I have historic air pressure line graphs and I have superimposed the dates on which I have suffered migraine attacks. I can see that my migraines are most likely to occur around 3 days after a sharp drop in pressure of more than 15 mBar over 3 days and sometimes about 1 day after a very large rise in pressure if it is more than about 30mBar over 3 days.

What is the best statistical test to apply to this to prove there is a connection?

I can see that I may have to change the barometric values into rates of change over time. I may also have to factor in that changes while pressure is low may be more likely to trigger attacks tban changes while it is high.

Also I may have to filter the fluctuations for frequency... Maybe I need to take say a 3 day running average from the barometer graph to put into the statisical test, or possibly start taking readings more often than once a day.

I am hoping to be able to reliably predict my migraines and then hopefully prevent them by taking medication only at the 'right time' instead of taking them more often than I need to.

  • $\begingroup$ If you search the medical literature for air pollution and migraine you can find articles which use mdoels with a variety of lags. sciencedirect.com/science/article/pii/… is an example and is not behind a paywall. You might be able to import some of those ideas into your case. I know you are interested in pressure rather than pollution but that should not matter. $\endgroup$ – mdewey Jan 25 at 13:40
  • $\begingroup$ Please be aware there does not exist a valid statistical test that uses the same data you have employed to construct your hypothesis. Thus, any test that might be recommended here has to be applied to future values. That can still be problematic because the subject is not blinded to the experiment! Regardless, it is important that you specify your hypothesis as clearly as possible: the many "maybes" and "hopes" in this question leave us wondering what that hypothesis might be. $\endgroup$ – whuber Jan 25 at 15:26

It’ll be hard to prove this, but kudos to you for being creative!!

Not sure how much data you have and I’m not confident there is a simple test for this but you can definitely make progress here.

What we want to know are two statistics:

Sensitivity: what fraction of your migraines occur after a pressure change of X

Specificity: What fraction of pressure changes precede a migraine.

If both statistics are high then you may be onto something.

As for statistical tests — there are bootstrapping and resampling approaches one could use but they require a computer — no nice formulas.


Not the answer you're looking for? Browse other questions tagged or ask your own question.