Sepsis kills more people globally than any individual cancer. The World Health Organization puts the number at 11 million deaths per year, making it the leading cause of preventable hospital mortality. Decades of awareness campaigns, sepsis bundles, and early warning protocols have not solved the fundamental problem: sepsis is hard to catch early. Its initial presentation mimics a dozen other conditions, and by the time it becomes obvious, organ damage has already started.
Earlier identification saves lives. Every hour of delayed antibiotic treatment increases mortality risk. The part that remains unsettled is how hospitals should actually monitor patients to catch those early signals. Checking vital signs every four to eight hours on general wards leaves gaps measured in hours where deterioration goes unnoticed. Camera-based vital sign monitoring through remote photoplethysmography is entering this space as a way to provide continuous, passive surveillance that picks up the physiological shifts tied to early sepsis.
"Heart rate variability provides non-invasive and continuous monitoring for early detection of deterioration in septic patients by reflecting the autonomic nervous system's response to infection." — Derivation and validation study, Biomedical Signal Processing and Control (2024)
Sepsis and the monitoring gap
Sepsis develops when the body's response to an infection spirals into systemic organ dysfunction. The Sepsis-3 definition, established by Singer et al. in JAMA (2016), redefined sepsis around organ dysfunction rather than the earlier systemic inflammatory response syndrome (SIRS) criteria. The practical implication: by the time sepsis is clinically recognized, the body is already in trouble.
On general hospital wards, where most sepsis cases first manifest, vital signs are checked intermittently. A patient's heart rate, blood pressure, respiratory rate, and temperature are recorded during nursing rounds — typically every four to eight hours. Between those measurements, there is no physiological data being collected. A patient could begin deteriorating minutes after a check and not be reassessed for hours.
Churpek et al. at the University of Chicago found that vital sign abnormalities are often present 24 to 48 hours before a rapid response team activation or ICU transfer. The data is there, in the patient's physiology. Nobody is measuring it during those windows.
Early warning scoring systems like NEWS (National Early Warning Score) and qSOFA (quick Sequential Organ Failure Assessment) were built to turn vital sign data into risk scores. They work, but only when the data is available. A NEWS score calculated from measurements taken eight hours apart captures a snapshot, not a trajectory. Research from the University of Oxford published in Resuscitation found that the rate of change in vital signs carries more prognostic information than any single measurement, but calculating rate of change requires frequent or continuous data points.
How vital signs change before sepsis becomes apparent
Sepsis produces measurable changes across multiple vital sign parameters before it becomes clinically obvious. What changes first, and by how much, matters for any monitoring strategy.
| Vital sign parameter | Typical change in early sepsis | Detection window before clinical recognition | Measurement method via rPPG |
|---|---|---|---|
| Heart rate variability (HRV) | Decreased SDNN and RMSSD; reduced parasympathetic tone | 12-24 hours before clinical deterioration | Extracted from inter-beat intervals in rPPG signal |
| Heart rate | Elevated, often subtle initially (5-15 bpm increase from baseline) | 6-24 hours before sepsis diagnosis | Direct rPPG measurement from facial blood volume pulse |
| Respiratory rate | Elevated above 22 breaths/min (qSOFA criterion) | 6-12 hours before clinical recognition | Motion analysis of chest or facial micro-movements |
| Blood pressure | Systolic drop, often a later finding | 2-6 hours before clinical recognition | Experimental via pulse transit time (not yet clinical-grade) |
| SpO2 | Variable; may decrease with respiratory compromise | Variable | Ratio of red/infrared rPPG signal components |
Heart rate variability is one of the earliest markers of impending sepsis in the research literature. Ahmad et al., in PLOS ONE, showed that HRV depression (reduced complexity, decreased high-frequency power) occurs before conventional vital signs cross abnormal thresholds. The autonomic nervous system responds to early infection by shifting toward sympathetic dominance, and this shift is measurable through beat-to-beat heart rate analysis hours before temperature spikes or blood pressure drops.
Respiratory rate is the other consistently early marker. Clinicians have called it the "neglected vital sign" because it is the most frequently skipped during manual observations — nurses often estimate rather than count breaths. Yet elevated respiratory rate (above 22 breaths per minute) is included as a qSOFA criterion specifically because of its predictive value for sepsis. Continuous camera-based monitoring captures respiratory rate passively through motion analysis, eliminating the estimation problem entirely.
