Does Hilo Core Track Blood Pressure Accurately Across the Day?

As a physician concerned about blood pressure and a wearable enthusiast, I wanted to share some observations on the PPG device called Hilo Core. It may represent the most officially studied and "validated" PPG wrist device available (see PMID  33675592, 37016925, 38997475, and 39927495). After my previous post dealing with the considerable hurdles to overcome for this type of tech to be accurate, I was quite interested to see how it did. The main issue, though is not having "ondemand" BP capability. After looking over my data, I wonder if this is by design to make this type of comparison difficult. However, we can outsmart the Hilo app and do a simultaneous cuff vs wrist PPG relatively easily.

Here is how I did it: The Hilo band will only measure "BP" when you are stationary. This rules out anything while exercising or even perhaps if the HR is high. Next, the app segments the readings in 2 hour blocks:

 

Therefore, if we have the situation above, when 6 PM rolls around and no reading is done, we have an opportunity to do a simultaneous cuff reading. What seems to work is opening the app, and doing a manual cuff measurement:

 

Usually (if you are motionless), the wrist PPG will also do a measurement within a minute or two. If the 6 PM field is already populated, it will not show you the separate readings, so you will have to wait for the 8 PM block. You can also achieve the same concept by just keeping the band off the wrist until you are ready to do a cuff reading.

I began by doing the recommended calibration with the supplied cuff while in a quiet, resting, early AM state, before starting my (hectic) day. My goal was to see if the daytime reading "agreed" with cuff measurements based on the company recommended calibration procedure. As I discussed in the prior post, this may not be the best way to go about calibration.

As the paired reading came in, I was disturbed at how disparate the numbers were. The daytime values were after exercise, food, or just done randomly. About 10 days into this, I decided to recalibrate but do it during the day, which is when my measurements occurred. The results are as follows:

 

Calibration time	Measure	n	Cuff mean	Hilo mean	Mean difference	Paired t	p
Daytime	Systolic BP (mmHg)	21	124.7	123.4	−1.3	−0.77	0.450
	Diastolic BP (mmHg)	21	74.2	72.7	−1.5	−1.36	0.190
	Heart rate (bpm)	21	63.4	63.5	+0.1	0.40	0.691
Morning	Systolic BP (mmHg)	19	126.8	115.3	−11.5	−5.65	<0.001
	Diastolic BP (mmHg)	19	74.1	70.5	−3.6	−3.73	0.0015
	Heart rate (bpm)	19	65.4	65.1	−0.3	−1.19	0.250

 Systolic Bias:

 

• There is a large bias with the morning calibration, which is not present with the daytime calibration group.

Diastolic Bias:

• Bias is better than systolic, and daytime calibration is better than morning. 

Correlation:

 

 

• Systolic regression is poor for morning calibration and poor for daytime calibration

• Diastolic regression is good for both calibrations.   

 

Heart Rate:

 

 

• Excellent agreement and correlation with HR cuff vs PPG

 

Key findings:

1. Systolic blood pressure showed a major calibration dependent difference

• With the daytime calibration, Hilo systolic pressure averaged only 1.3 mmHg lower than the cuff.
• With the morning calibration, Hilo averaged 11.5 mmHg lower than the cuff.
• This approximately 10.2-mmHg difference in measurement bias between the two calibration periods was statistically significant:

• The 95% confidence interval for the difference was approximately 4.8 to 15.6 mmHg.
• This was the clearest finding in the dataset.

2. Daytime calibration largely removed the average systolic bias

• Mean Hilo−cuff systolic difference: −1.3 mmHg

• Paired comparison: p = 0.45

• Therefore, there was no evidence of a systematic average difference between Hilo and the cuff during this period.

        During the morning-calibrated period:

• Mean difference: −11.5 mmHg

• Paired comparison: p < 0.001

Hilo therefore substantially underestimated systolic pressure during the morning-calibration period.

3. Better average agreement did not mean good agreement for individual readings

• Although daytime calibration almost eliminated the average systolic bias, the Bland–Altman limits of agreement remained relatively wide:
• Daytime calibration: −16.9 to +14.2 mmHg
• For the morning calibration they were even wider:
• −29.0 to +5.9 mmHg
• This means that individual Hilo systolic measurements could still differ considerably from the cuff even when the average bias was close to zero.

In other words, calibration improved accuracy of the average but did not make individual Hilo and cuff readings interchangeable.

4. Systolic correlation remained poor under both calibration conditions

• The correlation between Hilo and cuff systolic measurements was:

• Daytime calibration: r = 0.29

• Morning calibration: r = −0.19

• Neither represents strong tracking of individual systolic changes.
• Importantly, the difference between these two correlations was not statistically significant.

