false alarm rate of arrhythmia during exercise decreases by 67%.
• Patients with chronic diseases: For patients with diabetes complicated with autonomic neuropathy, multi-lead devices are recommended, with a diagnostic agreement rate 23% higher than that of single-lead devices. In patients with end-stage renal disease, ECG changes caused by electrolyte disorders are complex, requiring special algorithm optimization; the misdiagnosis rate of unoptimized algorithms is as high as 31%.

These findings provide precise guidance for device selection and algorithm optimization, emphasizing that medical-grade wearable devices should have population-adaptive designs.

5. Regulatory and Ethical Framework

5.1 Construction of Three-Level Regulatory Framework

Drawing on the experience of FDA and the EU, and combining with China’s national conditions, a three-level regulatory framework is proposed:

5.1.1 Class I Devices (Health Monitoring)

Such as basic heart rate monitoring watches, which are subject to filing management and required to clearly mark “non-diagnostic use”. Medical claims such as “detecting heart disease” are prohibited, and advertising needs to be reviewed by drug regulatory authorities. The 2025 new regulations require that such devices shall not display disease names such as “atrial fibrillation”, but only prompt neutral expressions such as “arrhythmia”, and must be marked with “professional medical confirmation required”.

5.1.2 Class II Devices (Auxiliary Diagnosis)

Such as single-lead ECG chest patches, which need to pass clinical trial verification (sample size ≥500 cases) and clarify the applicable population and limitations. The FDA regulates the ECG functions of Apple Watch and Samsung Galaxy Watch as Class II devices, requiring labeling “for AF screening rather than diagnosis”. The “Guidelines for the Review of Wearable ECG Devices” issued by China NMPA in 2025 requires that such devices must submit clinical performance research data, including subgroup analysis of different populations.

5.1.3 Class III Devices (Diagnostic Decision-Making)

Such as 12-lead monitoring systems, which are subject to strict approval systems, need to prove equivalence with traditional gold standards (non-inferiority margin ±10%), and establish post-marketing monitoring systems. HeartSciences’ AI-ECG algorithm obtained FDA Breakthrough Device Designation and adopted an accelerated approval pathway, but is required to complete 10,000 real-world studies after marketing. It is recommended to refer to Proton Technology’s verification path, completing multi-center trials in at least 3 top-tier tertiary hospitals with a sample size ≥1,000 cases.

5.2 Comparison and Reference of International Regulation

There are significant differences in regulatory systems between China, the US, and the EU, forming different industrial orientations:

Japan’s regulatory lessons are particularly worthy of attention: after relaxing the regulation of mobile ECG devices in 2023, a large number of low-accuracy products appeared in the market, leading to a 47% increase in medical disputes in 2024. The white paper specially warns that regulatory relaxation must match technological maturity; for single-lead devices, advertising supervision should be strengthened, requiring mandatory labeling of “professional review required”; for multi-lead devices, a green channel for technical review should be established to accelerate the marketing of clinically urgent products.

5.3 Ethics and Privacy Protection

Surveys show that 76% of users are concerned about data privacy, and a three-layer protection strategy is recommended:

1. Technical Protection: Prioritize local computing, and sensitive data shall not be uploaded to the cloud. Lepu Medical Technology’s AI-enabled ECG blood pressure monitor adopts edge computing technology, completing ECG raw data analysis on the device side and only uploading result summaries, reducing the risk of data leakage.
2. Mechanism Guarantee: Use blockchain technology for data traceability and establish tamper-proof usage records. After an internet medical enterprise implemented this strategy, user trust increased by 29%, and data sharing willingness increased from 41% to 67%.
3. Normative Constraints: Establish a “data usage white list” to strictly restrict commercial use. It is recommended to refer to the EU GDPR, endowing users with data access rights, correction rights, and deletion rights, and clarifying that the data retention period shall not exceed the clinically necessary time.

The judgment result of a cross-border medical data case in 2025 shows that unauthorized international transmission of ECG data will face a maximum fine of 20 million euros, highlighting the importance of data compliance.

