The PPG, Blood Pressure Paradox: Why Hilo and Oura May See the Pulse but Not the Pressure:
With the recent wave of wearable devices claiming some form of blood-pressure tracking, from Oura’s nighttime BP dipping to Whoop, Galaxy Watch, and now even the Pixel Watch 4/5, it seems like a good time for a reality check: what are these devices actually measuring, what do their BP-related numbers really mean, and how accurate are they?. To that end, first we will discuss the
theoretical background of optical BP and then mention a more "dedicated" unit, the
Hilo Core wristband vs the Oura ring, which I’ve used for the past few weeks.
What does a PPG blood-pressure device actually measure?
Photoplethysmography (PPG) has become one of the most widely
used sensing technologies in consumer wearables. By lighting the skin with an
LED and measuring changes in reflected or transmitted light, PPG detects the
pulsatile changes in blood volume that accompany each heartbeat. This signal is
well suited to estimating heart rate and, under favorable conditions,
beat-to-beat timing for heart-rate variability. More recently, however, PPG has
been extended to a much more ambitious target: blood pressure estimation
without an inflatable cuff. This development has considerable appeal.
Conventional ambulatory blood-pressure monitoring requires repeated cuff
inflation, which can be uncomfortable, disruptive during sleep, and limited to
intermittent measurements. I can attest to this personally. Several months ago
I took part in a Pixel watch ambulatory BP study and wore a cuff for 24 hours.
Every time it inflated, it woke me from sleep and I ended up just removing it
before it drove me nuts. Therefore, a small wristband, watch, or ring capable
of estimating blood pressure passively could potentially provide many
measurements across days or weeks with little burden on the wearer. Devices
such as Aktiia/Hilo and newer PPG-based features from Oura illustrate how
rapidly this field is moving.
The important physiological distinction is that PPG does not
measure pressure directly. A conventional cuff estimates arterial pressure by
mechanically compressing an artery and observing the pressure at which arterial
flow or oscillations change. A PPG sensor instead records an optical waveform
generated by pulsatile changes in peripheral blood volume. Blood pressure
influences that waveform, but it is only one of many factors that do so.
The shape of a PPG pulse contains substantial cardiovascular
information. Features such as the systolic upstroke, systolic peak, dicrotic
notch, reflected or diastolic wave, pulse width, and timing between waveform
landmarks can change as arterial pressure changes. These relationships provide
the physiological foundation for cuffless BP algorithms. Modern systems may
combine many such waveform characteristics, along with heart rate and other
sensor information, using statistical or machine-learning models to infer blood
pressure or blood-pressure-related patterns.
The difficulty is that these waveform characteristics are not
uniquely determined by blood pressure. Peripheral vascular tone, arterial
stiffness, pulse-wave velocity, wave reflection, stroke volume, heart rate,
skin temperature, local perfusion, body position, autonomic activity, and even
the pressure of the sensor against the skin can modify the PPG waveform. Thus,
a change in pulse shape does not necessarily mean that blood pressure changed
by a corresponding amount.
Conceptually, the PPG signal might be represented as:
PPG waveform = a function of: (blood pressure, vascular
tone, arterial stiffness, wave reflection, stroke volume, heart rate,
temperature, perfusion, sensor conditions, …)
A cuffless BP algorithm attempts to solve the inverse
problem:
blood pressure ≈ summated function⁻¹(PPG waveform)
That inverse relationship is considerably more difficult. If
several physiological variables can produce similar changes in the optical
pulse, there may be no unique blood-pressure value associated with a particular
waveform. The problem becomes especially important when the physiological state
changes substantially, for example during sleep, exercise, medication
treatment, heat exposure, or sympathetic activation. This distinction helps
explain an apparent paradox in the cuffless-BP literature. A PPG device can
demonstrate superb agreement with a conventional cuff during controlled
resting measurements yet perform much less well when asked to follow blood
pressure changes over time. Calibration can place the estimated pressure close
to the true value at baseline, but good baseline agreement does not necessarily
demonstrate that the algorithm has the correct "dynamic response" or "gain" when blood pressure subsequently rises or falls.