Thus, recalibration primarily improved the absolute bias, rather than producing a dramatic improvement in measurement-to-measurement correlation.

5. Concordance remained weak for systolic pressure

Lin's concordance correlation coefficient, which assesses both correlation and agreement, was:

• Daytime calibration: CCC = 0.26

• Morning calibration: CCC = −0.05

The daytime value represents an improvement, but agreement remained weak.

This again indicates that a good average value does not necessarily mean Hilo reliably tracks each individual cuff measurement.

6. Morning calibration showed evidence of proportional systolic bias

• The Bland–Altman analysis also examined whether error changed depending on the actual blood pressure.
• With morning calibration, there was significant proportional bias:
• p = 0.022
• This means the Hilo−cuff difference was not simply a constant offset; the amount of disagreement changed across the BP range.
• With daytime calibration, proportional bias was weaker and narrowly missed conventional statistical significance:
• p = 0.060
• This suggests that daytime recalibration may have reduced both the fixed bias and some of the pressure-dependent error.

7. Diastolic pressure was less affected by calibration timing

• The average Hilo−cuff diastolic difference was:

• Daytime calibration: −1.5 mmHg

• Morning calibration: −3.6 mmHg

• The difference between calibration periods was approximately 2.1 mmHg, but this was not statistically significant:
• p = 0.155

Therefore, the strong calibration-dependent effect seen for systolic pressure was not clearly present for diastolic pressure.

8. Diastolic correlation was considerably better than systolic correlation

• Cuff and Hilo diastolic values correlated moderately well:

• Daytime calibration: r = 0.60

• Morning calibration: r = 0.63

• This was substantially better than the systolic relationship.

However, morning-calibrated diastolic measurements still showed significant proportional bias, illustrating why correlation alone does not establish agreement.

9. Heart-rate agreement was excellent under both conditions

• Heart rate served as an informative comparison because both devices measured HR at the same observations.
• The Hilo−cuff HR difference was only:

• Daytime calibration: +0.14 bpm

• Morning calibration: −0.32 bpm

• Correlations were extremely high:

• Daytime: r = 0.94

• Morning: r = 0.96

• Lin's concordance coefficients were similarly strong:

• Daytime: 0.94

• Morning: 0.96

This suggests that the poor systolic agreement was not simply caused by gross temporal misalignment between measurements.

10. The data are consistent with a calibration-state effect

The original Hilo calibration was performed early in the morning, while many subsequent comparisons were performed during the daytime and around exercise.

After recalibration under daytime conditions, the average systolic error improved from approximately:

−11.5 mmHg to −1.3 mmHg

This raises the possibility that the relationship Hilo uses to estimate blood pressure may depend partly on the physiological state present at calibration.

Potential contributors could include changes in:

• vascular tone

• sympathetic activity

• arterial stiffness

• peripheral vasoconstriction or vasodilation

• temperature

• posture

• recent exercise

A cuffless system based on pulse-wave characteristics may therefore perform best near the physiological conditions under which it was calibrated.

11. What does this mean for someone trying to determine whether they have mild hypertension?

This may be the most practically important implication of the experiment.

Under the current 2025 AHA/ACC blood-pressure classification, adult BP is categorized as:

• Normal: systolic <120 and diastolic <80 mmHg

• Elevated: systolic 120–129 and diastolic <80 mmHg

• Stage 1 hypertension: systolic 130–139 or diastolic 80–89 mmHg

• Stage 2 hypertension: systolic ≥140 or diastolic ≥90 mmHg

What is sometimes casually described as “mild hypertension” therefore generally corresponds to the Stage 1 range beginning at 130/80 mmHg. The guideline classification is based on averages of multiple careful measurements rather than a single isolated reading.

This is precisely the range in which measurement error can become clinically important.

For example, consider someone whose true cuff systolic pressure averages 132 mmHg, just inside the Stage 1 hypertension range.

With the −11.5-mmHg average systolic bias observed during the morning-calibrated portion of this experiment, a pressure of 132 mmHg could, on average, be represented by Hilo as roughly 120–121 mmHg.

That would move the apparent result from Stage 1 hypertension into the “elevated” category.

This calculation is only an illustration—the individual Hilo error cannot be predicted simply by subtracting the average bias—but it demonstrates the potential magnitude of classification error.

The opposite problem is also possible because the Bland–Altman limits were wide. Some individual Hilo readings were substantially above the corresponding cuff value.

Therefore, depending on the particular reading, a person near a diagnostic threshold could potentially be classified in either a higher or lower BP category.