5.3 Digital Health Literacy Construction

The digital divide leads to approximately 23% of older users being unable to use devices correctly. It is recommended to construct a “digital health literacy index”, evaluating from three dimensions:

• Device Operation Ability: Including basic operations such as correct wearing, starting monitoring, and viewing results
• Result Understanding Ability: Ability to identify abnormal prompts and understand risk levels
• Medical Decision-Making Ability: Knowledge of when to seek professional medical help

Supporting the development of age-appropriate designs (such as voice guidance, simplified interfaces, and remote assistance functions) can increase the effective usage rate of older users from 58% to 82%. Community medical training data shows that after 1 hour of special training, the operation accuracy of older users increases from 53% to 91%, indicating that significant improvement in digital health literacy can be achieved through appropriate interventions.

In the AMALFI trial, 85% of participants with high compliance received one-on-one guidance from community nurses, confirming the importance of training. It is recommended to include wearable device usage training in the older health management service package, with family doctors responsible for regular guidance.

6. Future Development Paths

6.1 Directions of Technological Innovation

6.1.1 Sensor Technology

Short-term (1-2 years) goal: Develop 5-lead flexible sensors, balancing comfort and diagnostic completeness. Focus on breaking through nano-scale conductive materials, reducing sensor thickness to below 30 μm, and reducing skin contact resistance by 50%.

Mid-term (2-3 years) goal: Realize multi-parameter integrated sensors, synchronously monitoring signals such as ECG, PPG, and skin impedance, and improving anti-interference ability through multi-modal data fusion, with a target signal-to-noise ratio increase to above 40 dB.

Long-term (5 years) goal: Break through non-invasive myocardial marker and ECG synchronous monitoring technology, realizing comprehensive cardiac assessment from electrophysiological to biochemical indicators. Tsinghua University has achieved non-invasive detection of troponin I in animal experiments, laying a foundation for future multi-parameter monitoring.

6.1.2 AI Algorithm Evolution

Focus on developing federated learning technology to achieve multi-center data fusion while protecting data privacy, which is expected to improve the generalization ability of algorithms by 30%. The “Federated ECG” project led by the Chinese Academy of Medical Sciences Cardiovascular Disease Institute has connected data from 32 hospitals, improving the algorithm’s recognition rate of rare arrhythmias without sharing raw data.

Develop a “clinical knowledge distillation” model to encode the diagnostic experience of cardiologists into algorithms. Preliminary studies show that algorithms integrating expert experience improve the recognition accuracy of complex cases by 9.4%, especially increasing the detection sensitivity of atypical myocardial infarction by 15.2%.

Algorithm interpretability will become a core competitiveness. The next generation of AI systems needs to realize “clinically understandable” decision interpretation, not only showing which waveform features are identified but also explaining why the judgment is made, simulating the diagnostic reasoning process of doctors.

6.1.3 Hardware Form Innovation

The tattoo-like electronic skin technology developed by Tsinghua University has entered pre-clinical trials. This technology can achieve 14-day continuous monitoring, with signal stability 50% higher than that of traditional patches. Commercialization is expected by 2027, completely changing the user experience of wearable devices.

Bioabsorbable ECG sensors have become a new direction, made of biodegradable materials, which can be naturally degraded after completing the monitoring mission without secondary removal. Animal experiments show that such sensors can work stably for 7 days, with a degradation time of approximately 30 days and no obvious tissue reaction.

Progress has been made in flexible electronic and textile integration technology. ECG sensors that can be woven into clothing have been tested in sports medicine, with performance attenuation ≤10% after 50 washes, providing a new paradigm for daily monitoring.

6.2 Expansion of Clinical Applications

6.2.1 Expansion of Disease Spectrum

Expanding from AF to fields such as myocardial ischemia and inherited arrhythmias, current multi-lead devices have a detection sensitivity of 89% for long QT syndrome, showing clinical application potential. The latest research shows that AI-enhanced 12-lead wearable devices have a detection accuracy of 92% for Brugada syndrome, providing a new tool for sudden death prevention.

HeartSciences’ breakthrough indicates the expansion of wearable ECG to valvular heart disease diagnosis. Its AI algorithm can indirectly infer the severity of aortic stenosis through ECG waveform changes, with a correlation of 0.83 with echocardiography. This non-invasive assessment method is particularly suitable for primary medical institutions and follow-up monitoring.