Nocturnal blood-pressure dipping provides a particularly
useful example. During sleep, blood pressure normally declines, but so do
sympathetic activity, peripheral vascular resistance, heart rate, skin
temperature, and other determinants of the peripheral pulse waveform. A PPG
algorithm therefore must distinguish waveform changes caused specifically by
falling arterial pressure from waveform changes caused by the broader
transition from wakefulness to sleep.
This issue is not confined to devices that report systolic
and diastolic pressure in mmHg. It also applies to algorithms that use PPG to
classify people as normal dippers or non-dippers. Such systems may avoid the
difficult task of reconstructing absolute blood pressure, but they still depend
on a relationship between nocturnal pulse morphology and the underlying
blood-pressure response.
The central question for PPG-based blood-pressure technology
is therefore not whether the optical pulse contains information related to
blood pressure, it clearly does. The more difficult question is whether an
algorithm can reliably separate the effect of blood pressure from the many
other physiological processes that alter the PPG waveform, particularly when
the cardiovascular state changes over time.
Hardware/Software Considerations:
Sampling requirements differ
substantially between ECG and PPG because the important information in the two
signals is different. In an ECG, heart-rate variability depends primarily on
accurately locating the R peak, a sharp and readily identifiable electrical
fiducial point. At a sampling rate of 250 Hz, samples are separated by 4 ms; at
500 Hz by 2 ms; and at 1000 Hz by 1 ms. Studies using ECG signals down-sampled
from 1000 Hz have generally found excellent agreement in HRV measures at
250–500 Hz, while lower sampling rates progressively increase timing
quantization error. Interpolation around the R peak can improve timing
considerably and permit useful HRV analysis even at lower sampling rates.
PPG presents a different problem. Beat timing can be
obtained from a single reproducible point on the pulse, such as the foot,
maximum upslope, or systolic peak and reasonable pulse-rate variability can
therefore be obtained at surprisingly low sampling rates. One study found that,
with an interpolated upslope fiducial point, PPG could be reduced to 50 Hz
without important changes in conventional variability indices. This does not,
however, mean that 50-Hz PPG adequately reproduces pulse morphology.
For blood-pressure estimation, the algorithm may depend on
the shape of the entire pulse: the systolic upstroke, systolic peak, dicrotic
notch, reflected wave, and the timing and amplitude relationships among these
features. At 50 Hz there is only one sample every 20 ms, and at 100 Hz one
every 10 ms. Thus, a relatively small feature such as the dicrotic notch may be
represented by only a few samples, and its apparent position can shift
substantially depending on where it falls relative to the sampling grid.
Interpolation can improve estimates of fiducial timing, but it cannot restore
morphological information that was never acquired, particularly in the presence
of noise or waveform distortion. Work specifically examining PPG fiducial
localization confirms that sampling rate, filtering, interpolation, and signal
quality all influence the localization of waveform landmarks used for BP
estimation.
Consequently, 100 Hz is quite reasonable for gross PPG
morphology and is the documented sampling rate of the validated Hilo system,
but higher rates such as 200–250 Hz provide substantially denser representation
of the pulse when precise timing of the dicrotic notch, reflected wave, or
other morphological features is important. This is different from ECG R-wave
detection, where a single sharp fiducial point can often be localized very
accurately even from a moderately sampled signal. A high sampling rate
therefore does not make PPG-derived blood pressure intrinsically accurate, but
inadequate sampling can add yet another source of uncertainty to an already
indirect physiological measurement. That said, Oura's historically reported
250-Hz nocturnal PPG could provide a prettier and more precisely sampled pulse
waveform than Hilo’s 100-Hz signal, yet it still cannot solve the fundamental
BP problem. Sampling improves measurement of the PPG waveform; it does not make
PPG morphology uniquely determined by arterial pressure.