12. Daytime calibration improves this problem—but does not completely solve it

After daytime calibration, the mean systolic bias was only −1.3 mmHg.

At the population-average level, that looks excellent. A cuff pressure of 132 mmHg would correspond to approximately 131 mmHg after applying the observed mean difference, leaving the measurement in the same Stage 1 category.

But the Bland–Altman limits remained approximately:

−16.9 to +14.2 mmHg

That range is considerably larger than the 10-mmHg width of the entire Stage 1 systolic category from 130 through 139 mmHg.

Consequently, a small average bias does not establish that an individual Hilo reading can reliably answer the question:

“Is my true blood pressure below or above 130 mmHg?”

The same issue applies to the 80-mmHg diastolic threshold.

For someone whose actual diastolic pressure sits near 80–85 mmHg, an error of only several mmHg could move the displayed result between normal/elevated and Stage 1 hypertension.

13. Classification is different from trend monitoring

This highlights an important contrast between two potential uses of a wearable BP device.

One use is:

“Is my blood pressure generally going up or down over weeks or months?”

A device might potentially provide useful longitudinal information even with some individual measurement error.

A much more demanding use is:

“Is my actual average BP 126, 132, or 142 mmHg, and therefore which hypertension category am I in?”

That requires sufficient absolute accuracy around clinically important thresholds.

The results of this N-of-1 experiment suggest caution about using Hilo alone for the second question, particularly when the user's BP is close to 130/80 mmHg.

14. This concern is consistent with current AHA guidance on cuffless devices

The 2025 AHA/ACC hypertension guideline specifically states that reliance on cuffless devices, including smartwatches, for accurate BP measurements should be avoided until they demonstrate greater precision and reliability.

The AHA's subsequent scientific statement on cuffless BP devices notes that these systems usually estimate BP indirectly, often relative to a calibration measurement, and that accuracy can be affected by factors including calibration, motion, posture, and real-world physiological conditions.

For home BP assessment, the AHA currently recommends a validated automatic upper-arm cuff device. It recommends taking multiple measurements under standardized resting conditions and recording results over time.

That recommendation is particularly relevant for someone wondering whether their average BP is just below or just above the Stage 1 hypertension threshold.

Practical implication for someone near 130/80 mmHg

If someone's Hilo readings consistently average, for example, 125/76 mmHg, 132/81 mmHg, or another value close to a diagnostic boundary, this experiment suggests that the wearable value alone may not be sufficiently precise to determine whether that individual truly has Stage 1 hypertension.

A sensible approach would be to use a validated upper-arm cuff as the reference measurement, under standardized conditions:

• rest quietly for at least five minutes

• avoid exercise, caffeine and smoking for at least 30 minutes beforehand

• support the arm at heart level

• take two readings approximately one minute apart

• repeat measurements across multiple days

For determining whether a person meets criteria for hypertension, reference BP should be measured under standardized resting conditions. However, evaluating the real-world validity of a cuffless wearable requires a different approach. Comparisons should also include physiologically challenging conditions—such as daytime activity, changes in posture, and post-exercise recovery—because these are precisely the situations in which vascular tone and pulse-wave characteristics may depart most from the calibration state. 

These procedures are consistent with current AHA home-monitoring recommendations.

The 2025 guideline also emphasizes that BP classification is based on an average of at least two careful readings obtained on at least two occasions, rather than an isolated measurement.

For someone close to 130/80, the average of properly obtained cuff measurements is therefore considerably more appropriate for answering “Do I meet the definition of hypertension?” than an isolated or cuffless wearable reading.

Overall conclusion

The most important observation was that Hilo systolic accuracy changed markedly after recalibration.

Morning calibration was associated with an average systolic underestimation of approximately 11.5 mmHg, whereas daytime calibration reduced that error to approximately 1.3 mmHg.

However, even after recalibration, Bland–Altman limits remained wide and correlation with individual cuff systolic measurements remained limited.

For someone whose main question is whether they have borderline or Stage 1 hypertension, these findings matter because the AHA threshold begins at 130/80 mmHg, while the observed individual disagreement between the Hilo and the cuff was often considerably larger than the few millimeters of mercury separating one BP category from another.

Thus, the data suggest that calibration may strongly influence Hilo's average systolic BP accuracy, while the remaining measurement-to-measurement variability limits confidence in using the wearable alone to determine whether an individual's true BP lies just above or below a hypertension threshold.

For that purpose, current AHA guidance continues to favor repeated measurements using a validated automatic upper-arm cuff.

 

Why should we be concerned about exercise and recovery BP dynamics?

