Barriers to translating continuous monitoring technologies for preventative medicine.

Biomedical technologies for continuous monitoring offer a unique opportunity for personalized preventative healthcare. Table 1 lists examples of successful translation. However, developing technologies that are capable of true continuous monitoring, or even periodic monitoring, remains challenging. A key limitation is the scarcity of diagnostic signals that can be continuously detected, which results in a narrow range of clinically translated applications. Remotely interfacing technologies primarily monitor physiological signals, digital records and, more recently, wastewater epidemiology. Directly interfacing technologies focus on vital signs, blood glucose and sweat, whereas internally interfacing technologies monitor vital signs, blood glucose and inflammation. Expanding these capabilities to new applications requires robust and reliable device and software design. Another common challenge is ensuring reliable and consistent monitoring over time. These technologies typically require patient adherence, which introduces design trade-offs—for example, a wearable device may prioritize comfort with breathable and skin-conforming materials at the expense of durability, whereas an implantable device may extend battery life through wireless charging but necessitate an external charging pad. This reflects a broader trade-off: longer monitoring periods require greater technological complexity, which can reduce ease of use and, consequently, adherence. Directly and internally interfacing technologies with successful clinical translation often impose operational burdens on users, such as frequent recharging, app-based interaction or secure wireless connectivity, all of which can hinder long-term adherence. Emerging research directions, such as battery-free wireless communication through ambient backscatter, offer promising strategies to enhance patient compliance. Most of the technologies that we have here reviewed are passive and, apart from the implantable devices, non-invasive. However, beyond continuous monitoring, there is substantial potential for active and closed-loop systems that interact with the patient. For example, ingestible devices have been developed separately for drug delivery to the stomach wall and for the probiotic-based detection of gastrointestinal bleeding. Integrating such capabilities could enable a new generation of biomedical devices. Overall, these technologies are not yet a substitute for in-person clinical care, which provides broader, more accurate and more reliable diagnostics. However, continuous monitoring technologies have value as a complement to traditional healthcare, offering caregivers critical insights into disease likelihood and progression. We envision a future in which biomedical technologies integrate seamlessly with existing health systems to support individualized preventative care.

Wearable and portable devices have been developed to continuously monitor a variety of signals, analytes and biomarkers 123 . These technologies use diverse sensing mechanisms and form factors that directly interface with the human body, extracting biophysical, electrophysiological and biochemical signals ( Fig. 3 ). Their success in translation stems from their ease of use, development and testing. Wearable technologies have been widely explored for continuous monitoring. Early wearables featured rigid electronics embedded in accessories, such as watches, shoes and headsets, for tracking physical activity and biophysical signals. Technological advancements have since diverged along multiple paths. One path has led to consumer health products, such as the Apple Watch and the Oura Ring, that have expanded access to continuous monitoring. Another path has led to clinical applications, with devices such as Biobeat and Corsano enabling medical-grade monitoring of vital signs. More recently, efforts have shifted toward skin-conformal wearables that reduce motion artefacts and improve comfort and compliance. These conformal devices have also enabled the non-invasive sampling of unconventional biofluids, including interstitial fluid, sweat, tears and saliva 124 . Such advances create new opportunities in preventative medicine, such as cytokine monitoring in sweat for detecting Crohn’s disease 125 or infections 126 . Continuous monitoring of electrophysiological signals, such as electrocardiography (ECG), electrooculography (EOG), electroencephalography (EEG), electromyography (EMG) and electrogastrography (EGG), has been achieved through mobile phone attachments, home-based portable devices and other form factors. Wearable electrophysiology monitoring could be transformative in terms of the capture and analysis of electrical signals generated by the body to provide critical insights into organ function. In contrast to traditional 12-lead ECG, portable ECGs typically use a pair of leads on the chest and a ground electrode to detect electrical changes during heart-muscle depolarization 127 , 128 , whereas wearable ECG sensors (such as those in the Apple Watch) provide data equivalent to only a single lead 129 . Although consumer devices from Apple and Samsung have expanded access, clinical success has been driven by medical-grade products, in particular the Biobeat patch 130 and the Zio patch by iRHYTHM (ref. 131 ). As with smartwatches, these patches provide single-lead ECG data, yet signal-amplification circuits and noise-reduction algorithms can mitigate the low signal-to-noise ratio of two-lead ECG designs 132 . iRHYTHM’s first-mover advantage eased translational barriers, which lead to clinical adoption. By the time Apple published results from the Apple Heart Study in 2019 (ref. 133 ), demonstrating that the Apple Watch can detect atrial fibrillation, Zio had already validated the detection of atrial fibrillation alongside 11 other arrhythmias 134 . Portable and wearable ECGs provide an affordable solution for continuous cardiovascular monitoring 135 , particularly when integrated into electronic skins 136 or smart clothing 137 to improve adherence. However, the popularity of consumer wearables, as exemplified by the Apple Watch, which received approval by the United States Food and Drug Administration (FDA) for detecting cardiac arrhythmias 138 , has reduced interest in non-wearable ECG technologies. EEG records the electrical activity of the brain through scalp electrodes by measuring voltage fluctuations from cranial neuron activity. It is used to diagnose epilepsy and sleep disorders, and supports brain–machine interfaces for cognitive and prosthetic control. EOG measures potential differences between the inner and outer canthi of the eye, enabling eye movement tracking for ophthalmological applications. Both EEG and EOG rely on direct current potentials, which are prone to drift owing to skin potentials (for EEG) or electrode polarization (for EOG). These issues can be addressed through drift-correction algorithms 139 or capacitance-based sensors that shield electrodes from environmental interference 140 . EEG and EOG electrodes are typically placed on the scalp, the nasopharynx or the ear, and have been incorporated into wearable formats such as electronic skins 141 , earpieces 142 and eyeglasses 143 . EMG detects electrical signals from contracting and relaxing muscle cells, and is used to diagnose neurogenic and myogenic diseases. Continuous EMG monitoring supports applications in rehabilitation, functional analysis and sports medicine. Wearable EMG devices include armbands such as the Myo armband by Thalmic Labs 144 , 145 , wearable patches such as the Pico EMG (ref. 146 ) and smart clothing such as STRIVE shorts 147 . EGG detects electrical activity in the gastrointestinal tract, particularly the stomach, for applications including motility assessment, vital-sign monitoring and the detection of slow-wave potentials 148 . A wearable patch with an electrode array for continuous and non-invasive EGG measurement 149 tested in 43 individuals differentiated EGG signals in patients with nausea and vomiting syndromes, indicating the device’s potential use for monitoring gastric dysmotility. Although the abundance of electrophysiological monitoring devices 150 complicates selection for clinical translation, there are evaluation frameworks 151 focusing on key translational challenges including convenience, accuracy, power requirements, biocompatibility and durability 150 . Devices that transfer data also face regulatory barriers related to privacy, security and storage, which are governed by the Health Insurance Portability and Accountability Act (HIPAA) and by the General Data Protection Regulation (GDPR). In the USA, HIPAA applies only if the device is provided by a covered entity, such as a healthcare provider, clearinghouse or business associate, and data are shared between the company and a covered entity 152 . Skin intolerance, which results in poor adherence, is another major translational barrier 153 – 155 . The device design must account for water repellency, stretchability and breathability. To improve comfort and compliance, electrophysiological sensors have been integrated into textiles 156 , 157 . For example, sensors were incorporated into a tailor-made long-sleeve shirt with custom sizing, which maximized comfort and usability 158 . Successful device commercialization also depends on shelf life and on the stability of key components, including membranes and coatings 159 . Recent advances in small biosensors for the detection and quantification of biomarkers in unconventional biofluids, such as saliva, exhaled breath condensate and urine, have expanded possibilities for non-invasive health monitoring 160 . These biosensors use four main types of receptors: enzymes, antibodies, nucleic acids and molecularly imprinted polymers (MIPs). Other biochemical recognition mechanisms include ion-selective membranes and the direct oxidation of electroactive targets. A range of transducers—electrochemical optical and piezoelectric—convert binding events into measurable electrical signals, enabling accurate quantification and real-time monitoring 160 . Enzyme-based biosensors catalyse specific reactions with target biomarkers, producing detectable signal changes. Electrochemical enzymatic biosensors are the most widely adopted wearable technology. Continuous glucose monitors (CGMs) sample interstitial glucose levels using microneedles 161 , and have been deployed at scale 162 , with over-the-counter approval for wearable CGMs such as Stelo by Dexcom in the USA 163 and Lingo by Abbott in Europe 164 . The widespread adoption of CGMs was driven by high diabetes prevalence—133 million Americans have diabetes or prediabetes 165 —and by the rising use of glucagon-like peptide 1 therapeutics such as Ozempic, which have increased the demand for glucose monitoring 163 , 166 . Beyond electrochemical transducers, enzymatic sensors have been integrated into optical and piezoelectric systems in patches, microneedles, mouthguards, earpieces and contact lenses for peripheral biochemical monitoring 167 . Wearable microneedle patches can track ketone fluctuations in interstitial fluids 168 ; notably, a CGM-integrated ketone sensor was introduced for ketoacidosis detection in diabetes management. Alternative biosensor form factors include thin films on teeth for the detection of H. pylori in enamel 169 . Exhaled and intestinal gas sensing is another emerging