Calibration: Why Good Agreement Does Not Necessarily Mean
Good BP Tracking
Most PPG-based blood-pressure systems face an immediate
problem: the optical waveform has no intrinsic scale in mmHg. A PPG sensor
measures changes in reflected light associated with pulsatile blood volume,
producing a waveform in arbitrary optical units. Some method is therefore
needed to connect the optical signal to an actual arterial pressure. The
simplest solution is calibration against a conventional cuff. This distinction
is particularly clear in the published Hilo/Aktiia validation studies. The
bracelet first analyzed the wrist PPG waveform to produce uncalibrated systolic
and diastolic BP estimates. During initialization, the first paired cuff
measurements were then used to calculate a subject-specific SBP offset and DBP
offset, converting the optical estimates into mmHg (PMID: 33675592). Conceptually,
this can be represented as:
Reported BP = PPG-derived BP estimate + individual
calibration offset
Suppose the optical algorithm estimates an individual's
resting systolic pressure as 112 arbitrary BP units, while the calibration cuff
measures 124 mmHg. An offset of +12 mmHg can make the bracelet report:
112 + 12 = 124 mmHg
The device now agrees perfectly with the cuff at
calibration. But that agreement tells us surprisingly little about what will
happen when the person's BP changes. Imagine that several hours later the true
systolic pressure falls by 20 mmHg:
Cuff: 124 → 104 mmHg
If the PPG morphology changes by an amount that the optical
algorithm interprets as only a 6-mmHg fall, the device will calculate:
112 → 106 optical units
and, after applying the same +12-mmHg calibration:
124 → 118 mmHg
So…. The initial value is exactly correct, but the
subsequent change is badly underestimated. The problem is not the calibration
offset. The problem is that the dynamic sensitivity of the PPG model to
changing BP is too small. An offset calibration primarily solves the problem: Where
should the BP estimate sit on the mmHg scale. It does not necessarily solve the
problem: How far should the BP estimate move when actual pressure changes?
If the correct gain is 1.0 but the PPG algorithm effectively
behaves as though it were 0.3, a true 20-mmHg change may appear as only about a
6-mmHg change. No adjustment of the baseline can correct that. This is
especially relevant because Aktiia's later 24-hour study describes these two
functions explicitly as separate. The investigators state that initialization
generates an offset, establishing the absolute reference around which
subsequent measurements vary, whereas the device's ability to track BP trends
is determined independently by the optical algorithm. Reinitialization changes
the baseline but does not change the trend-tracking mechanism (PMID: 39927495).
Population accuracy is not the same as within-person
tracking
This also illustrates an important limitation of
conventional device-validation statistics. Suppose 100 people undergo seated
testing. Each device is individually calibrated near the subject's resting BP.
Across the population there may be people with pressures ranging from 100 to
180 mmHg. If calibration places each person's optical estimate near his or her
cuff pressure, the device can show:
- · very small mean bias,
- good correlation across subjects,
- acceptable average error,
- an impressive Bland–Altman plot.
The original Aktiia validation study, for example, reported
average differences of approximately 0.46 mmHg for SBP and 0.39 mmHg for DBP
during standardized seated measurements after subject-specific calibration (PMID:
33675592). Those are encouraging results, but they primarily demonstrate that
the system can provide reasonable calibrated absolute BP estimates under the
tested conditions. They do not automatically demonstrate that, within a
particular individual:
ΔBP from PPG ≈ ΔBP measured by the reference method.
This is a fundamentally different validation question. A
device could therefore show excellent cross-sectional agreement because
subjects with high BP generally receive high estimates and subjects with low BP
receive low estimates, while still substantially underestimating the magnitude
of BP changes occurring within each person.
Why estimating an individual slope is much harder
In principle, calibration could determine both the intercept
and the slope of the true BP vs PPG waveform value. But this requires reference
measurements obtained across a meaningful range of BP. If cuff calibration
measurements are: 121, 123, 120, 124 mmHg they provide excellent information
about the person's resting baseline but very little information about how the
PPG should behave at: 100 or 150 mmHg.
To estimate an individual PPG-to-BP slope reliably, paired
PPG and reference measurements would ideally span substantially different
pressure states. That might require
controlled perturbations such as:
- · changes in posture,
- exercise or recovery,
- pharmacological BP changes,
- cold or heat exposure,
- sustained handgrip,
- lower-body pressure manipulation,
- or naturally occurring day-night changes.
But this introduces the next problem: these interventions
alter vascular physiology as well as BP. The relationship between PPG
morphology and pressure may itself change between conditions. Consequently,
even a personalized linear slope may not be sufficient. The true relationship
may be state dependent: BP = function of (PPG, vascular tone, temperature,
posture, arterial stiffness, autonomic state, …) rather than a single fixed
calibration equation.
