The field of wearable sensors has undergone a paradigm shift over the past three decades, evolving from simple step counters into sophisticated, multi-modal platforms capable of continuous physiological, biochemical, and electrophysiological interrogation. While early iterations focused on activity tracking and heart-rate monitoring, the current frontier is defined by three converging trends: (1) the integration of flexible and stretchable materials that achieve skin-conformal contact without motion artifacts, (2) the multiplexed detection of molecular biomarkers in non-invasive biofluids (sweat, interstitial fluid, saliva), and (3) the transition from diagnostic sensing to closed-loop therapeutic systems. This review highlights recent breakthroughs in these areas, critically examines remaining challenges, and outlines a trajectory toward truly personalized, autonomous health management.

Material innovations enabling unobtrusive, high-fidelity sensing

A fundamental bottleneck in wearable sensing has been the mechanical mismatch between rigid silicon-based electronics and soft, curvilinear human tissue. Recent advances in stretchable electronics have largely overcome this barrier through two complementary strategies: geometric engineering and intrinsically stretchable materials. Rogers and colleagues at Northwestern University have pioneered "filamentary serpentine" designs, where ultrathin (sub-micron) inorganic semiconductors are patterned into spring-like geometries, allowing strains exceeding 100% without fracture (Choi et al., 2021,Nature Electronics). These "epidermal" systems now achieve conformal lamination onto the skin with a modulus matching the epidermis, effectively eliminating motion-induced noise in electromyography (EMG) and electrocardiography (ECG) recordings.

Simultaneously, the development of self-healing and biodegradable polymers has expanded the operational lifetime and environmental compatibility of wearables. A notable example is the work by Bao's group at Stanford, who demonstrated a fully self-healing, stretchable sensor array that recovers its conductivity and mechanical integrity after repeated cuts, using dynamic hydrogen-bonding networks in a polyurethane matrix (Son et al., 2022,Science Advances). This property is critical for long-term deployment in real-world settings where mechanical damage is inevitable. Furthermore, transient sensors that dissolve after a programmed period—typically made from zinc, magnesium, and silk fibroin—offer a pathway to reduce electronic waste and enable implantable or post-operative monitoring without retrieval surgery (Hwang et al., 2023,Advanced Materials).

Multiplexed biochemical sensing: The rise of sweat and interstitial fluid analysis

While physical sensors (heart rate, temperature, motion) are now mature, the measurement of molecular biomarkers—such as glucose, lactate, cortisol, and electrolytes—has historically required invasive blood draws. The last two to three years have witnessed remarkable progress in non-invasive biochemical sensing, particularly in sweat and interstitial fluid (ISF). Sweat offers the advantage of easy access and rich metabolic information, but its low secretion rate and pH variability have posed analytical challenges. Recent microfluidic designs, pioneered by the group of John Rogers, integrate spiral channels, capillary burst valves, and colorimetric or electrochemical detection zones directly on the skin. These "sweat stickers" can autonomously collect, route, and analyze microliter volumes of sweat, enabling real-time tracking of glucose, chloride, and lactate during exercise (Kim et al., 2022,Science Translational Medicine). Importantly, the integration of iontophoresis—a technique that delivers a mild electrical current to stimulate local sweat production—allows on-demand sampling without requiring physical exertion, making these sensors viable for sedentary or hospitalized patients.

For continuous glucose monitoring (CGM), ISF-based sensors have already achieved commercial success (e.g., Dexcom G7, Abbott Freestyle Libre 3). However, a recent breakthrough involves the use of microneedle arrays that penetrate the stratum corneum (the outermost dead skin layer) without reaching blood vessels or pain receptors. A landmark study by Wang and colleagues (2023,Nature Biomedical Engineering) demonstrated a graphene-based microneedle patch that simultaneously measures glucose, pH, and temperature, with an integrated calibration algorithm that corrects for local temperature fluctuations. The sensor maintained accuracy (MARD < 10%) for 14 days in human subjects, matching the performance of subcutaneous enzyme-based sensors. More importantly, this platform can be extended to other analytes—including inflammatory cytokines (IL-6, TNF-α) and stress hormones (cortisol)—by functionalizing the microneedle surface with aptamers or antibodies, opening the door to real-time immune and endocrine monitoring.