approach. Electronic noses use electrochemical sensors to detect redox reactions with target gases. Proof-of-concept devices have identified musty odours on infant skin (indicative of phenylketonuria) and acetone-like odours in breath (associated with diabetes mellitus) 13 , as well as markers for uraemia 14 . Continuous biosensing primarily relies on enzymatic reactions, ion-selective recognition and direct oxidation to detect key analytes such as electrolytes, glucose and lactate. However, these mechanisms are less effective at detecting trace biomarkers. For example, sweat-based biosensing is challenging because relevant biomarkers, such as hormones and proteins, are present at nanomolar concentrations or lower. Notable examples are a handheld device that used silver nanoparticles and Raman spectroscopy for the detection of creatine and cortisol in sweat with high specificity 170 , and a wearable skin-conforming patch inducing sweat through iontophoresis that used graphene electrodes with MIPs to monitor multiple metabolites, including all essential amino acids 171 ( Box 2 ). Moreover, sweat-based biomarker detection is hindered by limited understanding of how sweat concentrations relate to blood or interstitial fluid levels. Protein concentrations, for example, can be 1,000× lower in sweat than in the blood, and sweat biomarkers may lag behind blood levels 162 . Another challenge is developing high-surface-area electrodes with specificity for target biomarkers. It is also crucial to develop sensors with ultrasensitive bioaffinity for specific disease biomarkers in biofluids other than sweat. For example, a wearable device combined affinity receptors (antibodies, aptamers and MIPs) with diverse transducers to enable trace-level biomarker detection in biofluids 172 . Another device monitored wound biomarkers alongside temperature and pH 173 . Most bioaffinity sensors are currently restricted to single-use point-of-care applications. Continuous in situ monitoring is limited by the slow dissociation kinetics of high-affinity recognition elements. To address this, synthetic receptors such as aptamers and MIPs balancing binding kinetics and sensitivity have been developed. Although sensitive transducers can be combined to enhance detection limits, the intrinsic affinity of these receptors constrains performance to micromolar-level sensitivity 174 . An emerging approach is receptor regeneration through chemical or electrochemical methods 175 . Thermal regeneration, which uses miniaturized heating elements, offers practical sensor reusability, whereas electrochemical regeneration preserves receptor integrity through pre-programmed electrical stimuli. Although semi-continuous, these sense-and-regenerate cycles can be tuned to match biomarker fluctuations. Wearable, portable and handheld biosensors share general translational barriers, yet each type of sensor faces unique challenges. Electrochemical metal-oxide sensors require a ‘burn-in’ time to heat semiconductors, which affects the packaging and power design. Crosstalk between electrochemical and MIP biosensors can reduce specificity by causing unintended interactions. Moreover, devices collecting biofluids such as tears and saliva carry an infection risk, which can limit most to proof-of-concept demonstrations with uncertain translation pathways 176 , 177 . Biophysical monitoring involves measuring and analysing physical parameters related to bodily functions and movement. Temperature monitoring is widely used to provide insights into thermal regulation and overall health. Mobile devices such as smartphones can sense skin temperature by repurposing battery temperature sensors 178 , whereas wearable devices—including patches, textile-based sensors, rings and wristwatches—use resistive temperature detectors (RTDs). RTDs made from thin-film metallic materials such as gold and platinum on flexible substrates rely on the linearity between resistance and temperature 179 . Wearable thermistors, which are typically composed of semiconductor materials such as metal oxides or of conductive nanomaterial-filled polymers, exhibit nonlinear positive or negative temperature coefficients 180 , 181 . RTDs offer high accuracy and stability over a broad temperature range, whereas thermistors provide faster response times. Optical temperature sensors based on infrared thermography 182 and thermochromic liquid crystals also enable temperature sensing and visualization 183 , 184 . A limitation of these technologies is that they measure skin temperature rather than core body temperature. However, some wearables can measure heat flux to estimate core temperature 185 . Despite this constraint, continuous regional surface-temperature monitoring has shown clinical value, particularly for monitoring wound healing and diabetic foot ulcers 186 . Motion and tactile sensing is another application of biophysical monitoring. Accelerometers, gyroscopes and pressure-sensitive materials embedded in watches, clothing, shoes and flooring can detect and quantify movement, step counts and exertion. Consumer wearables already promote physical activity and help prevent conditions such as cardiovascular disease and metabolic syndrome 187 . Accelerometers in smartwatches and belts also facilitate fall detection and prediction 188 , 189 , particularly for geriatric care 190 . In home environments, sensors embedded in floorboards 191 or carpets 192 can detect falls and monitor gait and balance 193 . And smart home systems can integrate multiple sensing modalities for unobtrusive and continuous monitoring 194 . Pressure and strain sensors are widely used for thes sensing of human activity, for tactile sensing, and for vascular dynamics sensing 195 , 196 . Most operate through