This concern is supported by the independent evaluation of
Aktiia by Tan and colleagues, in which antihypertensive treatment produced an
approximately 19.7/11.5 mmHg reduction by cuff-based home BP monitoring, while
Aktiia detected only about 1.0/0.8 mmHg of change (PMID: 37016925). Although
this medication-change subgroup was very small and requires replication, the
result is important because it illustrates precisely the distinction between
obtaining a reasonable calibrated BP value and accurately tracking a
substantial within-person BP change.
The same issue appears in nocturnal measurements. In a later
24-hour comparison, conventional ambulatory monitoring showed a substantially
larger nighttime systolic decline than Aktiia, despite relatively close
agreement in daytime mean pressure (PMID: 39927495). An independent ambulatory
comparison similarly found marked underestimation of normal nocturnal BP
dipping (PMID: 37016925). These findings raise the possibility that the optical
system can track the *direction* of some BP changes while compressing their
magnitude.
Thus, the crucial question for a calibrated PPG
blood-pressure system is not simply: “Does its BP value agree with the cuff
after calibration?” It is: does the optical signal correctly track both the
direction and the magnitude of subsequent changes in blood pressure? The two
are not equivalent—and distinguishing them is essential when interpreting
validation studies of cuffless BP technology.
Oura Nighttime Blood Pressure: A Different Approach, but the
Same PPG Limitation
Oura has taken a somewhat different approach to PPG-based
blood-pressure assessment than Hilo/Aktiia. Rather than attempting to provide
systolic and diastolic pressure in mmHg, Oura's Nighttime BP feature analyzes
nocturnal PPG patterns and estimates the degree to which blood pressure appears
to fall during sleep. The distinction is significant. Oura is not claiming that
the ring directly measured, for example: Daytime SBP = 126 mmHg → nighttime SBP
= 108 mmHg. Instead, it classifies the nighttime pattern into categories such
as:
- · typical dipping: 10–20%
- reduced dipping: <10%
- pronounced dipping: >20%
- rising: no fall or an increase overnight.
The displayed pattern is based on approximately 30 nights of
data, rather than a single night's recording. Oura explicitly states that nighttime BP is a wellness feature, does not display BP in mmHg, and should not
replace conventional blood-pressure monitoring. This represents a more modest
target than reconstruction of absolute BP. However, it does not eliminate the
fundamental physiological problem inherent in PPG-derived blood pressure: Oura
is still inferring BP from a waveform that is not determined solely by BP!
As noted, the optical pulse recorded at the finger is
affected by blood pressure, but also by many other variables (vascular tone,
arterial stiffness, wave reflection, stroke volume, heart rate, temperature,
peripheral perfusion, autonomic activity, sensor conditions, …) All of these
factors can change during the transition from wakefulness to sleep. Therefore,
if Oura observes a nocturnal change in pulse-wave morphology, that change
cannot automatically be attributed to a corresponding change in arterial
pressure. This is the same fundamental inverse problem encountered by other
PPG-based BP systems. The American Heart Association's recent scientific
statement on cuffless BP measurement emphasizes that PPG-derived pressure
estimates depend on physiological relationships that may be altered by changes
in vascular properties and measurement conditions and concludes that currently
available cuffless devices have not yet been adequately validated across all
typical use conditions (Cohen et al., 2026; PMID: 41376592).
What Oura's validation actually shows
Oura reports that its science team compared Oura Ring 4 PPG
data with 48-hour ambulatory BP monitoring in 134 participants. For
discrimination between dippers and non-dippers, Oura reports:
- · Sensitivity: 84%
- Specificity: 69%
- Area under the ROC curve: 0.87
These results are encouraging for a noninvasive screening
signal. However, as of August 2026, this 134-participant Oura validation has
not been identified as a PubMed-indexed peer-reviewed publication, and
therefore no PMID is currently available. The performance figures are reported
by Oura. The results also need to be interpreted according to what was actually
tested. An AUC of 0.87 means that the PPG-derived features contain substantial
information related to ABPM-defined dipping status. It does not mean that Oura
measured nighttime blood pressure with 87% accuracy. Likewise, the reported 69%
specificity means that, at the chosen classification threshold, approximately 31%
of ABPM-defined non-dippers would not be correctly identified as non-dippers. That
is meaningful uncertainty if the feature is interpreted at the level of an
individual. More importantly, important methodological details remain
unavailable or insufficiently described publicly, including:
- · the exact PPG features entering the dipping
model,
- the sampling rate used specifically for the
Nighttime BP algorithm,
- signal filtering and pulse-rejection procedures,
- the proportion of nighttime data rejected for
poor signal quality,
- Whether sleep stages enter the model,
- the continuous relationship between estimated
and ABPM-measured percentage dipping,
- Bland–Altman limits for percentage dipping,
- and performance in clinically important
subgroups.
