Engineering trustworthy closed-loop bioelectronic systems for cardiovascular disease

Main text

Establishing closed-loop clinical management for cardiovascular disease (CVD) remains more challenging than in many other medical domains. The cardiovascular system couples electrical, mechanical, biochemical, and neurohumoral signals across rapid, tightly interconnected timescales, creating a high-risk environment in which uncertainty can rapidly propagate into therapeutic error. CVD therapy must distinguish pathological states from normal physiological variations in real time, setting it apart from other mature closed-loop systems (e.g., insulin delivery) that manage a single variable over minutes (Figure 1A). Cardiovascular closed-loop therapy must be conceptualized not as integrated hardware but as a trustworthy, life-saving control system that converts uncertain physiological information into actionable, clinically relevant interventions within a safe operating range. The next frontier is no longer integration alone but how to engineer reliability, interpretability, and adaptive decision intelligence into the therapeutic loop.

Figure 1.

Schematic illustration of closed-loop systems for CVD management and their evolutionary pathway

(A) Comparison of complexity: conventional vs. cardiovascular closed-loop systems.

(B) Pioneering prototypes: (i) monitoring system for cardiac graft integration. Inspired by Aoyama et al.¹ (ii) Closed-loop system for blood-pressure management. Inspired by Zou et al.²

(C) Future development toward technological breakthroughs and clinical translation.

Pioneering prototypes and the path toward integration

Recent breakthroughs in prototype bioelectronics have established the feasibility of integrated sophisticated sensing with localized therapeutic action. Lee et al. developed a dual-component platform of ultraflexible nanoelectronics¹ to monitor the integration of transplanted human induced pluripotent stem cell-derived cardiomyocytes (hiPSC-CMs) within host cardiac tissues (Figure 1B). This research offers a long-term, cell-resolved observation window into the electrophysiological mechanisms of arrhythmogenesis. By revealing how these engineered tissues interact with the native myocardium, such platforms provide the essential diagnostic infrastructure required to understand and eventually mitigate post-transplantation risks. While these systems focus on the observational dimension of regenerative medicine, they establish the necessary technological foundation for future architectures that might incorporate active rhythm-management capabilities.

Zou et al. demonstrated a closed-loop bioelectronic patch for intelligent blood-pressure management (Figure 1B).² In this system, nitric oxide release or termination was triggered by the elevation and normalization of blood pressures, allowing immediate therapeutic output in response to physiological input. This represents a step in moving closed-loop cardiovascular control from conceptual possibility to tangible engineering reality, illustrating the translational potential of wearable and unobtrusive therapeutic devices. The decision logic here still relies on a single physiological threshold. It cannot describe comprehensive cardiovascular dynamics. An elevation in blood pressure during intense activity represents a healthy adaptive response, unlike a hypertensive surge at rest.

These prototypes demonstrate that cardiovascular closed-loop systems are no longer speculative: high-fidelity sensing can be coupled to localized therapeutic action. The next challenge is not further integration but the design of controllers that remain reliable under context-dependent cardiovascular dynamics. This challenge involves three tightly coupled layers: the tissue-device interface, the energy-computation architecture, and the decision intelligence that governs therapy (Figure 1C).

Technical barriers for reliable and trustworthy closed-loop systems

The first bottleneck lies at the tissue-device interface, where mechanical mismatch between electronic materials and cardiovascular tissues is a contributor to chronic inflammation, interface instability, and signal degradation during long-term implantation. For cardiovascular closed-loop systems, the mechanical mismatch between electronic materials and pulsating cardiac tissues is the major cause of chronic inflammation and signal degradation during implantation. Unlike devices placed in mechanically quiet tissues, cardiovascular implants must remain functional on continuously beating, deforming, and fluid-exposed surfaces. This places simultaneous demands on mechanical modulus matching cardiac tissue (10–100 kPa), stable adhesion under cyclic strain (millions of heartbeats), resistance to biofouling for signal fidelity, and long-term electrochemical stability. Hybrid conductive hydrogels, such as poly(3,4-ethylenedioxythiophene):polystyrene sulfonate (PEDOT:PSS), offer favorable conductivity and mechanical strength but lack antifouling durability. Other “star” materials are also gaining attention, including liquid metals (e.g., eutectic gallium-indium alloys) and MXene-based composites, owing to their high conductivity, large surface area, and tunable surface chemistry. Despite the above-mentioned developments in materials science, a practical path forward is to further investigate hybrid architectures, including conductive polymer substrates with surface-grafted antifouling layers or interpenetrating networks that embed conductive nanomaterials within antifouling hydrogel matrices. Such strategies allow independent optimization of each constraint. For devices intended for a defined therapeutic window, fully biodegradable materials that disintegrate upon external triggering can circumvent chronic interface problems altogether by removing the foreign body after therapy completion.³