capacitive, piezoresistive or piezoelectric mechanisms. Capacitive pressure sensors detect strain-induced changes in dielectric thickness, with elastomeric materials such as polydimethylsiloxane, Ecoflex and polyurethane providing high mechanical sensitivity 197 . Piezoresistive sensors convert pressure or strain into resistance changes through conductive nanomaterial-filled elastomers (such as carbon nanotubes or silver nanowires) or conductive polymers 198 , 199 . Planar piezoresistive sensors have low sensitivity, which can be improved through structural modifications 200 , 201 . Piezoelectric sensors generate charge under stress owing to dipole reorientation. Wearable versions use flexible polymer films such as P(VDF-TrFE) 202 , ultrathin inorganic films such as PZT 203 or nanowires 204 . The optimal sensing modality depends on the pressure range of the target biophysical signal: intraocular pressure monitoring (<10 kPa) requires contact lens-based technologies 205 , whereas gait-monitoring pressure-sensitive shoes or carpets must function up to 2,000 kPa (ref. 206 ). The monitoring of vascular dynamics, typically around 100 kPa, is implemented in limb-conformal form factors such as wristbands 207 , electronic skins 208 and neck patches 203 . Key cardiovascular biomarkers—including heart rate, blood-oxygen saturation, blood pressure and respiration rate—can be continuously tracked using optical, pressure/strain or magnetoelastic generator sensors 209 , 210 . Wearable optical sensors use PPG to measure blood-volume changes in peripheral vessels 211 , a method that is used in commercial products such as the Oura Ring. Pressure sensors monitor blood pressure through subtle arterial-pressure fluctuations, whereas strain sensors on the neck or chest track respiratory rate by detecting respiration-induced movements, which is relevant to conditions such as cardiac events, pneumonia and dyspnea 212 . Ultrasound-based sensors enable deep-tissue monitoring. The sensors can capture central blood pressure 213 , deep-tissue haemodynamics and cardiac performance 214 , 215 and organ-volume changes 216 . Recent advances have integrated adhesive hydrogel systems into wearable ultrasound devices for the continuous imaging of organs 217 and breast tissue 218 . These systems use piezoelectric transducer arrays and bioadhesive hydrogel couplants that are durable, compliant and non-dehydrating and thus allow for ambulatory use 219 . As with electrophysiological monitoring, a distinction exists between commercially successful products (such as Garmin and Fitbit) and clinically validated devices (such as Corsano and Empatica). The success of Corsano and Empatica in overcoming translational barriers was enabled by first-mover advantage, validation 220 and subsequent FDA approval 221 rather than from superior technology. Several translational challenges hinder direct biophysical monitoring. Motion and gait sensors based on inertial-motion units suffer from sensor drift and noise, which requires regular calibration or sensor fusion for long-term accuracy. Pressure and strain sensors face a material trade-off: low mechanical modulus improves sensitivity but reduces durability due to inelastic stretching. Ultrasound sensors require high power for deep-tissue penetration, limiting portability. In fact, powering is a broader challenge across sensing modalities, as signal processing and filtering for the extraction of more biomarkers requires additional electrical components and increases energy demands. Moreover, because many biophysical signals can be monitored remotely, these technologies face added competition during clinical translation.

Remotely interfacing technologies enable continuous monitoring without direct patient contact, providing non-invasive, safe and convenient options for patients and clinicians. These technologies span several categories ( Fig. 2 ), differing in their underlying mechanisms and monitoring objectives. Informatics, which use either individual-level electronic health records (EHRs) or population-level wastewater data, show promise for tracking disease risk and progression. They gained initial traction during the COVID-19 pandemic, when they helped to direct resources and investments towards monitoring virus transmission (through waste water) and assessing the pandemic’s impact on population health (through EHRs) 16 – 21 . However, translational challenges remain, particularly for EHR-based approaches, as they often rely on machine-learning models that reduce transparency and may limit physician acceptance. Researchers have also explored video-based health monitoring. Previous research has demonstrated the feasibility of extracting heart rate from subtle variations in facial skin colour and extracting respiratory rate from slight body movements. However, these systems remain sensitive to lighting and environmental variations, limiting real-world application. Robotics have been deployed as mobile platforms for computer vision systems: in one example, a quadruped robot mounted with RGB and infrared cameras as well as a teleconferencing system was used for patient triage during the COVID-19 pandemic 22 . However, cost is a considerable translational barrier for scalable, real-world deployment. More recently, off-body sensors have been designed to leverage environmental radio waves for the monitoring of vital signs 23 , motor symptoms 24 , sleep stages 25 , 26 and medication adherence 27 . These devices passively collect physiological signals in the background while patients go about their daily activities, making them well suited for long-term monitoring. Some have already been deployed in pharmaceutical clinical trials to track disease progression 24 , 28 – 30 and to