Why are doctors wary of wearables? | Hacker News Reader

Why are doctors wary of wearables?

bbc.com

33 pointsNN8851 comments

The problem is people having devices monitor different metrics of their body without the knowledge to know what to do about what they see.

As mentioned in the article, wearables monitoring heart rate variability do quite a good job of telling people ahead of time that your body might be trying to fight something. During covid devices monitoring HRV would spike before people started to present with the symptoms of COVID-19.

Monitoring every aspect of your body and going to the hospital at any sign of "out of baseline" metrics is silly. There will always be a group of people that get a "high heart rate" notification and think its an emergency, or start to make generalised diagnosis of issues.

This is the premise of the issue that the author assumes, that doctors are somehow gatekeeping a cure to the common cold/flu -- a premise which is generally untrue.

But, as the author's example, a cold/flu; is not generally useful to doctors, because doctors can not fight a cold. At worst, they get more visits from people with colds in their waiting rooms, are pressured into over prescribing antibiotics - which impact the individual's gut biome and society's antibiotic resistance.

What's extremely odd, is an article like this in the time of telemedicine; advantageous because it uses less of a provider's time, is typically lower cost, and reduces chance of office visit infections; and of which there are very few actionable outcomes.

Humans are complex in their emotions, they might see a metric on their wearable device that alarms them that shouldn't and they visit a hospital out of concern. People who are more "hypochondriacs" might be more likely to buy wearables and hawk eye anything that becomes out of baseline. You can't likely help those people unless you address their underlying sensitivity to their health.

At the end of the day, smart devices have made people more aware of their bodies particular vitals and therefore will create more people asking for help out of concern, of which for the majority isn't likely to cause anything of concern

> As mentioned in the article, wearables monitoring heart rate variability do quite a good job of telling people ahead of time that your body might be trying to fight something. During covid devices monitoring HRV would spike before people started to present with the symptoms of COVID-19.

I think the article was talking about changes in heart rate, e.g., increases or decreases in the beats per minute. It certainly makes sense from an English point of view to call that heart rate variability but it turns out there is an actual medical term "heart rate variability" (HRV) that means something a little different.

Heart rate: count the number of beats in a minute.

Heart rate variability: how the beat to beat interval changes.

So say if you have a rate of 60 beats per minutes the average time between beats will be 1000 ms, but the actual times will vary around that. HRV is a measure of that variation.

Some wearables do track both rate and HRV.

No I was absolutely talking about heart rate variability in my comment and I wasn't confused by the term.

The article was talking about wearables being able to measure these types of metrics and what average people do in response to those metrics.

The article spoke about a Ring which indicated to the author that they were becoming sick, this watch also uses HRV and uses HRV amongst other metrics to determine things like illness.

Yes wearables measure both HR and HRV, I was talking specifically about HRV.

I find my HRV drops when I have a virus but my resting heat rate goes way up

Makes a lot of sense given that a lower HRV from your baseline is considered to be a heightened stress state

So I have high blood pressure (middle age it seems sucks), and have been under a cardiologist for 18 months. The first thing he said was “this wearable is good”, so of course I bought it and it’s been on my wrist consistently.

The company is legit but has made considerable effort not to allow the data out of their eco-system (exports only to pdf, no ability to send to Apple Health etc).

So I notice that I send a quarters worth of pdfs to the doctor, whose secretary prints them out and he flips through them with a strange expression trying to work out what’s going on.

I have taken a few frustrating lunchtimes to parse out the pdfs and have gotten to the point I can get the data and graph it - but what to usefully do with it is interesting

Company: Aktiia - look for mikado-aktiia on github

Out of curiosity, what is the value in constant heart rate measurements your doctor sees? I have taken blood pressure meds for about a decade now and don't think that's ever been suggested for me but also my PCPs never acted particularly considered about my high BP readings given my age for a long time. I kinda had to self advocate for meds that (mostly) control it.

I was hoping losing a lot of weight (now in the normal range) would help but still seem to need the meds and by all indications I have a consistently high heart rate which concerns me a bit as I get older.

