5 ms·

This is a common concern in virtually all clinical studies and it's unlikely the researchers just ignored this. In principle these studies attempt to decorrelat

by joshvm 5y ago

This is a common concern in virtually all clinical studies and it's unlikely the researchers just ignored this. In principle these studies attempt to decorrelate confusing factors; from the abstract: > We used Cox proportional hazards models to estimate the associations between coffee/tea consumption and incident stroke and dementia, adjusting for sex, age, ethnicity, qualification, income, body mass index (BMI), physical activity, alcohol status, smoking status, diet pattern, consumption of sugar-sweetened beverages, high-density lipoprotein (HDL), low-density lipoprotein (LDL), history of cancer, history of diabetes, history of cardiovascular arterial disease (CAD), and hypertension. It's extremely difficult to do this in practice because people lie, there are other factors they might have missed and so on.

tpoacher 5y ago
The controls give you information about confounders, but it still says nothing about the causal direction. Typically the easiest way to turn it around is to switch where the authors chose to place the word "risk" and see if the new statement (with some appropriate changes, and adding "predisposition" to account for future events) still makes sense. E.g. Predisposition to dementia lowers lifetime "risk " of coffee consumption.
- sharperguy 5y ago
  The interesting thing is that assuming that there is a confounding variable that correlates with coffee consumption which is the real cause of the effect, the correlation can still be useful in some contexts. For example, while that would mean that simply increasing coffee consumption wouldn't directly lower your risk of dementia and stroke, we can improve our assessment of the risk of a certain patient developing these conditions by asking whether they consume coffee. Even though the coffee isn't the cause, since it correlates with the cause, we still get information by doing that.

by joshvm 5y ago

tpoacher 5y ago

The controls give you information about confounders, but it still says nothing about the causal direction. Typically the easiest way to turn it around is to switch where the authors chose to place the word "risk" and see if the new statement (with some appropriate changes, and adding "predisposition" to account for future events) still makes sense. E.g. Predisposition to dementia lowers lifetime "risk " of coffee consumption.

sharperguy 5y ago
The interesting thing is that assuming that there is a confounding variable that correlates with coffee consumption which is the real cause of the effect, the correlation can still be useful in some contexts. For example, while that would mean that simply increasing coffee consumption wouldn't directly lower your risk of dementia and stroke, we can improve our assessment of the risk of a certain patient developing these conditions by asking whether they consume coffee. Even though the coffee isn't the cause, since it correlates with the cause, we still get information by doing that.