assess treatment outcomes, with early translational success 31 – 35 . Beyond pharmaceutical applications, off-body sensors hold potential for clinical care, particularly in chronic disease management. However, widespread adoption will require further evidence linking their measurements to actionable clinical decisions, along with insurance coverage. As with EHR-based informatics, devices’ reliance on machine learning may also affect physician confidence. Telehealth is another category of remotely interfacing technologies. By incorporating remote sensors (such as weight scales) and contact-based devices (such as blood-pressure cuffs), telehealth enables the remote exchange of clinical information through video interfaces. By providing clinicians with diagnostic and therapeutic data, telehealth enhances patient care through improved accessibility, convenience and cost-efficiency while overcoming geographical barriers. Wastewater analysis for continuous or periodic monitoring has seen substantial translational success in recent years. In the USA, the COVID-19 pandemic spurred interest in using waste water to forecast disease transmission 36 . Since 2020, daily samples have been collected in Massachusetts 16 , demonstrating the feasibility of long-term population monitoring. Wastewater networks have previously served as biosensors for disease prevalence and substance use, but most studies have been limited to individual treatment plants. Expanding biosensing to include intermittent sampling from both treatment plants and downstream sites, such as sewer manholes, allows for adjusting resolution and sampling catchment areas to target public health priorities. Ideal wastewater biomarkers include pathogens excreted renally or faecally. Xenobiotics can also be detected, provided that they undergo renal or hepatic metabolism. A recent innovation—detecting glucuronidated xenobiotic metabolites—has further improved precision by ensuring that measurements reflect human use 37 . During the pandemic, wastewater-based epidemiology proved to be a highly cost-effective tool for monitoring disease transmission, allowing policymakers to prepare health systems ahead of infection surges 17 , 18 , 38 . Multiple studies showed that SARS-CoV-2 RNA-fragment concentrations could predict hospital presentations several days in advance 39 , 40 . These insights may also inform public-health measures for high-risk populations and settings 41 , 42 . Beyond infectious-disease monitoring, wastewater analysis has been used to estimate substance-abuse prevalence 43 – 45 . By measuring glucuronidated drug metabolites in wastewater, these systems provide indicators of human consumption. Maps from geographical information systems overlaid on wastewater drug-concentration data enable longitudinal biosensing, and have revealed patterns of drug use, the emergence of new substances and the effectiveness of public-health interventions 46 , 47 . The translation of COVID-19 wastewater epidemiology was rapid; funding from the American Rescue Plan Act repurposed an existing foodborne pathogen-surveillance network to sequence wastewater samples 48 . However, this swift adoption was driven entirely by the urgency of the pandemic. To build on these lessons, the United States Centers for Disease Control and Prevention (CDC) established the National Wastewater Surveillance System to develop infrastructure for monitoring emerging pathogens 49 . Today, wastewater analytics in the USA continue to track seasonal pathogens, emerging diseases (such as monkeypox) and drug use. In Europe, the European Monitoring Centre for Drugs and Drug Addiction funds wastewater monitoring across major cities to track substance use trends 50 . Before the pandemic, several barriers limited the translation of wastewater epidemiology. Effective systems must not only identify robust analytical targets but also account for bacterial degradation. They also require high-resolution detection at large volumes to enable continuous monitoring. Moreover, monitoring sensitive substances or pathogens raises bioethical concerns that must be addressed before implementation. Data mining of search-engine queries is another approach to population-level monitoring. It uses correlations between infectious-disease incidence and topical internet-search terms. This method has successfully been used to predict the spread of Coxsackie virus 51 , influenza 52 and SARS-CoV-2 (ref. 53 ). During the COVID-19 pandemic, it was also applied to mental-health-related searches to track anxiety symptoms and mental-health trends 54 . Combining search-engine data mining with machine learning can further improve prediction accuracy 53 . At the individual level, analyses of EHRs enable continuous monitoring of disease incidence and risk. As health records are updated, machine-learning models can process large datasets to identify health trends. EHRs have been used to monitor disease incidence 55 , to assess osteoporotic fracture risk 19 , 21 and to classify patients according to diabetes type and subtype 20 , 56 . Other applications include using administrative claims, pharmacy records, service requests and laboratory results to track health conditions and diabetes risk 57 – 59 . In clinical settings, EHR-based models have supplemented inpatient monitoring to predict sepsis risk and mortality from it 60 . However, a key limitation of EHR analysis is the absence of social determinants of health, which are critical for translational informatics 61 . As a result, conclusions drawn from EHR data may be incomplete. The primary translational barrier to applying machine learning to EHRs is generalizability: a model may perform well on the data it was trained on but may not reliably extend to broader