An exaggerated blood pressure response during exercise has been associated with masked hypertension, future development of hypertension, and higher cardiovascular risk, even in people whose resting BP appears normal.

A 2023 meta-analysis found that people with masked hypertension were more than three times as likely to show an exaggerated BP response to exercise compared with truly normotensive individuals (OR 3.33), suggesting that exercise BP can expose hypertension that is not apparent at rest (PMID: 36980313).

Longitudinal studies support this relationship. In 3,742 initially normotensive men, a peak exercise systolic BP above approximately 181 mmHg predicted a higher risk of developing hypertension over about five years (PMID: 25824452). An earlier study of 5,386 normotensive men likewise found that an exaggerated exercise BP response independently predicted future hypertension after accounting for resting BP and other risk factors (PMID: 9467632).

A systematic review including more than 35,000 normotensive participants concluded that exaggerated exercise BP is associated with both future hypertension and cardiovascular events (PMID: 28511070). Exercise SBP has also been associated with adverse vascular and metabolic characteristics, including greater arterial stiffness and poorer endothelial function (PMID: 34738988).

Importantly, some investigators argue that submaximal exercise BP may be particularly useful, because moderate workloads resemble the cardiovascular demands of normal daily activity more closely than maximal exercise. Practical guidance has therefore proposed measuring BP at a standardized moderate workload as a way to identify individuals whose hypertension may not be evident from resting measurements alone (PMID: 35270514). 

BP context	Established role
Standardized seated BP	Primary classification/diagnosis
Home BP	Diagnosis/management
24-h daytime ABPM	Established diagnosis + risk prediction
Submaximal exercise BP	Risk marker; can uncover masked HTN
Peak exercise BP	Risk marker, but thresholds less standardized
Recovery/post-exercise BP	Emerging/established prognostic marker, not diagnostic criterion
BP during unrestricted activity	Interesting physiologically, less standardized

 

Hilo PPG BP, the bottom line:

Cons:

• Poor agreement and correlation to daytime BP when calibrated per recommendation (in AM)
• Minimal bias, but still a large limit of agreement and weak correlation when calibrated for daytime.
• Can easily miss mild HTN
• Only measures at rest with no motion
• Can't use pre/post/during exercise 
• Difficult to do on-demand readings.
• AHA recommends avoidance of PPG BP usage
• Cost and subscription model

Pros:

• Minimal footprint and weight
• Does not require inflation (especially important at night)  
• If calibrated properly, overall average BP does match with cuff 

 

Relevance to Smartwatch BP:

The major smartwatch companies are now moving aggressively into blood-pressure monitoring, but their approaches differ substantially.

Google's Pixel Watch 5 has begun rolling out a new Blood Pressure Trends feature as part of Google Health's “Health Guardian” system. Rather than displaying conventional systolic and diastolic numbers, the watch analyzes physiological signals continuously and looks for changes in blood-pressure patterns over time. Google describes this as a trend-monitoring feature rather than a replacement for cuff measurements. The feature is also being extended to Pixel Watch 3 and 4.

Apple Watch takes a similar approach. Apple now offers hypertension notifications on supported watches. The optical heart sensor analyzes vascular responses over approximately 30-day periods and can notify a user when it detects a pattern consistent with chronic hypertension. Importantly, Apple does not provide an estimated 128/82-style BP reading. If a hypertension notification occurs, Apple recommends confirming the finding with a conventional BP cuff over seven days.

Samsung takes a different approach. Its Galaxy Watch blood-pressure feature is now available to U.S. users on compatible watches and provides estimated systolic and diastolic BP values. Like Hilo, however, it requires calibration against an upper-arm cuff and requires recalibration every 28 days. Samsung explicitly states that the feature is not intended to diagnose disease. Samsung has also announced passive blood-pressure trend monitoring for later in 2026.

Accuracy issues become particularly important near diagnostic boundaries such as 130/80 mmHg, where an error of only several millimeters of mercury can change BP classification.

For this reason, claims that a smartwatch can detect hypertension, monitor blood-pressure trends, or estimate systolic and diastolic pressure should not automatically be interpreted as evidence that it can replace a validated upper-arm cuff.

The critical question is not simply whether the device performs well in a controlled resting validation study. It is whether it remains accurate across the conditions in which people actually intend to use it—morning, daytime activity, sleep, emotional stress, exercise and post-exercise recovery—and whether it can correctly identify people close to clinically important BP thresholds.

Until independent studies demonstrate that level of real-world performance, smartwatch BP and hypertension features are best viewed as screening or supplementary tools rather than definitive measurements of blood pressure or proof that hypertension is present or absent.

Blog Index