Agreed on the “how do I fix this” issue - I run about four miles a day which I hoped would help, and need to lose more weight but it seems mostly “middle age - cannot be fixed only managed”

As for the constant monitoring - I don’t know. I think he would be happy with weekly manual blood pressure or something - because as someone points out downthread, that’s what is comparable ans what has decades of medical experience (research/ empirical experience)

But one believes that monitoring more is going to lead to more effective or earlier decision making.

I worked in research on wearables for more than a decade, including research with medical doctors. The number one reason we got from doctors about not using wearable data was that they didn't know what to do with it, what it meant. If blood pressure is too high when they've had you sit still for a while and taken a measurement in their office with their machine, they know how to interpret that. If you show them a continuous blood pressure trace from a persons whole day, they have no idea what to do with it. How much should your BP go up when you are doing something strenuous? How much should it go down when you sleep? They don't know. The volume of data also dismayed them. Present a years worth of heart rate data and who has time to look through it all and understand what it means? You need to know the context of the readings, what the person was doing, what was going on in the environment, even what the person was thinking, and wearable data comes with little context (mainly just accelerometer and gyro data, but it can't tell if you are eating dinner or watching tv or typing on a keyboard). The UI's for looking at the long term data are terrible, and it requires a lot of processing power to get a reasonable response time, and there is no ability to query the data, such as looking for time-offset correlations between one type of reading and another. The data just is not very usable without context. A person looking at their own data for the day remembers the context, what they did that day, but how about a month ago? A year ago? Sensor accuracy still needs improvement also (e.g., did your body temperature go up today because you're getting sick or because you spent the day outside when it was hot instead of staying inside in the air conditioning? If the sensor is affected by ambient temperature there is no way to tell.) Wearable health sensing is good enough to be somewhat useful to an individual, but it needs to be developed a lot more to be useful to a doctor (though it can provide clues that are useful sometimes). There need to be a lot of RCT's to understand just what the data means, what is normal for a person, what indicates a problem. There is also the fear that if the doctors help develop this technology it will replace a lot of their diagnostic function. That could be good, but it could also result in everyone trusting the machine's decisions rather than looking at what is actually happening with the person.

To give an example, a wearable might indicate a sudden change in gait for an elderly person. Is that a sign of mental decline? Muscle deterioration due to age? A stroke? Or were they playing baseball two days ago with a grandchild and injured themselves? New shoes? It could be any of these and more besides. If the machine does not evaluate all the possibilities it can only choose from the ones it does implement, which raises the likelyhood it will be wrong.

> How much should your BP go up when you are doing something strenuous? How much should it go down when you sleep? They don't know.

The point about BP and strenuous tasks highlights how useless this data is on its own: Random data points from uncontrolled scenarios mean very little. Blood pressure measurements performed in standard conditions (seated, stationary for several minutes, doing nothing else) are comparable. Blood pressure measurements taken in unknown conditions are not.

Hypochondriacs are especially bad at misinterpreting data from wearables and self-testing. The more data points they have, the more likely they are to convince themselves that they have a number of conditions.

True, data can increase paranoia. It can have good effects also. One of the most common effects from wearing an activity tracker is people suddenly realize they are hardly moving at all. 95% of the time they are sitting (often 99% of the time). They thought they were an active person, but they only remember the active times, not the sitting times. Seeing the data can convince a person they need to move more. Maybe clinicians should recommend NOT using a tracker to people they see are hypochondriacs? (Better mental health evaluations should be a bigger part of medicine, though privacy is an issue.) Most people seem to use trackers for sports activity tracking. BP measurements throughout a day may actually have meaning, we just haven't done the research to know what that might be yet. There might be generalities, like a generally accepted "typical" peak BP value when working out in a gym or running. Or there may be things that are more specific that can be discovered, like BP should not peak and stay high for an hour after just standing up for a minute. We can probably answer some questions like what is the typical effect of a cup of coffee or a full meal on BP? Data inference can happen on mobile computers now, a smartwatch can tell when you are asleep without consulting a supercomputer. In the future these wearables will not report raw data, they will report inferences from algorithms. Figuring out those algorithms is just beginning.

