4 ms·

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 we

by rapjr9 2y ago

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.

godelski 2y ago
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
- rapjr9 2y ago
  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://destrin.tech.cornell.edu/ https://people.cs.umass.edu/~dganesan/ 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.

by rapjr9 2y ago

godelski 2y ago

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

rapjr9 2y ago
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://destrin.tech.cornell.edu/ https://people.cs.umass.edu/~dganesan/ 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.