Camera-based monitoring as continuous surveillance
Remote photoplethysmography extracts physiological signals from video captured by standard cameras. It picks up changes in skin color caused by blood volume fluctuations with each heartbeat, changes too small for the eye to see but measurable computationally.
For sepsis detection, no single measurement matters as much as having continuous data. A camera positioned near a patient's bed captures heart rate, HRV metrics, and respiratory rate around the clock without any sensor attached to the patient. This generates a continuous physiological record rather than isolated data points.
Jorge et al. at the University of Oxford published a study in npj Digital Medicine (2022) evaluating video-based monitoring of 15 postoperative ICU patients for an average of 16 hours each. They demonstrated that heart rate and respiratory rate could be estimated with mean absolute errors of 2.5 bpm and 2.4 breaths per minute respectively. The system captured data regardless of patient position — lying in bed, sitting, moving, or receiving routine care.
Wang and Huang (2024), in a study published in IETE Journal of Research, investigated using existing CCTV camera infrastructure in hospital ICUs for contactless physiological monitoring. Their approach repurposed cameras already installed for patient safety surveillance, adding rPPG processing to extract vital signs from the same video feed. Using cameras that already exist for patient safety surveillance to also extract vital signs addresses the implementation cost problem that has slowed adoption of new monitoring technologies.
A double-center clinical study in IEEE Journal of Biomedical and Health Informatics tested a camera-based platform for multi-parameter vital sign monitoring in hospital ICUs. The system achieved mean absolute errors of 1.89 bpm for heart rate and 19.04 ms for SDNN, a key HRV metric. That second number matters here: it means camera-based systems can capture the variability patterns relevant to sepsis detection, not just raw heart rate.
HRV as the early sepsis biomarker
HRV has a strong evidence base as an early marker of sepsis. The physiology is straightforward: healthy autonomic nervous systems maintain high variability in heart rhythm, reflecting adaptive responses to internal and external stimuli. When systemic infection takes hold, the autonomic response shifts, and variability decreases — often before other clinical signs appear.
A 2024 study published in Biomedical Signal Processing and Control developed machine learning models using HRV features for early sepsis prediction. The researchers derived and validated models specifically using HRV parameters extracted from continuous monitoring data, finding that HRV-based algorithms could identify patients progressing toward sepsis hours before conventional screening tools flagged them.
Buchan et al. at the University of Virginia developed HeRO (Heart Rate Observation), a system that monitors HRV patterns in neonatal ICU patients to predict late-onset sepsis. Their work, published across multiple journals including Pediatric Research, demonstrated that reduced heart rate variability and transient decelerations — detectable only through continuous monitoring — preceded clinical sepsis diagnosis by up to 24 hours in neonates. The HeRO monitor, now commercially available, reduced mortality by up to 22% in one randomized controlled trial. While HeRO uses contact-based ECG, the HRV parameters it relies on are the same parameters that rPPG systems can extract contactlessly.
Fairchild et al. (2023), extending this work to adult populations, published findings showing that heart rate characteristics including reduced variability and abnormal decelerations are predictive of deterioration in adult ICU patients. The signal processing methods translate directly to rPPG-derived heart rate data.
AI-enhanced early warning from continuous vital sign streams
Continuous vital sign data paired with machine learning produces a detection system better than either piece on its own. Manual early warning scores like NEWS require a human to collect data, calculate a score, and decide whether to escalate. Automated systems process continuous streams of physiological data and apply trained algorithms to detect subtle patterns that precede deterioration.
Park et al. (2025), publishing in JMIR, evaluated a remote early warning system (R-EWS) framework that analyzed continuous vital sign data from ward patients. Their system processed real-time heart rate, respiratory rate, and oxygen saturation streams to calculate rolling deterioration scores, identifying patients at risk of sepsis or other critical events hours before traditional intermittent monitoring detected a problem. The study demonstrated that moving from four-hourly spot checks to continuous data analysis fundamentally changed the timing of clinical alerts.
A systematic study published in JAMIA Open (2022) evaluated 240 combinations of four vital signs — heart rate, respiratory rate, blood pressure, and temperature — for sepsis prediction performance. The analysis found that models combining heart rate and respiratory rate together outperformed models using any single parameter, and that adding blood pressure data provided marginal additional benefit. This finding supports the use of camera-based monitoring, which captures heart rate and respiratory rate continuously, as a foundation for automated sepsis screening even when blood pressure measurement is not available through the camera.