populations 62 . One challenge is racial and ethnic bias. The use of commercial medical-informatics algorithms—including those already in clinical use—can lead to systematic biases against minority populations 63 . Other models perform less accurately in under-resourced populations, potentially exacerbating healthcare disparities 64 . A related challenge is model calibration, which ensures that predicted probabilities correspond to true event frequencies. Recent studies have proposed methods to improve calibration, enhancing the interpretability of model outputs 65 , 66 . Another major challenge is EHR interoperability. Formats vary widely, and datasets contain thousands of potential predictors. In the USA, policies such as the Trusted Exchange Framework and Common Agreement have established technical networks for EHR data exchange, whereas the United States Core Data for Interoperability standardizes data elements. However, stringent regulations and limited stakeholder cooperation continue to hinder interoperability. Even more fundamental is the issue of EHR adoption. Despite the benefits of EHRs, financial misalignment remains a major barrier: physicians must cover the costs of outpatient EHR systems, yet the primary financial beneficiaries are EHR vendors 67 , 68 . As a result, many clinicians remain reluctant to use EHRs, limiting their potential for continuous monitoring. Telehealth enables healthcare delivery through digital technologies and telecommunications, allowing patients to consult clinicians virtually, and clinicians to access electronic information remotely. Telehealth offers convenience, and can integrate with wearable or portable devices to capture physiological data at a volume or frequency that is unfeasible for in-person visits. Continuous monitoring through telehealth requires a virtual care team to regularly track clinically relevant physiological signals or biomarkers. For example, Form Health, a telehealth company focused on obesity management, uses digital scales to transmit weight data from patients to an external care team for virtual clinical recommendations. Teladoc, which offers a broad range of continuous monitoring services, enables patients to track health metrics, such as blood glucose (for diabetes management), blood pressure (for hypertension) and weight. These data can inform virtual consultations or trigger digital interventions if abnormalities are detected. A key enabler of the success of Form Health and Teladoc is their eligibility for insurance coverage in the USA. However, although Teladoc may increase access to healthcare for patients by not requiring in-person providers 69 , its users do not disproportionately come from underserved communities 70 . Telehealth extends beyond monitoring and screening. The rapid collection and transmission of data can further enable pre-emptive and intensive care. For example, many people with hypertension fail to reach treatment goals that minimize cardiovascular risk 71 , 72 . Traditional in-person care limits treatment adjustments to in-person visits every 3–6 months, restricting opportunities for timely intervention. By contrast, telehealth allows for continuous blood-pressure monitoring, substantially increasing the frequency of treatment adjustments. Evidence of hypertension control from clinical trials supports the effectiveness of remote blood-pressure monitoring 73 – 75 . Another advantage of telehealth is its capacity for personalization and precision 76 . For example, continuous glucose monitoring allows for the transmission of real-time data to patients, family members and clinicians, and can therefore enable improvements in glycaemic control and diabetes-related quality-of-life measures 77 . Telehealth has been shown to enhance care quality and cost efficiency while maintaining high patient and clinician satisfaction 78 – 80 . Remote data transmission to external care teams is gaining acceptance, with regulatory standards expanding medical-billing permissions for telehealth services 81 . Adoption surged during the COVID-19 pandemic 82 , but substantial translational barriers remain. Privacy and security concerns arise from the inclusion of personal identifiers in digital health data 83 . Most telehealth services operate independently of health system networks, and therefore require further integration to facilitate seamless clinical use across providers. Although studies indicate that telehealth diagnoses can match in-person accuracy 84 , the digital medium has limitations—for example, wireless bandwidth affects the reliability of fine-motor assessments, potentially leading to misdiagnoses 85 and, although some healthcare systems use telehealth to reach remote or rural populations, others resist its adoption because reimbursement rates are lower than for in-person care 86 . Video data contain diagnostically relevant signals that can be measured remotely. The most well-known application is remote photoplethysmography (rPPG), which uses RGB cameras to detect changes in skin reflectance. These changes correlate with capillary dilation and constriction, enabling estimation of vital signs such as heart rate, respiratory rate, blood-oxygen saturation and blood pressure 87 . rPPG allows for continuous and contactless monitoring, and has been tested in emergency rooms 88 , 89 . However, rPPG requires the individual to remain within the camera’s field of view. To address this, robotic systems—such as ground robots 90 , drones 91 and miniature robotic blimps 92 —can be used to track the individual and perform rPPG. The COVID-19 pandemic accelerated research into robotic vital-sign monitoring, but deployment remains costly and impractical at scale 93 – 95 . Computer-vision methods can also