The second engineering hurdle arises from the trade-offs among power consumption, computational capacity, and device size. Cardiovascular monitoring demands high-frequency data acquisition to capture transient-but-critical events such as premature ventricular contractions, which is a substantial burden on battery life. One viable pathway is the use of self-powered and/or low-power edge computing. Intracardiac generators that harvest biomechanical energy from heart motion have already achieved root-mean-square outputs above 100 μW and passed accelerated lifetime tests equivalent to a decade of operation, demonstrating that self-powered autonomy is within reach.⁴ More recent solutions employ distributed computing architectures with decoupled functionality. Urgent functionalities, including sensing, preprocessing, communication, and actuation layers, operate locally on the device, while more computationally intensive processes, including long-term trend analysis, can be offloaded to an external hub or cloud system. This decoupling reduces the size of individual modules, conserves power, and provides fault isolation.

The third and perhaps most vital concern is the reliability of black-box artificial intelligence (AI), which demands a shift toward interpretable and adaptive decision intelligence. In practice, clinicians must understand the rationale behind each AI-assisted recommendation and intervention, treatment plans need to be transparent and interpretable to patients, and regulators also demand traceable chains of logic for each decision-making process. We therefore advocate hybrid control architectures that integrate mechanistic physiological models with modern machine-learning techniques. Mechanistic models define strict safety boundaries to ensure that therapeutic outputs never exceed safe physiological ranges, while machine learning handles the recognition of patient-specific patterns and short-horizon forecasting.⁵

Within this hybrid framework, algorithm selection follows functional layering. While the safety boundary layer relies solely on mechanistic models, the machine-learning-assisted pattern recognition can extract clinical features from noisy signals to provide additional personalized forecasting. As discussed, interpretable models are of particular importance for cardiovascular treatments. For example, Physics-informed neural networks (PINNs) improve decision robustness by embedding known physiological knowledge. Large language models (LLMs) may add value at the supervisory level by integrating heterogeneous information streams, explaining model outputs, supporting clinician interaction, and strengthening auditability; however, they are not yet suitable as primary controllers for high-risk therapeutic actuation.

Beyond algorithm design, software-governed cardiovascular systems will also require explicit management of cybersecurity, data governance, post-deployment performance drift, and update traceability. These considerations are integral components of the safety architecture, because instability in software behavior may directly translate into instability in therapeutic control.

Perspective and translational horizon

The transition of cardiovascular closed-loop systems from laboratory prototypes to mainstream practice will require coordinated advances across clinical, regulatory, and socio-technical landscapes (Figure 1C).

From a clinical perspective, not all cardiovascular indications are equally ready for closed-loop translation. Early clinical deployment will likely be most feasible in conditions with clearly measurable physiological targets, relatively rapid feedback, and well-defined therapeutic objectives, such as refractory hypertension, selected high-risk arrhythmia settings, or early decompensation management in heart failure.

From a regulatory perspective (e.g., in regions such as China), these autonomous devices fall under the stringent requirements for class III medical equipment, which are governed by frameworks established by the National Medical Products Administration. Navigating this advancement requires not only functional evidence but also rigorous validation of software lifecycles and traceability of machine-learning algorithms.

Equally important, integration into clinical workflows will depend on human factors and social trust. Rather than pursuing total machine autonomy, the next decade of development should focus on systems that augment medical expertise by providing actionable insights and well-bounded decision support. Patients, in turn, are also more likely to accept implantable or semi-implantable closed-loop devices when the cognitive burden is low and the treatment process is transparent.

Finally, the long-term impact of trustworthy closed-loop systems will be measured not only by technical sophistication but also by accessibility. Although early implementation may concentrate in high-resource clinical centers, the ultimate goal should be to create scalable, cost-effective solutions for underserved populations. The evolution of cardiovascular care from reactive monitoring to proactive and autonomous intervention is not merely a technical challenge but a social imperative. The ultimate success of this field will depend on whether we can build systems defined as much by reliability, accountability, and responsibility.

Funding and acknowledgments

S.Z. and Z.T. acknowledge the support from National High Level Hospital Clinical Research Funding (grant nos. 2025-PUMCH-C-006 and 2025-PUMCH-A-112), Beijing Natural Science Foundation Program (grant no. 7254403), Peking Union Medical College Hospital Talent Cultivation Program (grant no. UHB12711), and Talent development plan for the future in Medical-Engineering Integration by BRA-CDCHE and ZTA (grant no. MBRC001205040). Y.L. acknowledges the “Hundred Talents Program” of the Chinese Academy of Sciences. S.Z. acknowledges the support from the Major Research Plan of the National Natural Science Foundation of China (grant no. 92570208). The funders had no role in study design, data collection and analysis, decision to publish, or preparation of the manuscript.

Declaration of interests

The authors declare no competing interests.

Published Online: June 5, 2026