I'm a physician. Agree with your comment... in a statistical sense this relates to screening test characteristics. As the sensitivity of a test increases, the specificity decreases.

A good example is screening for atrial fibrillation, an abnormal heart rhythm that is not uncommon especially as you age. All our guidelines right now on treatment and stroke prevention (a potential consequence of AF) is based on the population who is currently most often diagnosed - symptomatic patients with palpitations, or lightheadedness, or chest pain, which provokes them to see a physician.

If everyone is now wearing an Apple watch and suddenly we have 2x or 10x the number of people diagnosed with AF, our current evidence about the benefits of treatment suddenly do not apply, because these newly found patients would never have been included in the original studies.

So what do we do with these (presumably lower risk) patients? No one yet knows.

Yes, exactly why new evidence is needed, new RCT's.

Certainly at this point we have some of these answers, right? I have to imagine a lot of medical researchers have been excited about all this data. So is that information just not making its way to them? Is there not software? I got to imagine there is if lawyers want the data

Sort of, a little. There was a lot of excitement about wearables for medical sensing early on, but the enthusiasm died down a lot when it became apparent that you can't just collect sensor data from the person and turn that into something meaningful in terms of health. For example, one project I worked on was measuring stress using heart rate and heart rate variability. You can get a fairly good idea of a persons stress level from this. But some stress is normal, like when you give a presentation in front of a crowd or do complex math. So is high stress good or bad? It's both, it depends on the context. What a persons stress level means in any random situation is something that is not currently known. It would require measuring stress over a wide variation of situations, for various types of people, to understand what correlations exist. So the research has narrowed a lot. A researcher might do a study to see if increases in stress measurements that are uncorrelated with activity level are correlated with inattention while driving. That's one possible study out of billions. Most research is done by grad students and they want to do something with impact, so many of the more mundane experiments will never get done, unless some kind of national lab is formed to systematically carry them out. There is no such lab yet for wearable data that I know of (though a few have tried to get funding for one). The manufacturers of the wearables are still working on improving the accuracy of the data (it's tough to get reliable sensor readings on a moving, active person, and nothing holds to skin well). So improvements in sensor accuracy will help, and then a more systematic evaluation of what can be inferred from that data will become more common. Some of it is happening, but it's a big job that is going to take decades (unless we start training a lot more grad students. I feel like much of the population of the world is going to waste, just trying to survive, when they could be trained and doing research.)

Or maybe we can just feed tons of raw data from everyone into an AI and it will Give Us The Answers (TM). But places like Apple and Google have had data like this for a decade and don't seem to have come up with much yet. Most wearables can't even reliably tell when a person is sleeping (though some can, I've had good luck with the Samsung G5, and the Oura Ring works if your sleep times are regular. The Mi Band 7 is pretty hit or miss.)

Here are the web pages of a couple of researchers working in this area to give you an idea of what current research looks like:

https://destrin.tech.cornell.edu/

https://people.cs.umass.edu/~dganesan/

Look at the proceedings of the MobiSys, SenSys, Ubicomp conferences to see more current research. Some of it is showing up in medical journals as well, so try a PubMed search on "health AND wearables". Moving something from a research demo to a product that a billion people can use and that clinicians will trust is not easy. I believe wearables are finding a place in physical rehabilitation.

I was recommended a HR monitor (Aktiia) by my cardiologist who then has been a bit overwhelmed by the voluminous pdfs they push out.

I have finally parsed out the data which I thought would be great, and have hit this very impass - basically now what?

I am going to try to provide this munged data to the cardiologist but it would be interesting to get any ideas on how to make it useful.

Or is the point that nothing is comparable unless I am sitting still each time?

