Month: September 2025

Last week Apple launched their Hypertension Notification Feature (HTNF) on Apple Watch Series 9 or later and Apple Watch Ultra 2 or later (excluding Apple Watch SE). According to Apple the feature is intended for people over 22 years of age who have not been diagnosed with hypertension. It is not intended to be used during pregnancy. The watch uses photoplethysmography (PPG) which is a 14 letter word (44 Scrabble points) which means that the watch looks at blood volume changes through the skin and uses that information to predict who has hypertension. The watch records 60-second segments of PPG signals as inputs. These signals are collected roughly every two hours throughout discreet 30-day evaluation windows. The watch uses the accelerometer to assure that only PPG signals collecting during rest are used are used to assess for hypertension. A blood pressure score is created for each segment. Segments scored during sleep are ignored. While creating the model, the actual blood pressure was assessed by home blood pressure measurement.

Apple’s white paper about the system provides a classic Table 1 of demographic information on training, validation, and testing cohorts:

I was not familiar The Fitzpatrick Scale. It is an assessment of skin color, V and VI are described by Wikipedia as “dark or brown” and “very dark or black” respectively. Pulse oximetry over estimates oxygen saturation in people with darker skin. Apple prospectively assessed hypertension detection for similar problems.

When determining who to alert for hypertension Apple decided to value specificity over sensitiviy to minimize false positives. From the white paper:

A receiver operating characteristic (ROC) curve was generated based on thedevelopment data, and an operating point was selected to prioritize high specificity — minimizing false positives — while also notifying a significant portion of people with hypertension

After the system was complete, Apple did an additional clinical validation study. They enrolled 2,229 people from two cohorts, all-comers without known hypertension risk factors and those at risk of hypertension based on historical blood pressure values and other risk factors. Each person was sent an Omron Evolv Wireless BP cuff (which scored very high in this review) and instructed to measure their blood pressure twice a day as well as wear the Apple Watch at least 12 hours a day.

Blood pressures were scored using the 2017 ACC/AHA hypetension guidelines:

The cohort for this part of the study was admirably diverse

The full set analysis is the original cohort. The Notification Analysis set only includes people who performed the home blood pressure assessments and wore the watch as instructed. They look reasonably similar except for the higher fraction of women in the Notification Analysis Set.

The investigators found that roughly a third of the Notification Set had hypertension and the watch notified about 40% of them. Only 99 of the 1,278 people without hypertension were given notices that they may have hypertension.

Here is how the test performed using a two-way table:

I feel a little bit out of my element since this is a test being being done in the background on people with no pre-test suspicion of disease. This is unlike any test I have ever ordered or really thought about. It is not a screening situation, where you want to maximize sensitivity at the expense of specificity in order to move a disease enriched population to more specific downstream studies. I think Apple’s description for their rational for tuning the ROC curves is appropriate. They want to capture people with hypertension while minimizing any unnecessary alarm and inconvenience for their customers with false positives.

The FDA application includes information from a 2-year study that is not further described. But it describes that specificity remains high month after month and that almost all of the false positives occur only in the first month.

I am curious, but the data is not provided, if the Hypertension Detection System increases sensitivity over the two years? Does it pick up additional cases of true positives? That would be cool if it did.

If you want to turn hypertension detection on for your Apple Watch, go into the Health App and scroll down to hypertension detection and answer the two questions and then follow through the four panels which describe how it works and to do home BP monitoring for more accurate diagnosis and then a generic warning that hypertension is not a heart attack. The last panel is not from the onboarding flow, but is another panel about hypertension that the health app provides. Nice.

I have hypertension, but I want to play with the warning system so I lied during set up regarding this diagnosis.

Apple received FDA approval for Hypertension Detection. You can read the regulatory filing and response here.

I love getting e-mails from interesting people. Here is a recent one:

Dear Dr. Topf:

First of all, congratulation to your recent promotion!

I have one question regarding the treatment of acute hyperkalemia. As an avid follower of your posts and blogs, I have implemented your “bag of saline with furosemide”-approach for several years in appropriate patients and it works like a charm. I particularly like the beauty of the physiology behind this supposedly simple regime, with flow-induced recruitment of BK (FIKS) supporting the job of ROMK.

