The standard line from scientists is correlation does not equal causation, but that seems to have been forgotten by the authors of this paper: ‘Birth Month Affects Lifetime Disease Risk: A Phenome-Wide Method’.

I came across this article in The Age (paywalled), but it has also been picked up by that respected scientific outlet Cosmopolitan.

Apparently data mining a large electronic database and using logistic regression on 1700 diagnosis codes (SNOMED-CT) combined with a Benjamini-Hochberg adjustment for multiple comparisons is a new procedure that deserves a new name SeaWAS: Season-Wide Association Study. This is a bit of a stretch, but let’s run with it.

So what does the procedure amount to? Individuals with the SNOMED-CT code were extracted, along with 10 individuals without the SNOMED-CT code (but still from the same database), and it appears that a logistic regression was then fit, with birth month (as an integer) as the predictor variable. SNOMED-CT conditions that had a p-value less than the Benjamini-Hochberg cutoff were deemed significant.

Unfortunately there’s not much more detail than that, but we can guess that this procedure is likely to lead to spurious results (as I think it does, more later): it would seem that the p-value referenced is likely the Wald p-value for the linear birth month term—but so what? What makes us think that birth month (on the logistic scale) should be linearly related to the SNOMED-CT condition?

Anyway, that’s what was used, and they found 55 (55!) conditions dependent on birht month. And within these 55, 16 completely novel conditions were found. Completely novel meaning that no-one else had apparently published an association between birth month and these conditions. And when you have a look at these completely novel conditions it’s no wonder! Have a look at their Table 2, and you’ll find that the following SNOMED-CT conditions are related to birth month:

  • Nonvenomous insect bite
  • Venereal disease screening
  • Vomiting

What?!! Relevance? Looking at their seasonal pattern graphic in the same table, it looks like if you were born in October then you’re out of luck, you’re more likely to receive a nonvenomous insect bite…

Sorry to be the harbinger of doom.

References

Mary Regina Boland, Zachary Shahn, David Madigan, George Hripcsak, Nicholas P. Tatonetti. “Birth Month Affects Lifetime Disease Risk: A Phenome-Wide Method” Journal of the American Medical Informatics Association Jun 2015. DOI: 10.1093/jamia/ocv046

comments powered by Disqus