estimate weight and body mass index by detecting anatomical landmarks, analysing body contours, extracting features with deep learning and integrating these elements for prediction 96 . Less-accurate approaches use only facial images, relying on facial geometry or deep-learning models 97 , 98 . Other applications of RGB cameras include monitoring the discomfort of infants through facial expressions 99 and nasal-versus-mouth breathing analysis for sleep disorders 100 . RGBD cameras (that is, cameras that provide both colour and depth), can precisely track physical behaviour, including gait, limb movement and fall detection 89 , 101 . RGB cameras can also be used for colorimetric assays, enabling unconventional applications such as the periodic monitoring of excreta 102 . Despite their accessibility, visible light-capturing methods face challenges in accuracy and reliability. As they rely on light reflectance, variations in lighting and motion artefacts can affect performance 103 . Although there are strategies to mitigate motion artefacts 104 , they remain condition dependent. Moreover, racial bias has been documented in computer-vision methods for continuous vital-sign monitoring, including state-of-the-art rPPG (ref. 105 ). Infrared cameras enable the direct monitoring of skin temperature 63 and have been explored for applications such as respiratory rate 106 , 107 , stress 108 and depression monitoring 109 , although with lower accuracy than clinical gold standards. Infrared monitoring presents several challenges—all objects emit and reflect thermal radiation, introducing noise and reducing image texture. Moreover, infrared cameras are sensitive to factors such as skin tone and distance, requiring calibration against a blackbody 110 . To address these limitations, a recent machine-learning-based approach decluttered heat signals to recover thermal-image texture, therefore enabling large-scale temperature monitoring and potentially improving infrared-based diagnostics 111 . Overall, although camera-based health monitoring is inexpensive, safe and accessible, its accuracy and clinical use remain limited. Moreover, few diagnostically useful signals beyond vital signs have been identified. For example, camera-based approaches, which rely on visible and near-infrared light, cannot be used to measure blood-glucose levels 112 . Radiofrequency (RF) signals can be used to monitor respiratory signals 113 , 114 , heart rate 115 , sleep stages and apnoea 25 , 26 , medication-administration errors 27 , and movement and behavioural symptoms 24 , 29 , 30 , 116 . These sensors detect wireless signal reflectance, which varies with body motion, without requiring wearable devices. In contrast to monitoring in the visual or infrared spectra, RF-based monitoring does not require line of sight, enabling easy integration into home and clinical settings. The passive and contactless nature of this approach also enables longitudinal monitoring without burdening patients. Early research focused on remote patient monitoring, whereas more recent work has leveraged this type of measurements to develop disease-specific digital biomarkers. For example, wireless monitoring of nocturnal respiration combined with a neural network trained on clinical datasets has been used to detect the occurrence and severity of Parkinson’s disease 28 ( Box 1 ). First-generation RF-based monitoring methods relied on signal processing, whereas more recent technologies use deep learning. A common challenge in RF-spectrum monitoring is entanglement: when multiple individuals are monitored in proximity, their signals interfere and overlap. To address this, RF-based sensors use waveform shaping through antenna arrays and frequency-modulated continuous wave technology to differentiate signals reflected from different individuals, which allows for continuous multi-individual monitoring 117 . These systems also apply independent component analysis to reduce residual interference and to accurately isolate individual signals, even when people are close together 23 . Certain RF-based technologies (such as Emerald); 23 , 26 , 118 , 119 have been validated (including third-party validation 120 ) against gold-standard methods for measuring breathing and heart rate, gait speed, sleep hypnograms and apnoea. These systems have also been deployed to monitor patients with Parkinson’s disease 24 , 28 , Alzheimer’s disease 29 , 30 , amyotrophic lateral sclerosis 33 , Crohn’s disease, atopic dermatitis 35 , endometriosis 121 , Rett syndrome 34 and facioscapulohumeral muscular dystrophy 122 . Some of these technologies have been commercialized and are used by the biopharmaceutical industry in clinical trials to assess disease progression and therapeutic outcomes 31 – 33 . Although RF-based monitoring has shown early success in drug development, widespread adoption will require regulatory approval. Further, although the technology holds promise in clinical care, particularly for risk assessment and the early detection of exacerbations in chronic diseases, integrating RF-based monitoring into healthcare systems poses considerable challenges. Moreover, while physiological signals captured by at-home monitoring systems can serve as non-invasive surrogate markers for health conditions, they generally have lower sensitivity and specificity than clinical gold-standard methods. The key advantage of RF-based systems is their ability to provide continuous and unobtrusive monitoring, which is advantageous for early detection and may help to prevent symptom escalation. Yet healthcare systems remain largely reactive, prioritizing treatment over prevention. Moreover, insurance coverage for such monitoring is limited, with few exceptions for specific conditions.