However, in my experience, this treatment is woefully underused in clinical practice and little known – even among nephrologists.

Are you aware of a publication substantiating its use? Which reference should I quote?

Looking forward to your reply!

Best wishes,

XXXXXXXXXX

Hmm? Data to support the use of loop diuretics in the treatment of hyperkalemia. Let’s see what we can find.

The first reference I found was this article from 1984

These authors looked at renal potassium excretion in patients following 40 mg of furosemide daily for three days. They did this experiment in 6 health volunteers five separate protocols:

1. Furosemide with a high salt diet (270 mEq of Na per day)
2. Furosemide with a low salt diet (15-20 mEq of Na per day)
3. Furosemide with a high salt diet and captopril, intended to prevent increased aldosterone release with diuresis
4. high salt diet and captopril, this time without furosemide
5. Furosemide with a high salt diet and water load, this was intended to suppress ADH, which has a kaliuretic effect

The authors found that in the acute phase, the first 3 hours after IV furosemide, patients excreted 16 mEq of additional potassium over baseline potassium excretion with the high sodium diet, 19 mEq over baseline in the low salt group (presumably due to higher aldosterone) and 13.5 mEq in the high salt and captopril group.

Interestingly, the authors found that following this acute phase of increased potassium excretion there was a compensatory period decreased potassium excretion in the high sodium group that resulted in just about neutral potassium balance.

It makes me wonder if adding fludrocortisone would be helpful to the “bag of saline with furosemide”-approach to hyperkalemia.

I posed this question to my favorite potassium expert, Melanie Hoenig, and she suggested these references.

First was a circulation manuscript from when furosemide was the new kid and we were still trying to feel it out. These authors were playing around with weekly infusions of furosemide to treat hypertension. And it worked. (Link)

But, helpfully, they also provided information on potassium excretion, but unhelpfully they provided the data in micro Equivalents of potassium per minute.

So we have to do the math

Let’s say:

188 µEq/min from 0 to 22 minutes = 4.1 mEq
180 µEq/min from 22 to 45 minutes = 4.1 mEq
122 µEq/min from 45 to 90 minutes = 5.5 mEq
62 µEq/min from 90 to 120 minutes = 1.9 mEq

So a total of 15.6 mEq in the first two hours with an average of 63 mg of furosemide

195 µEq/min from 0 to 22 minutes = 4.3 mEq
201 µEq/min from 22 to 45 minutes = 4.6 mEq
166 µEq/min from 45 to 90 minutes = 7.5 mEq
91 µEq/min from 90 to 120 minutes = 2.7 mEq

And 19.1 mEq in the first two hours with an average of 210 mg of furosemide for you cowboys.

The next reference she sent was from the Canadian Medical Association Journal in 1968 where they took 115 male students and gave them one of three diuretics:

Hydrochlothiazide 50mg
Hydrochlorothiazide/triamterene 50/25
Furosemide 40 mg

And then tracked them for 24 hours. The results are interesting. Here are the potassium results:

73 mEq of potassium. Pretty impressive. But look how hydrochlorothiazide does just as well, though it is backloaded with most of the kaliuresis coming later in the monitoring period. Hydrochlorothiazide actually resulted in more sodium excretion, over 24 hours than a single dose of furosemide.

The last reference she sent was a 1964 manuscript from Circulation where the authors were playing with the, then novel furosemide. They were using it in edematous patients resistant to available diuretics (acetazolamide, spironolactone, meralluride, and thiazides) as well as normal patients. Here are the daily electrolyte losses with various doses of oral furosemide, from 50 to 600(!) mg.

Good to know that doses beyond 100 mg don’t seem to add much kaliuresis.

So to answer the question, a slug of furosemide seems to be good for 15 mEq of potassium removal acutely, given that the extracellular volume of 70 kg man is about 17 liters, this should drop the serum potassium by a little less than 1 mEq/L.

None of these studies look at patients with hyperkalemia, so I would really like to see any experimental evidence with that, so if you know of any, hit me up on socials.