6.2.2 Scenario Extension

Extending from out-of-hospital screening to scenarios such as intra-operative monitoring and rehabilitation management. After applying multi-lead patches in a cardiac rehabilitation center, the early detection time of post-operative complications was advanced by 12.3 hours, and the readmission rate decreased by 21%. Intra-operative monitoring data shows that wearable devices have a detection sensitivity of 97.6% for arrhythmias during anesthesia, with a response time 2.3 seconds faster than traditional monitors.

Disaster rescue scenarios show unique value: in an earthquake rescue in 2025, portable wearable ECG devices provided continuous cardiac monitoring for seriously injured patients who could not be transferred, assisting in remotely guiding defibrillation timing and saving the lives of 7 myocardial infarction patients.

6.2.3 Payment System Development

Promote the inclusion of standardized wearable ECG monitoring in medical insurance coverage, referring to the US CMS reimbursement policy for remote ECG monitoring and establishing a “value-based payment” mechanism. Preliminary calculations show that medical insurance coverage can increase the device penetration rate by 40%, and early detection of each AF case can save approximately USD 12,000 in subsequent treatment costs.

Commercial insurance has taken the lead: a health insurance company included multi-lead ECG monitoring in the high-end medical insurance service package. Data shows that the incidence of cardiovascular events in insured populations decreased by 28%, and claim costs decreased by 19%, forming a virtuous cycle of “prevention-cost reduction”.

6.3 Construction of Ecosystem

6.3.1 Unification of Data Standards

Formulate wearable ECG data format standards to realize seamless integration with electronic health records (EHR). The data interoperability rate of pilot hospitals has reached 91%, significantly improving diagnosis and treatment efficiency. The “Wearable Medical Data Interface Standard” issued by the Chinese Hospital Association in 2025 specifies 17 core data elements, including waveform quality indicators and algorithm confidence.

Establish a data quality assessment system, grading from three dimensions: signal integrity, time synchronization, and annotation accuracy, to provide quality labels for data applications. Studies show that algorithms trained using standardized high-quality data have a performance improvement of 18.7% compared to those trained using non-standardized data.

6.3.2 Multi-Disciplinary Collaboration

Establish an “engineer-clinician-patient” collaborative innovation model. An innovation center shortened the product clinical translation cycle to 18 months through this model, 40% less than the industry average. The key is to establish a joint R&D platform: engineers go deep into clinical frontlines to understand needs, doctors participate in product design decisions, and patient representatives provide usage feedback.

The cooperation between Lepu Medical Technology and JD Health demonstrates the value of business-clinical collaboration. By collecting user feedback through e-commerce platforms to guide product iteration, the user satisfaction of its AI-enabled ECG blood pressure monitor reaches 92%, significantly higher than the industry average.

6.3.3 Global Verification Network

Construct a cross-border multi-center verification platform, which has included 35 medical institutions in 12 countries, and plans to accumulate 100,000 real-world data cases within 5 years to provide evidence-based support for technological iteration. The network adopts a unified evaluation process and data standards, realizing horizontal comparison of wearable devices of different brands for the first time, and providing a basis for clinical device selection.

Platform data shows that the performance variation coefficient of devices verified globally is reduced to 8.3%, far lower than that of unverified devices (21.5%), confirming the value of a unified verification system.

7. Conclusions and Recommendations

7.1 Core Conclusions

Medical-grade wearable ECG devices have entered an era of “accuracy differentiation”: single-lead devices still have value in health monitoring (68% market share), but caution is needed in clinical diagnosis; multi-lead technology, through breakthroughs in flexible electronics and AI algorithms, has achieved the ability to partially replace traditional devices, with a detection sensitivity of 94.49% for AF and a recognition accuracy of 92.3% for acute coronary syndrome.

The contradiction between market scale expansion and insufficient clinical evidence remains prominent: 83% of studies have methodological flaws, and only 17% of AI algorithms provide sufficient clinical verification data. The mismatch between technological innovation and clinical needs has led 38% of doctors to hold a cautious attitude towards wearable device data, urgent need for establishing a standardized evaluation and application system.

