DiscoveringaNewLayerofVitalsforHumansintheAgeofAIandAgents

Abstract

Artificial intelligence is increasingly capable of generating health guidance, yet its usefulness is limited by the structure of real-world health data. Most day-to-day health decisions occur in a “middle zone” between (i) traditional physiological vital signs that are objective but nonspecific (heart rate, blood pressure, oxygen saturation, temperature) and (ii) medical states or conditions that are meaningful but poorly measurable at home (e.g., dehydration, stress/anxiety, “not feeling well,” inflammation, erectile dysfunction). This gap forces both humans and AI systems to rely on qualitative narratives that are noisy, inconsistent, and difficult to operationalize. Here we propose Microvascular Vitals: a new class of measurable, repeatable, baseline-personalized physiological indicators derived from optical bio signals (wearable photoplethysmography, PPG; and camera-based remote PPG, rPPG). Microvascular Vitals are designed to map directly to common everyday medical situations, expressed as trajectories with uncertainty and quality gating. We argue that microvascular dynamics constitute an underutilized physiological interface layer that can enable safer AI-driven monitoring and guidance—earlier and more meaningfully than threshold-based macrovascular vital signs alone.

1. The simplest question AI cannot answer well

A foundational health question— “Am I dehydrated?”—illustrates the core limitation of current macrovascular based vitals and clinical questions driven health advice. In daily life, people answer such questions using subjective proxies (“thirsty,” “tired,” “headache,” “dry mouth”) and generic heuristics. These descriptors are affected by context, expectation, reporting bias, and inter-individual variability. As a result, an healthcare professions and AI systems operating primarily on narrative inputs is forced to infer physiological state indirectly, which limits both accuracy and actionability.

Health management—especially when delegated to AI agents—requires more than language. It requires measurable physiological state with known uncertainty and repeatable interpretation over time.

2. The “missing layer” between vital signs and medical reality

Today’s health data ecosystem is dominated by two data types that frequently fail to connect.

Traditional vitals are objective and standardized but often too generic to characterize early or ambiguous states. Heart rate, blood pressure, and oxygen saturation can shift for many reasons, and their clinical meaning often depends on context not captured by the measurement itself.

Medical situations and conditions—such as dehydration, stress/anxiety, “not feeling well,” inflammation, and erectile dysfunction—are meaningful to patients and clinicians, but are often expressed as labels or symptom narratives rather than quantified physiological states measurable at home.

The consequence is a practical gap: there is no widely adopted, standardized middle layer that translates physiology into state-level measures linked to common medical situations, particularly outside clinical environments.

3. Five everyday medical situations that define the problem space

We focus on five high-frequency states that lie at the boundary of wellness to proper medicine, where early detection and longitudinal monitoring matter but measurement is currently weak.

Dehydration / low effective circulating volume Subjective phenotype: thirst, fatigue, dizziness, reduced performance. Why traditional vitals fail: HR/BP changes are nonspecific and often late; single readings ignore recovery kinetics after rehydration. Target representation: baseline-normalized perfusion/vascular tone signatures with recovery curves after fluid intake, reported with confidence. Stress or anxiety (sympathetic dominance) Subjective phenotype: “wired,” palpitations, restlessness, poor sleep, panic-like episodes. Why traditional vitals fail: elevated HR can reflect many causes; BP is episodic and context-dependent. Target representation: autonomic–vascular coupling metrics, capturing sympathetic dominance trajectories and response to interventions. “Not feeling well” (pre-symptom / ambiguous malaise) Subjective phenotype: off-baseline fatigue, fogginess, “coming down with something.” Why traditional vitals fail: fever or oxygen changes may be absent; thresholds are designed for overt illness. Target representation: deviation-from-baseline signatures that detect physiological drift prior to clear symptom clusters. Inflammation / vascular stress Subjective phenotype: heaviness, soreness, skin “flare,” post-exertional malaise, subclinical inflammatory states. Why traditional vitals fail: often normal in chronic or low-grade inflammation; single-point measures miss oscillatory or microdynamic behavior. Target representation: microvascular variability and reactivity features separating inflammatory dynamics from superficial appearance changes. Erectile dysfunction (vascular/autonomic function state) Subjective phenotype: reduced erectile quality and reliability; strong sensitivity to sleep, stress, cardiometabolic health. Why traditional vitals fail: resting HR/BP poorly reflect functional vascular responsiveness. Target representation: longitudinal vascular responsiveness and autonomic readiness metrics linked to function, not just static vitals.

Across these domains, the scientific requirement is consistent: state inference should be expressed as quantitative scores (or vectors) grounded in physiology, personalized to baseline, and evaluated as trajectories with uncertainty and quality gating.