The CDC's Hospital Sepsis Program Core Elements, updated through 2025, emphasize that effective sepsis programs require systematic monitoring and performance evaluation across the organization. While the guidelines do not yet specifically address contactless monitoring technology, they establish the framework for institutional monitoring protocols that continuous camera-based systems could support.
The road from research to bedside deployment
Camera-based sepsis detection faces several technical and implementation hurdles before widespread clinical adoption. Low-light performance remains a challenge — hospital rooms are often dimmed at night, precisely when monitoring is most needed and staffing is thinnest. Patient movement, bedding covering the face, and variable ambient lighting all affect signal quality. Jorge et al. (2022) reported that after removing privacy periods and low-quality recordings, their system captured usable vital sign data for approximately 44-51% of total recorded time, meaning roughly half of the monitoring period yielded no data.
Privacy is another consideration. Continuous video monitoring in hospital rooms raises ethical questions around patient consent, data storage, and secondary use of video recordings. The Frontiers in Digital Health review by Di Lernia et al. (2024) specifically flagged privacy and data governance as barriers to clinical rPPG adoption, noting that regulatory frameworks have not kept pace with the technology.
Skin tone bias in rPPG accuracy is an active area of research. Dasari et al. at Carnegie Mellon University, publishing in npj Digital Medicine (2021), evaluated biases in remote photoplethysmography methods and found measurable accuracy differences across skin tones. Addressing these disparities is essential for any system deployed in diverse hospital populations, particularly since sepsis disproportionately affects underserved communities.
Still, improved deep learning algorithms, cheaper camera hardware, and growing demand from hospitals for continuous monitoring are collectively pushing camera-based vital sign technology closer to clinical readiness. Debnath and Kim (2025), in a comprehensive review of 145 articles published in BioMedical Engineering OnLine, noted that deep learning methods have consistently outperformed traditional signal processing for rPPG accuracy, with the gap widening as larger and more diverse training datasets become available.
Where sepsis surveillance is heading
The direction is toward multi-modal continuous monitoring: combining camera-derived vital signs with other ambient sensors, electronic health record data, and lab results to build composite risk scores that update in real time. Several clinical trials are testing this approach now.
The Nature Pediatrics Research group published work in 2025 on next-generation non-contact vital sign monitoring for neonatal intensive care, exploring how RGB-D cameras can simultaneously capture heart rate, respiratory rate, and body position in critically ill newborns — a population where sepsis is both common and difficult to detect early.
A clinical trial registered at ClinicalTrials.gov (NCT07307521) is currently evaluating AI-assisted video monitoring for predicting clinical deterioration in ICU patients. The study aims to determine whether continuous video analysis can identify patients at risk of rapid deterioration, including sepsis onset, earlier than standard monitoring protocols.
Circadify has developed camera-based vital sign measurement technology that captures heart rate, HRV, respiratory rate, and SpO2 from a standard device camera. What the technology measures and what clinicians need for early sepsis identification are converging, and the evidence base keeps growing.
Frequently Asked Questions
Can camera-based monitoring detect sepsis?
Camera-based monitoring does not diagnose sepsis directly. It continuously measures vital signs like heart rate, heart rate variability, and respiratory rate — parameters that change measurably in the hours before sepsis becomes clinically apparent. By tracking these markers continuously rather than intermittently, camera-based systems can flag physiological changes earlier than traditional spot-check monitoring.
What vital signs are most predictive of sepsis onset?
Research identifies heart rate variability as one of the earliest physiological markers of sepsis, often decreasing before other vital signs change. Respiratory rate elevation is another strong early indicator. The combination of heart rate, HRV, respiratory rate, and blood pressure changes together provides stronger predictive performance than any single parameter alone.
How does contactless monitoring improve sepsis detection compared to standard care?
Standard care on general wards involves manual vital sign checks every four to eight hours. Sepsis-related physiological changes can develop and progress during those gaps. Continuous contactless monitoring eliminates blind spots by tracking vital signs passively around the clock, giving clinicians access to trend data and early alerts.
Is rPPG technology accurate enough for clinical sepsis monitoring?
Clinical studies demonstrate that rPPG systems can measure heart rate with mean absolute errors below 2.5 bpm and respiratory rate within 2.4 breaths per minute compared to contact-based references. While these systems are positioned as supplementary monitoring rather than replacements for standard devices, their accuracy is sufficient for trend detection and early warning applications.