Machine Learning can be of some use in analyzing all this data. However with ML there are usually false positives, false negatives, and accuracy below 100%. If the ML says you have a 61% chance of having had a heart problem, is that meaningful to anyone? How many false alarms or missed alarms can be tolerated? These things already happen with physicians, but diagnosis is a process, and they can keep evaluating over time. ML could also analyze long term data, but that means it could take a decade to gather and train it, and it might only be valid for the people it is trained on. I'm sure we'll figure some of this out eventually, but it takes RCT's and some cleverness to figure out how to turn data into information. Also people vary a lot, and it is increasingly apparent that one treatment or diagnosis method doesn't fit all people. It's a tough set of problems. That doesn't mean no one should work on it. I think the areas where the most progress will be made soon are likely to be those areas where there currently is no good diagnostic, or when the risks of a wrong answer are less harmful.

You can try to make some sense of the data yourself. You know what you are doing at any one time, and can probably remember what you did during the current day. So you know when you were walking, sitting still, driving, etc. If there was a period of elevated heart rate and you were sitting still typing on a computer what else happened then that might affect your heart rate? Food you were eating? Stress from people around you? You can look for patterns where your HR changes when something in the environment changes. Physicians don't have a good idea of how environment affects people. They'll ask general questions like "is anything bothering you lately?" or "how are you doing at work?" but they don't see what each moment of your day was like. But you yourself know that. So perhaps HR data would be useful for looking for environmental influences on health. Until someone tries it, nobody knows.

I suspect it might be more useful tying this to accelerometer data form my phone etc. I will get the code into some decent shape over next few lunchtimes and see how to get historic accelerometer data / location data - it must be feasible somehow.

However I suspect that Apple already has 50 people working on it

Sounds like yet another way in which doctors are inferior to machine learning algorithms: A log of the patient's activities and symptoms, or some kind of daily journal would also be nice, but there's a lot of predictive anomaly detection algorithms that would do just fine with the raw data, finding patterns of activity and rest and adjusting for them.

There is definitely a world in which various sensors send data somewhere and a thing crunches the numbers and figures out you have a trend to somewhere bad. It's just the doctors were trained in a world before sensors, and they're understandably wary of people googling symptoms, and sensors aren't that good, and the software is limited, and we as regular humans aren't used to understanding this sort of information about ourselves.

The only question is: are we going to go back to a world without wearables, and the potential benefits and drawbacks they bring? Or are we going to improve them and adapt ourselves to them?

While I'm sure a wearable could be occasionally useful or even save a life, most don't do much. It's telling that mainstream fitness wearables tout their data collection ability but never use this data in marketing to prove they make people healthier. I think most fitness wearables are cashing in on pseudo medical device branding as a way to charge hundreds of dollars for electronics that cost very little to manufacture. I'd be wary of those kinds of devices if I were a doctor.

Both whoop band and Apple Watch + https://bevel.health plus a bunch of other fitness apps, give awesome abilities to folks, to improve their sleep, balance their energy cycle, reduce injuries by knowing when to rest more for recovery.

- Sleep

- Properly recovering each day for Consistent Cardio

- Stress minimisation with Tracking Triggers/Causes, Foods

These things alone are monumentally impactful in increasing longevity and health, more than any current available medicine, or therapy. Just these 3 factors alone.

And fitness bands combined with effective apps that make use of these metrics, do a pretty darn good job already to improve all these 3 critical factors that allow you to

- Live a healthier day

- Perform better at work

- Identify and mitigate causes that hurt your sleep (bad sleep -> bad day)

- Balance exercise with recovery

- Managing stress levels and making healthier choices

The sleep monitoring in particular is complicated: monitoring and especially alerting on sleep quality can exacerbate insomnia.

strictly speaking, I think apnea monitoring as done on the Apple Watch is pretty non-invasive and not too bad.

One thing that ended up being really noisy and irritating was a continuous glucose monitor. It turns out it's possible to stick those things in you in a way that can false alarm blood sugar drops during your sleep, at which point your CGM app freaks the hell out and wakes you up.

The Ringconn Gen 2, supposedly can detect sleep apnea. https://ringconn.com/products/ringconn-gen-2

The reviews I've read/watched about wearables ability to track sleep patterns seem to show great variability between what one wearable would say versus another.

Sleep apnea detection is mostly about detecting changes in blood oxygen and paused breathing, so if you can do that, you can diagnose sleep apnea.

