[m12k]

The question is, will all these data aggregated together actually give us any actionable data that we can and will act on? If there's one thing Google Analytics has taught me, it's that it's quite possible to have a lot of detailed data at your fingertips, but still not be able to answer any valuable questions with it.

[alwillis]

I’m using the Watch and with my iPhone. The Health and Activity apps work great together.

The visualizations are clear and understandable. For example, I’m significantly less active this spring than last spring—average of 2.2 miles walking per day in 2019 vs. about 1 mile in 2020.

That’s mostly due to stay at home orders due to COVID-19 the past 3 months. Once the weather got better and the number of cases subsided, I started taking walks in my neighborhood due to the actionable data I had.

Remember, the Watch and the iPhone can do machine learning on the device.

What’s missing is an easy way to do more advanced health data processing on my Mac if I wanted to.

I’d like my public health department release an app for COVID-19 exposure notification, using the Apple/Google API.

You could also envision broad clinical trials of treatments for COVID-19 using these components.

[paulcole]

What valuable question did you answer with the data? You didn’t need a watch to know you were walking less.

> Remember, the Watch and the iPhone can do machine learning on the device.

Great, what is the machine learning?

[alwillis]

It’s not just the amount of walking; it was also calories burned, heart rate not getting into an aerobic range, my VO2 max and other data that I’d have no idea about if I didn’t have a device monitory this during my waking hours.

Regarding on device machine learning: https://developer.apple.com/machine-learning/

[paulcole]

Neat! You managed to not answer either of my questions:

1. What valuable question was this data answering?

2. What is the machine learning?

[clairity]

you should never have stopped walking. the chances of covid infection outside on a walk are essentially zero.

this is one of the many little harms of blanket stay home orders rather than just distancing (with masks only when you can’t), where you get nearly all of the benefits with nearly none of the downsides.

[alwillis]

I live in New England; in addition to the stay at-home orders, it was also cold and rainy much of the spring, in addition to living in a state with the third most infections and deaths from COVID-19 in the country during March–May.

As of today, we’ve had 8,860 confirmed cases of COVID-19 in my county alone.

I have pre-existing conditions and live with a septuagenarian that clearly made it not worth the risk to be out and about.

[clairity]

i completely sympathize with the need to be cautious in the face of elevated risk, but taking a walk outside doesn't by itself present additional risk. the virus doesn't exist ambiently in the air. you need an infected person breathing into your face for some period of time for the risks to accumulate enough to successfully transmit the disease.

[nbashaw]

Totally agree! It's going to take a whole new sub-specialty of ML to make any sense of it. But I think there's a positive feedback loop where more data → better algorithms → more data, because more people opt into the system in order to get access to the algorithms.

[m12k]

I think there might be an even bigger problem in coming up with any kind of response that we didn't already know. E.g. the advice "Don't smoke, stop eating before you're full, eat more vegetables, eat less bread and pasta, drink less alcohol, prefer water instead of drinks like juice, milk or soda, stop eating processed sugar, avoid saturated fats, get regular cardio, avoid stress, get enough sleep and do so at roughly the same time every night" is likely to be correct for the vast majority of people. Having top-of-the-line ML crunch a bunch of data only to spit out the exact same recommendation to everyone doesn't really accomplish much. Maybe more specific recommendations would be possible in a decade or two if we start having detailed data to look at, but considering how hard a time we've had just answering simple questions like 'are eggs good or bad for you?' in general, I'm not really all that optimistic about our ability to make precise individual recommendations that are not already obvious.

[nbashaw]

Generally I think you're right. But also I think there are a ton of edge cases we could discover based on complex interactions of genetics, environmental factors, disease history, etc.

Also there's a lot more urgency when an algorithm detects an anomaly and asks you to schedule a blood test!

[ralph84]

Do you really need ML to tell you to eat better and exercise?

[SirHound]

No but you might to pull together a bunch of noisy data points until a confident prediction of early stage serious disease.

[alwillis]

This happens relatively frequently actually.

Guy buys an Apple Watch and gets notified of an arrhythmia, sees his doctor and he averts a potentially serious incident: https://www.medicaldevice-network.com/news/apple-watch-atria...