Until those data are published, the reported sensitivity, specificity, and AUC are best regarded as promising preliminary validation,
rather than definitive evidence of quantitative BP-dipping accuracy.
Is the finger a better site for BP:
There are reasons why Oura could potentially perform better
than wrist PPG. The finger generally provides a stronger peripheral pulse
signal, and Oura has demonstrated very good performance for nocturnal heart
rate and interbeat-interval measurement. But obtaining a high-quality waveform
and interpreting its physiological meaning are two different problems. Higher
sampling frequency, lower motion artifact and a clearer dicrotic notch can
provide: a more accurate measurement of the PPG waveform without necessarily
providing a more accurate measurement of blood pressure.
The Aktiia studies are particularly relevant because they
demonstrate experimentally that PPG morphology can contain enough information
to produce plausible BP estimates while still substantially underrepresenting
nocturnal BP changes. That physiological limitation does not disappear when the
PPG sensor is moved from the wrist to the finger.
Thirty night averaging: potentially an advantage, but also a
different measurement
Oura's decision to average nighttime patterns over
approximately 30 nights is potentially valuable. Dipping status itself is not
perfectly reproducible from night to night. Burgos-Alonso and colleagues
examined 225 high-cardiovascular-risk patients who underwent four 24-hour ABPM
recordings over five months. Only about two-thirds of participants retained
their initial systolic dipper/non-dipper classification across subsequent
recordings, leading the authors to characterize individual dipping
reproducibility as modest (Burgos-Alonso et al., 2021; PMID: 33591600). A
second large study by McGowan and colleagues examined 512 untreated patients
with two ABPM studies approximately 29 months apart. Binary dipper/non-dipper
classification remained unchanged in 76% of participants, but agreement was
relatively weak. In contrast, nocturnal dipping expressed as a continuous
percentage was considerably more reproducible, supporting the concept that
dipping may be better regarded as a continuous physiological variable rather
than a rigid categorical state (McGowan et al., 2009; PMID: 19641455). Repeated
passive PPG measurements, therefore, offer a genuine potential advantage: they
can characterize the person's typical nocturnal physiology over many nights
without repeatedly inflating a cuff and disturbing sleep. However, this also
raises an important validation issue.
Oura's reported reference study used 48-hour ABPM, whereas
the consumer feature characterizes patterns over approximately 30 nights. Those
are not necessarily the same phenotype. A 30-night PPG average could
conceivably be a more stable measure of nocturnal vascular physiology than a
one- or two-night ABPM classification—but demonstrating that would require
prospective outcome studies, not simply agreement with a short ABPM recording.
Oura's approach is conceptually more conservative than
attempting to reconstruct systolic and diastolic BP in mmHg. A PPG-based
classifier may indeed identify useful nocturnal cardiovascular patterns even
when PPG cannot accurately reproduce absolute blood pressure. But the
underlying physiological limitation remains: The nocturnal PPG waveform is
affected by blood pressure, but it is also strongly affected by the same
autonomic and vascular changes that accompany sleep. Oura may therefore be
measuring a valuable surrogate of nocturnal BP physiology, rather than
nocturnal BP itself. That distinction should remain central when interpreting
the new Nighttime BP feature.
Some screen shots from my Hilo Core and Oura 4 ring from the same night:
Oura 4
compared to the Hilo:
Both show some nocturnal dips, but as noted above, YMMV. I do like the form factor of the Hilo and hope to see improvements in calibration over time.