The actual medical protocol is sleeping a night with a pulse oximeter on.

> most don't do much

Directly, maybe, not. But people love games and wearables gamify fitness. I think that's a net positive. My mother looks at her Apple Watch step count daily.

Are they wary of wearables? Whenever I go to a specialist these days, I usually see an Apple Watch on their wrist.

The premise of the story is that doctors are worried that people are over-monitoring raw health details, not that doctors are worried that OLED screens are replacing jeweled watch movements.

> would the data from my wearable have helped healthcare professionals with my treatment?

This presumes doctors practicing medicine 2.0 have any effective treatment. For the flu, sure there is tamiflu; or COVID, paxlovid; but absent those 2, for the short term interference; the best options are somewhere between zinc lazenges, creatine, water, and sleep.

Both the flu and COVID now have vaccines, which are more effective and preventative than interfering treatments. Anti-biotics should, generally, rarely be given at early stage, unless there are significant factors, as they have both individual and population level side effects.

Now, can a glucose monitor or testing cortisol or liver proteins more often tell you something? Yes, but not with much more actionable items than eat healthy, work out, sleep more, drink less (Yes, there is hrt, glp1s - but those are unlikely to be prescribed if the issue is not significant over a longer term)

>This presumes doctors practicing medicine 2.0 have any effective treatment. For the flu, sure there is tamiflu; or COVID, paxlovid;

don't know if you were being intentionally sarcastic when naming those two as effective treatments

>How effective is Tamiflu? Unfortunately, the effectiveness of Tamiflu is marginal, as it cannot “cure” the illness. Most studies have shown that the medication will reduce the length of symptoms by only 12 to 24 hours, and if started after two days of symptoms, it does not help at all. It is important to keep effectiveness in mind, as expectations for Tamiflu are often very high and, frankly, overhyped.

https://health.mountsinai.org/blog/is-it-worth-it-to-take-ta...

>Paxlovid does not significantly alleviate symptoms of COVID-19 compared with placebo among nonhospitalized adults, a new study published on April 3 in The New England Journal of Medicine found.

The results suggest that the drug, a combination of nirmatrelvir and ritonavir, may not be particularly helpful for patients who are not at high risk for severe COVID-19. https://www.medscape.com/viewarticle/study-shows-nirmatrelvi...

which links to https://www.nejm.org/doi/full/10.1056/NEJMoa2309003 and this little nugget

>CONCLUSIONS

The time to sustained alleviation of all signs and symptoms of Covid-19 did not differ significantly between participants who received nirmatrelvir–ritonavir and those who received placebo.

the answer, as always, is hypochondria. i used to be a hypochondriac and it is crazy how much more im seeing my peers have hypochondria

I know a few folks who tend towards hypochondria, curious what changed for you? I feel like the people I know who I'd consider hypochondriacs^, are going to be like that forever.

^I'm not a doctor

Personal anecdote from knowing a few cases. It seemed to result from too much time alone and ruminating. Getting out and doing stuff with people made their hypochondria fade into the background.

Author should have induced a artificial ahead of time fever. Sauna, hot bath, insulation (bed rest well wrapped). Go all in, guns blazing before it goes all out.

Doctors are obsessed with job security

Doctors don't worry much about job security. Unemployment rate for licensed physicians is ~0% and demand is growing faster than supply. Greater use of wearable devices will only create more work for them, not less.

Do you know doctors IRL?

Yes

because they are gimmicks and its fair to be wary of gimmicks

I don't think so. I upgraded my Fitbit to a Fitbit 5 when I started getting arrhythmias a few years back. I used the ECG capture whenever I felt something was wrong. As the irregular beats graduated into episodes of what the medics identitied as SVT (due to a worsening thyroid, as as it transpired at a later point), I captured each trace. I brought the printout of the PDF to the GP, who ordered some blood tests and referred me to cardiology. The cardiologist wanted me to email the PDF traces too as a starting point. Nobody treated it as a gimmick; in the early stages it was the difference between data and no data, because the episodes had been unpredictable.

Reply on news.ycombinator.com