4. Why current vitals are “too late” and “too vague”

Traditional vitals remain indispensable, particularly in acute care. However, their limitations become clear when applied to early, personalized, at-home decision-making:

Nonspecificity: The same vital sign pattern can correspond to multiple physiological states (e.g., elevated HR from dehydration, anxiety, fever, caffeine, pain, sleep loss). Population ranges over individual baselines: “Normal” by population standards can obscure meaningful deviation for an individual. Snapshot bias: Clinic-style single measurements underrepresent dynamics; for many states, trajectory is more informative than absolute value. Late-stage detection: Threshold-based logic is optimized for overt deterioration, not subtle physiologic drift.

Thus, while traditional vitals function well as safety alarms, they are poorly suited to characterize the everyday physiological states that drive most health decisions.

5. AI agents require quantitative trajectories, not anecdotes

AI and autonomous agents change the operating model of health management: instead of episodic clinical visits, continuous or frequent micro-decisions become possible (hydration prompts, stress regulation, recovery optimization, symptom monitoring, escalation guidance). This model is only safe and effective if the underlying data are:

Quantitative and repeatable Longitudinal (time-series) Baseline-personalized Quality-gated (signal integrity assessment; rejection of unreliable data) Uncertainty-aware (confidence intervals and probability-of-state rather than binary labels)

In effect, AI agents require physiological “state estimation” analogous to navigation systems: a reliable signal with uncertainty and drift correction, not narrative descriptions alone.

6. Proposal: Microvascular Vitals as a new physiology layer

We propose Microvascular Vitals:

A class of condition-linked physiological indicators derived from microvascular dynamics, measurable at home, repeatable across time, personalized to baseline, and expressed with uncertainty and quality gating.

Microvascular Vitals are defined not by recreating traditional vitals, but by mapping optical physiology to medical situations using target-specific features and longitudinal modeling.

Core properties

At-home measurable (phone camera rPPG; wearable PPG) Repeatable and protocolizable (standardized capture conditions) Personalized (baseline normalization and adaptive calibration) Condition-targeted (distinct representations for dehydration vs stress vs inflammation) Actionable outputs (score + confidence + recommended next steps and escalation rules)

7. Why microvasculature is the right interface layer

The microvasculature is the functional interface for tissue perfusion, oxygen/nutrient exchange, thermoregulation, and inflammatory signaling. Many everyday physiological states may manifest early as microvascular changes, including:

Perfusion redistribution and capillary recruitment patterns Vascular tone shifts and reactivity changes Autonomic modulation signatures reflected in pulsatile dynamics Inflammatory microdynamics affecting variability and responsiveness

This motivates a central hypothesis: many “I feel off” moments are microvascular moments first, preceding overt changes in conventional vital signs.

8. Enabling measurement: wearable PPG and smartphone rPPG

Two sensing modalities can operationalize microvascular measurement at scale:

Wearable PPG: continuous, passive data suited for baseline establishment and trend monitoring.

Smartphone rPPG: ubiquitous, low-friction “check-in” measurements that can be standardized and scaled without dedicated hardware.

The key opportunity is not measuring heart rate again; it is extracting condition-relevant microvascular features and tracking their trajectories to infer state.

9. Why the field has stalled: the “pulse-ox framing”

Despite substantial progress in PPG/rPPG, the dominant research framing has emphasized reconstruction of traditional vital signs (HR, SpO₂, respiratory rate). This emphasis is scientifically valuable, but it leaves critical gaps underexplored:

Condition-targeted signatures and feature families Baseline personalization and adaptive calibration Trajectory modeling (response curves, recovery kinetics) Quality gating and uncertainty quantification Validation anchored to everyday medical situations and longitudinal outcomes

In short, optical biosensing has often been treated as a cheaper pulse oximeter rather than a route to a new physiology layer designed for state inference and agentic health management.

10. Safety, trust, and scope: what this is not

Microvascular Vitals should not be positioned as a diagnostic engine. Their intended role is an early-state guidance and monitoring layer that supports:

earlier detection of deviation from baseline personalized intervention tracking conservative escalation pathways when risk thresholds are met or uncertainty is high

For safe deployment, Microvascular Vitals must include (at minimum): quality gating, explicit uncertainty reporting, conservative escalation rules, and transparent user controls. Privacy-preserving governance and user ownership of data are prerequisites for adoption.

Outlook

As AI and agents become embedded in daily life, the limiting factor in health will increasingly be the availability of structured, quantitative physiology that is meaningful at the level where humans actually live: hydration, stress, malaise, inflammation, and functional vascular health. Microvascular Vitals offer a plausible path to fill the missing middle layer—bridging generic vital signs and clinically meaningful states through at-home, repeatable, baseline-personalized measurement. Realizing this vision will require a shift in research priorities from reconstructing classic vitals toward validating condition-targeted microvascular state indicators, expressed as trajectories with uncertainty, and benchmarked against outcomes that matter to patients and clinicians.

This perspective is intended as a conceptual framework for a broader research program. Detailed domain-specific validation should be developed separately for individual targets such as hydration, autonomic stress, inflammatory states, and functional vascular health.