4 ms·

Worth noting case control studies like this have inherent bias because your sample is of very sick people. It’s like saying out of all car accidents a dispropor

by madrox 6y ago

Worth noting case control studies like this have inherent bias because your sample is of very sick people. It’s like saying out of all car accidents a disproportionate amount of them were Uber drivers. Being an Uber driver won’t get you in an accident. Uber drivers just drive more. This is correlation at best. That said, this was also the type of study that found the original link between smoking and lung cancer. So while being vitamin D deficient may not cause me to end up in the hospital if I get COVID, I’m going to be taking a supplement just in case.

headmelted 6y ago
Is it though? An Uber driver is likely to spend more time in a car, and therefore more likely to be in a car accident. Is a person more likely to spend time around sick people if they’re Vitamin D deficient? Yes it’s correlation, but correlation specifically without known other causes (Unlike the Uber example).
- madrox 6y ago
  Yes, that’s the point of the Uber example, and while it’s trivial to realize what the hidden correlation is in this case it’s pointing out there could be other hidden correlations in this case we don’t automatically know. For example, vitamin D deficiency can be a symptom of an underlying issue that is actually what causes sicker patients. Or what’s even more likely, vitamin D deficiency is more prevalent among elderly and African Americans, who have been disproportionately hit harder by COVID because they’re more vulnerable to disease in general (former) or are more likely to be unable to isolate due to being an essential worker (latter). And before we start debating these specific possibilities, the point is to illustrate we can’t just assume more than the sampling method dictates. Different studies are needed.
  - headmelted 6y ago
    Gotcha, with you now. Now that you’ve phrased it this way I see your point and I’ve actually been thinking this myself for a while now. Also, you wouldn’t believe the amount of conversations I’ve had with people who are convinced it “only kills old people” who’s eyes glaze over when I explain that there are other outcomes besides living and dying. Everyone in the world seems to believe it’s a magical death lottery that instantly switches the lights off for 0.01% of people who catch it, and the other 99.99% of people are therefore completely fine. The whole world’s gone mad.
k_bx 6y ago
It’s not “correlation at best”. It’s an almost canonical example of the Bayesian error: given a reverse probability (if you’re sick => you’re 80% likely to have deficiency), people make a reverse conclusion (if you’re deficient => you’re 80% sick) In reality (due to Bayes theorem), we need to compute the probability of being sick when having deficiency as population_with_covid (1/47 for Spain) * 80% / population_with_deficiency. If, say, there’s 60% vitamin D deficient people in Spain, it’d give us 2% chance of getting a disease if you have deficiency. Not much!
- Jugurtha 6y ago
  I'd argue it's not even a Bayesian error but a logic error stemming from difficulty with contraposition.[0] It checks out even if you remove the probability aspect of it and use a simple implication. - [0]: https://en.wikipedia.org/wiki/Contraposition https://en.wikipedia.org/wiki/Contraposition
- faeyanpiraat 6y ago
  I believe you are wrong about what we want to calculate. It's not about whether you have a higher chance of getting infected if you are vitamin D deficient. It's about whether you need medical attention in a hostpital if you do. So if we have X people who got COVID, and out of those X people Y% needed to spend time in a hospital, and 80% of them had D vitamin deficiency, then it does not matter what is the Y% value. The important thing is that the more severe cases had D vitamin deficiency, therefore you have a higher chance to get a severe case if you are deficient. Right?
  - 542354234235 6y ago
    > So if we have X people who got COVID, and out of those X people Y% needed to spend time in a hospital, and 80% of them had D vitamin deficiency, then it does not matter what is the Y% value It does matter because you need to know what the percent that didn’t need to spend time in the hospital. For instance, if 80% of the people that did not have to be hospitalized also had a Vitamin D deficiency, then there is nothing meaningful about the deficiency because it is the same between the two groups. I could say pretty confidently that 80% of people hospitalized for COVID have brown eyes. Why? Because the overall population is about 80% brown eyed. We need to look at the difference first and foremost. > The important thing is that the more severe cases had D vitamin deficiency, therefore you have a higher chance to get a severe case if you are deficient. Even if we found a large difference, that is far too little information to assume causation. For instance, it could be that the elderly are more likely to be Vitamin D deficient due to being sedentary and spending most of their time indoors, and the elderly are more likely to be hospitalized due to their age. So it wouldn’t be Vitamin D deficiency, it would be age that was the underlying cause of both phenomenon. We need to be very careful about making these leaps in logic, since it can send us down pointless rabbit holes and have people spending time and resources on things that do not actually help.
rhacker 6y ago
It's really not worth noting. There's a real link between vitamin D's usefulness that's part of the broader picture which is why vitamin D articles are getting increasing attention not just in hacker news. Your comment would definitely be worth noting if this was an article siting that many very sick covid patients have had multiple marriages (because they're old).