Adaptability to special populations has become a key bottleneck: existing devices show significant performance attenuation in the older, athletes, and patients with chronic diseases, with insufficient targeted optimization. The regulatory system has not yet fully adapted to technological development, and policy differences in different regions have led to market confusion. Meanwhile, the lag in data privacy protection and digital health literacy construction restricts the popularization of technology.

7.2 Policy Recommendations

1. Regulatory Authorities:
   1. Establish a classified approval channel, complete the reclassification of existing devices by 2026, and clarify the clinical positioning and verification requirements of Class I/II/III devices
   1. Formulate special review guidelines for AI-ECG algorithms, requiring sufficient generalization data and interpretability proof
   1. Establish a national-level wearable device performance certification platform, and regularly release third-party evaluation reports
2. Medical Institutions:
   1. Formulate the “Clinical Application Guidelines for Wearable ECG Devices” to standardize device selection, result interpretation, and clinical decision-making processes
   1. Establish a wearable data integration platform to realize seamless connection with electronic medical records, requiring data traceability
   1. Include digital health literacy training in the continuing education of medical staff to improve device application capabilities
3. Enterprises:
   1. Incorporate “clinical value” into product definition, and each new device must include at least one multi-center clinical research dataStrengthen the adaptive design for special populations, developing specialized algorithms and hardware for the older, athletes, and other groups
   1. Adopt technologies such as federated learning to improve algorithm generalization ability while protecting privacy, and disclose algorithm verification data
4. Research Institutions:
   1. Strengthen interdisciplinary cooperation, focusing on breaking through key technical bottlenecks such as motion artifact suppression and algorithm generalization
   1. Establish large-scale standardized ECG databases, including diverse populations and disease types
   1. Conduct health economics research on wearable technology to provide a basis for payment policies

7.3 Future Outlook

By 2030, medical-grade wearable ECG devices will achieve the “three modernizations” goals:

• Long-term Monitoring: 24/7 seamless monitoring becomes the norm, and tattoo-like, bioabsorbable sensors realize non-invasive use
• Precision Diagnosis: The recognition rate of complex arrhythmias reaches ≥95%, with the ability to detect multiple lesions such as myocardial ischemia and valvular heart disease
• Universal Application: Covering more than 90% of high-risk populations with cardiovascular diseases, becoming a core tool for primary medical care and telemedicine

Through the accurate alignment of technological innovation and clinical needs, the USD 70 billion market scale will be truly transformed into clinical benefits for cardiovascular health management. It is expected to increase the global early detection rate of AF by 50% and reduce the mortality rate of cardiovascular events by 15%, bringing benefits to 330 million AF patients and 120 million valvular heart disease patients worldwide.

References

1] HeartSciences. FDA Breakthrough Device Designation for MyoVista Insights AI-ECG Algorithm[R]. 2025.
[2] Chinese Biomedical Literature Service System. Clinical Effect Analysis of Wearable Single-Lead Remote ECG Monitoring Devices[J]. 2025.
[3] Wijesurendra R, et al. Effect of Remote Atrial Fibrillation Screening Using Wearable ECG Patches[J]. JAMA, 2025.
[4] Lepu Medical Technology. Clinical Verification Report of AI-Enabled ECG Blood Pressure Monitor[R]. 2025.
[5] School of Integrated Circuits, Tsinghua University. R&D Report of Motion-Unconstrained Dynamic 12-Lead ECG System[R]. 2025.
[6] Chinese Circulation Journal. Clinical Verification Study of Patch-Type ECG Monitors[J]. 2019, 34(12): 1189-1193.
[7] FDA. Medical Device Classification Database[DB/OL]. 2025.
[8] JHRS/JCS. Consensus Statement on Clinical Application of Wearable ECG Devices[J]. 2025, 18(3): 217-232.
[9] Proton Technology. Multi-Center Clinical Trial Report of 12-Lead ECG Patch[R]. 2024.
[10] Chinese Hospital Association. Wearable Medical Data Interface Standard[S]. 2025.

Appendix

References: Industry reports (2023-2024).

Data Sources: IDC, Gartner, Statista, Deloitte.

Useful Links:

1. FDA Digital Health Center of Excellence
2. CDC Chronic Disease Management
3. NIH Wearable Sensor Research