Conclusion:
- PPG-based blood-pressure technology is attractive because it
promises something conventional cuff measurement cannot easily provide:
frequent, unobtrusive cardiovascular monitoring across daily life and sleep.
The optical pulse clearly contains information related to arterial pressure,
and increasingly sophisticated algorithms can extract timing, amplitude, and
morphological features that correlate with BP and vascular state.
- The central
limitation, however, is physiological rather than merely technical. A PPG
sensor does not measure arterial pressure directly. It measures changes in
peripheral blood volume, and the resulting waveform reflects the combined
influence of blood pressure, vascular tone, arterial stiffness, pulse-wave
reflection, stroke volume, heart rate, temperature, perfusion, autonomic
activity, body position, and sensor conditions. Blood pressure is therefore one
contributor to the waveform, not its sole determinant. That distinction has
practical consequences.
- A device can agree closely with a cuff after
calibration without necessarily tracking subsequent BP changes accurately.
Calibration may establish the correct baseline or offset while leaving errors
in the dynamic gain of the PPG-to-BP relationship. The Aktiia/Hilo literature
illustrates this particularly well. Controlled seated measurements can show minimal average bias, while independent and ambulatory studies have
demonstrated substantial attenuation of nocturnal BP dipping and, in limited
data, large medication-induced changes in BP (PMID: 33675592; PMID: 37016925;
PMID: 39927495). This should change how cuffless BP validation is interpreted.
Low mean bias around the calibration condition is important, but it does not
answer the more physiologically relevant question: Does the device correctly
reproduce the magnitude of within-person BP changes across different
physiological states? For many applications, including nocturnal dipping,
medication response, BP variability, exercise recovery, and stress
responses—that may be the more important test.
-
Nighttime monitoring is especially challenging because sleep
alters many of the same physiological variables that determine PPG morphology.
Sympathetic activity, vascular resistance, arterial wave reflection, skin
temperature, peripheral perfusion, heart rate, and stroke volume all change
during sleep. A nocturnal change in the optical pulse therefore cannot
automatically be interpreted as a proportional change in arterial pressure. This
limitation also applies to devices such as Oura that take a more conservative
approach and infer BP-dipping patterns rather than reporting absolute SBP and
DBP. Classification is an easier problem than reconstruction of pressure in
mmHg, and a PPG-derived dipping phenotype may ultimately prove useful. But
successful classification does not establish that the ring directly measured
the magnitude of the BP decline. The algorithm may instead be detecting
autonomic and vascular changes that are associated with normal dipping.
-
Signal quality and sampling rate matter, but they do not
solve the underlying physiology. A higher sampling rate can improve
identification of the systolic peak, dicrotic notch, reflected wave, and other
pulse landmarks. Better sensor placement can reduce noise and motion artifact.
More sophisticated machine learning can identify complex patterns that simple
regression cannot. Yet none of these improvements make PPG morphology uniquely
determined by blood pressure. Likewise, aggressive signal-quality filtering can
create another potential source of bias. Wearables commonly reject periods
containing motion, poor perfusion, atypical pulse morphology, or inadequate
sensor contact. This improves technical signal quality but may also
preferentially exclude unusual physiological states. A device may therefore
provide very clean estimates during selected low-motion periods while
incompletely representing the full variability of blood pressure across the day
or night.
-
For future validation studies, the most informative analysis
may therefore be less about absolute agreement after calibration and more about
tracking change. Studies should deliberately expose participants to a broad
range of pressures and physiological states. None of this means that PPG-based
BP technology lacks value. On the contrary, PPG may ultimately prove extremely
useful for long-term cardiovascular phenotyping, detecting trends, identifying
individuals who warrant formal BP evaluation, and recognizing physiological
patterns that conventional intermittent measurements miss. A multi-night
optical phenotype may even prove more reproducible or prognostically
informative than a single night of cuff-based monitoring.
-
“Can PPG predict blood pressure?” It clearly can to some
degree. The more important question is: “Can a particular PPG algorithm
distinguish a true change in arterial pressure from the many other
physiological changes that alter the peripheral pulse waveform—and can it do so
accurately enough for the intended use?” That is the standard by which
PPG-based blood-pressure technology should ultimately be judged.