by madrox 6y ago

headmelted 6y ago

Is it though? An Uber driver is likely to spend more time in a car, and therefore more likely to be in a car accident. Is a person more likely to spend time around sick people if they’re Vitamin D deficient? Yes it’s correlation, but correlation specifically without known other causes (Unlike the Uber example).

madrox 6y ago
Yes, that’s the point of the Uber example, and while it’s trivial to realize what the hidden correlation is in this case it’s pointing out there could be other hidden correlations in this case we don’t automatically know. For example, vitamin D deficiency can be a symptom of an underlying issue that is actually what causes sicker patients. Or what’s even more likely, vitamin D deficiency is more prevalent among elderly and African Americans, who have been disproportionately hit harder by COVID because they’re more vulnerable to disease in general (former) or are more likely to be unable to isolate due to being an essential worker (latter). And before we start debating these specific possibilities, the point is to illustrate we can’t just assume more than the sampling method dictates. Different studies are needed.
- headmelted 6y ago
  Gotcha, with you now. Now that you’ve phrased it this way I see your point and I’ve actually been thinking this myself for a while now. Also, you wouldn’t believe the amount of conversations I’ve had with people who are convinced it “only kills old people” who’s eyes glaze over when I explain that there are other outcomes besides living and dying. Everyone in the world seems to believe it’s a magical death lottery that instantly switches the lights off for 0.01% of people who catch it, and the other 99.99% of people are therefore completely fine. The whole world’s gone mad.

k_bx 6y ago

It’s not “correlation at best”. It’s an almost canonical example of the Bayesian error: given a reverse probability (if you’re sick => you’re 80% likely to have deficiency), people make a reverse conclusion (if you’re deficient => you’re 80% sick) In reality (due to Bayes theorem), we need to compute the probability of being sick when having deficiency as population_with_covid (1/47 for Spain) * 80% / population_with_deficiency. If, say, there’s 60% vitamin D deficient people in Spain, it’d give us 2% chance of getting a disease if you have deficiency. Not much!

Jugurtha 6y ago
I'd argue it's not even a Bayesian error but a logic error stemming from difficulty with contraposition.[0] It checks out even if you remove the probability aspect of it and use a simple implication. - [0]: https://en.wikipedia.org/wiki/Contraposition https://en.wikipedia.org/wiki/Contraposition
faeyanpiraat 6y ago
I believe you are wrong about what we want to calculate. It's not about whether you have a higher chance of getting infected if you are vitamin D deficient. It's about whether you need medical attention in a hostpital if you do. So if we have X people who got COVID, and out of those X people Y% needed to spend time in a hospital, and 80% of them had D vitamin deficiency, then it does not matter what is the Y% value. The important thing is that the more severe cases had D vitamin deficiency, therefore you have a higher chance to get a severe case if you are deficient. Right?
- 542354234235 6y ago
  > So if we have X people who got COVID, and out of those X people Y% needed to spend time in a hospital, and 80% of them had D vitamin deficiency, then it does not matter what is the Y% value It does matter because you need to know what the percent that didn’t need to spend time in the hospital. For instance, if 80% of the people that did not have to be hospitalized also had a Vitamin D deficiency, then there is nothing meaningful about the deficiency because it is the same between the two groups. I could say pretty confidently that 80% of people hospitalized for COVID have brown eyes. Why? Because the overall population is about 80% brown eyed. We need to look at the difference first and foremost. > The important thing is that the more severe cases had D vitamin deficiency, therefore you have a higher chance to get a severe case if you are deficient. Even if we found a large difference, that is far too little information to assume causation. For instance, it could be that the elderly are more likely to be Vitamin D deficient due to being sedentary and spending most of their time indoors, and the elderly are more likely to be hospitalized due to their age. So it wouldn’t be Vitamin D deficiency, it would be age that was the underlying cause of both phenomenon. We need to be very careful about making these leaps in logic, since it can send us down pointless rabbit holes and have people spending time and resources on things that do not actually help.

rhacker 6y ago

It's really not worth noting. There's a real link between vitamin D's usefulness that's part of the broader picture which is why vitamin D articles are getting increasing attention not just in hacker news. Your comment would definitely be worth noting if this was an article siting that many very sick covid patients have had multiple marriages (because they're old).