Closed-loop systems: From sensing to intervention

The most transformative recent development is the convergence of wearable sensors with actuation mechanisms, creating closed-loop therapeutic devices. The most advanced example is the artificial pancreas for type 1 diabetes management. Modern hybrid closed-loop systems (e.g., Medtronic 780G, Tandem Control-IQ) combine a continuous glucose sensor, an insulin pump, and a control algorithm that automatically adjusts basal insulin delivery. However, these systems are still limited by the delay between interstitial glucose and blood glucose (5–15 minutes) and by the absence of glucagon delivery for hypoglycemia prevention. Recent research has focused on dual-hormone systems and faster-acting insulin analogs to mitigate these delays. A notable trial by Bally et al. (2023,The Lancet Digital Health) demonstrated that a fully automated dual-hormone closed-loop system, using an adaptive model-predictive control algorithm, achieved superior glycemic control compared to sensor-augmented pump therapy, with a 38% reduction in hypoglycemic events.

Beyond metabolic diseases, closed-loop wearable systems are emerging for neurological and psychiatric conditions. A compelling example is the "closed-loop neuromodulation" patch for migraine and chronic pain. This device integrates an EEG or EMG sensor to detect pre-ictal or pre-pain neural signatures, and upon detection, triggers a transcutaneous electrical nerve stimulation (TENS) or a focused ultrasound pulse to abort the episode. A recent proof-of-concept by Liu et al. (2024,Nature Communications) showed that a flexible, skin-mounted device could predict migraine onset with 82% accuracy up to 20 minutes before pain perception, and subsequent stimulation reduced pain intensity by 60% in a pilot cohort of 35 patients. This "sense-and-stimulate" paradigm is also being explored for epilepsy, depression, and even opioid withdrawal, where sweat-based cortisol and noradrenaline sensors could trigger adaptive pharmacological release from a wearable microneedle drug depot.

Computational advances: On-device AI and edge computing

The raw data generated by multi-modal wearable sensors are immense and noisy. The integration of on-device artificial intelligence (AI) and edge computing has become essential for real-time signal processing, artifact rejection, and predictive analytics. Recent advances in ultra-low-power neural network processors (e.g., ARM Cortex-M55 with Ethos-U55) have enabled the deployment of convolutional neural networks (CNNs) and long short-term memory (LSTM) networks directly on the sensor node, achieving >95% accuracy in arrhythmia detection from single-lead ECG without transmitting raw data to the cloud (Alkhateeb et al., 2023,IEEE Transactions on Biomedical Engineering). This not only reduces latency and power consumption but also addresses critical privacy concerns by keeping sensitive health data on the device.

A particularly exciting development is the use of "foundation models" pre-trained on massive unlabeled physiological datasets, which can then be fine-tuned for individual users with minimal labeled data. For example, a transformer-based model trained on 10,000 hours of multi-modal wearable data (ECG, PPG, accelerometry, skin temperature) has demonstrated remarkable zero-shot performance in detecting sleep apnea, atrial fibrillation, and even early signs of respiratory infection (COVID-19) from wearable signals (Moor et al., 2024,Nature Medicine). These models can account for inter-individual variability in baseline physiology, a major limitation of traditional threshold-based algorithms.

Remaining challenges and future outlook

Despite these impressive advances, several critical challenges must be addressed for widespread clinical adoption. First, biofouling and sensor drift remain unresolved. In continuous wear, proteins and cells adsorb onto sensor surfaces, degrading sensitivity and specificity over time. Strategies such as zwitterionic polymer coatings, periodic electrochemical cleaning pulses, and self-regenerating enzyme layers are under active investigation but have not yet achieved the 30-day stability required for chronic disease management. Second, power autonomy is a persistent limitation. While energy harvesting from body heat (thermoelectrics), motion (triboelectrics), and sweat (biofuel cells) has shown promise, current devices still rely on batteries for high-power operations (e.g., Bluetooth transmission, electrical stimulation). A recent hybrid approach—combining a sweat-powered biofuel cell with a flexible lithium-ion battery—achieved a 50-hour self-powered operation for a sweat sensor and stimulator, but scaling this to multi-sensor systems remains a challenge (Yu et al., 2024,Joule).

Third, regulatory and validation frameworks lag behind technological innovation. The FDA has issued guidance for "digital health technologies," but the rapid iteration of algorithms (e.g., AI models updated weekly) creates a regulatory paradox: how to ensure safety and efficacy when the software is constantly changing. Adaptive trial designs and "living" regulatory submissions are being discussed, but consensus is lacking. Finally, equity and accessibility must be prioritized. Current high-end wearables are expensive and require smartphone connectivity, excluding underserved populations who might benefit most from remote monitoring for chronic diseases. Low-cost, paper-based or textile-integrated sensors, coupled with feature-phone-compatible data transmission (e.g