Owing to their proximity to tissues and organs, internally interfacing technologies enable highly sensitive and specific monitoring of biomarkers, analytes and diagnostic signals. However, their development and clinical translation remain challenging, as implantable and ingestible devices are often invasive, pose clinical risks and are subject to strict regulatory oversight. For intermittent or continuous monitoring to be clinically beneficial, device longevity must be maximized, yet it remains constrained by the functionalities and finite lifetimes of the sensing, powering and encapsulating materials. Currently, few implantable devices, and even fewer ingestible devices, are approved for continuous patient monitoring. In this section, we highlight technologies with translational potential and discuss common challenges. Implantable devices for continuous monitoring interface with tissues, organs or organ systems to collect physiologically relevant data. Most are implanted in the arms, chest or abdomen, although other sites are possible ( Fig. 4 ). FDA-approved neural implants record electrical signals for closed-loop seizure detection and control, vagal-nerve stimulation or deep brain stimulation 222 . Other implantable sensors include intraocular pressure monitors 223 and smart knee implants that detect infection and loosening 224 . Devices are typically inserted through minimally invasive procedures, such as catheterization 225 , endoscopy, laparoscopy or subcutaneous incision or injection 226 , 227 , yet may also be implanted during required surgical procedures 228 . Once inserted, they are secured using barbs 229 , sutures or clips 230 , adhesives, co-implanted structures (such as scaffolds, meshes or stents 231 ) or surgically defined pockets. Devices remain in situ for their operational lifetime before either persisting as non-functional implants, dissolving partially or fully 232 , 233 or being removed. Implantable continuous monitors have been explored for a range of applications. Devices can track haemodynamic parameters (such as heart rate 234 , and blood volume, pressure and flow rate 225 , 230 , 235 ) in patients with arrhythmias 234 or progressive heart failure 225 , as well as intraocular 223 or bladder 236 pressure in patients with glaucoma or urinary incontinence. The loop recorder is widely used to monitor cardiac rhythms and palpitations 234 . Implants have also been investigated for early anomaly detection requiring intervention, such as monitoring wound healing 233 and graft oxygenation 228 or detecting post-surgical complications 224 . Another application is overdosing detection: implants have been developed to monitor vital signs associated with opioid toxidromes 237 – 240 . Continuous monitoring is additionally integral to closed-loop systems, such as artificial pancreases that sense interstitial glucose to regulate insulin dosing 241 . There is also growing interest in applying devices containing transplanted cells or cell clusters for continuous monitoring. In such cell therapy-based approaches, the transplanted cells, inherently responsive to target bio-factors, mediate the continuous monitoring, so that the encapsulating device instead serves to support the via bility and functionality of the cells 242 , 243 . Implants may use active or passive electronics, or function without electronics 244 . Active electronic implants require a co-implanted or external power source, with energy generated, delivered or stored on-site 245 . Passive or non-electronic implants lack an integrated power source and transmit data only when externally stimulated by electrical 225 , ultrasonic 246 or optical 227 signals. Beyond loop recorders 247 and continuous glucose sensors 248 , 249 , few implantable monitors are approved for clinical use ( Box 3 ). Advances in this area will require highly selective sensors with rapid and reversible binding as well as the amplification of the transduced signals 174 . Long-term operation demands materials that ensure biocompatibility, minimize fouling and prevent humidity ingress 250 . Inadequate testing or regulatory oversight has direct consequences for patient safety 234 . Wireless data transmission also raises privacy and security concerns, necessitating low-power security solutions 251 . Moreover, patient acceptability and ethical considerations present translational barriers, particularly because sham-controlled trials of implantable devices are unethical 252 . The gastrointestinal tract is rich in diagnostic markers, and ingestible sensors have been developed to detect gastric inflammation, diabetes and vital signs 253 – 255 . However, rapid gastric transit (2–5 h in the stomach and 2–6 h in the small intestine) limits continuous monitoring unless the devices have prolonged gastric residency 256 . Two approaches for gastric retention involve expanding hydrogels 257 , 258 or mechanical arms 259 , 260 to prevent device passage through the pylorus. One ingestible sensor used magnetic hydrogels (fabricated with NdFeB microparticles) and an external wearable magnet to maintain gastric positioning. The device, which encapsulated biosensing Escherichia coli Nissle 1917, fluoresced in haem solutions and enabled the continuous detection of gastrointestinal bleeding 261 . Despite hydrogel corrosion in stomach acid, biocompatibility remained intact, and mice studies showed seven-day device retention with viable probiotic bacteria. Other proof-of-concept applications include sensing inflammatory biomarkers 262 , 263 , monitoring drug intake 264 , tracking internal temperature 265 and ensuring adherence to tuberculosis therapy 266 and to psychiatric medications 267 . A major limitation of ingestible continuous monitoring is power supply. To be ingestible, devices must conform to the osmotic-controlled-release oral-delivery-system form factor (15 mm in length and 9 mm in diameter), which has an estimated retention risk of 1 in 76 million 268 . Most devices, including capsule endoscopes, follow the larger #000 capsule form factor (26 mm in length and 10 mm in diameter), which carries a 1.4% retention risk 269 . For powered devices, two strategies have been explored: wireless power transfer and energy harvesting. Wireless power has been demonstrated in capsule endoscopes 270 and in electrochemical sensors 271 , 272 . Energy harvesting from gastric fluid has enabled metabolite biosensing 273 , although efficiency remains low, which limits clinical applicability 245 . Additional challenges include durability, biocompatibility and fibrosis 274 . As with implantable devices, the successful translation of ingestible sensors depends on multiple factors, particularly patient acceptability, power efficiency and safety 245 , 275 . Moving from proof-of-concept studies to clinical trials remains difficult, particularly in the USA, where FDA regulations evaluate devices on a case-by-case basis 276 . Nonetheless, the clinical adoption of ingestible devices such as the PillCam capsule endoscope has set a precedent for future translational efforts 277 , 278 .