Getting yellow fingernails from holding many cigarettes is not uncommon. However, I guess that few people would intuitively buy the argument that yellow nails can cause lung cancer. But if asked if cell phones can cause cancer, then it is not all that intuitively clear.
Or how about coffee. A Google search of “coffee can give you..” results in 113.000.000 hits, so I guess you can set in whatever you like (or perhaps better not like), but luckily for coffee drinkers does the search term “coffee can prevent” give 150.000.000 hits.
Correlations are part of our everyday life, and it is so easy to be misleading.
I must admit that I have not searched the many hits in the Google searches for coffee for co-occurrence on both search, but I am not in doubt they are there.
Correlations are easy to calculate using the software included on most computers, or even using smartphones. Correlations can be a powerful tool to find meaningful associations, but they never demonstrate causal connection and they are very easy to misinterpret if not cautious. That yellow nails and lung cancer are easy to dismiss since a more logical explanation is obvious. However, many topics in medicine and healthcare are much harder to have an intuitive feeling about. Just take a look at discussions concerning vaccines or statins.
To make it even more complicated does a relevant medical correlation not necessarily imply that the alternative is better. Increasing body weight is associated with increased risk for diabetes, but that does not prove losing weight reduces that risk. Luckily we have clinical trials using randomization demonstrating that lowering body weight, in fact, decreases the risk for diabetes. Correlations can be used to generate ideas and hypothesis, but must be followed by proper conducted clinical trials. Following correlations alone would lead to healthcare running in all directions at the same time, and find it difficult to maintain an overview. Which unfortunately often is the case, partly due to many news headlines reporting correlation analysis.
The power of correlation is often worth looking at, since the stronger the correlation, the more likely the finding is to be relevant. Like smoking and lung cancer. But high correlations does not prove causality. As an example is it possible to calculate a very strong! Correlation between the number of iPhone in use and the drop in deaths from cardiovascular disease. However, it is hard to believe that there should be any causal relationship.
So – as always – we need to look at the data, not just the result
Analysis including many persons makes it much easier to find “significant” correlations. In fact, you can set up a computer to analyse data in multiple ways to find just something (while you take a cup of coffee). With increased availability of “big data”, including information from wearables, we will, without doubt, see more and more of this kind of analysis, and we must subsequently be very careful with these results.
This makes the detailed analysis of data and results even more valuable