I read with interest this interesting study and the commend the authors for their detailed work. However, I do feel ultimately the conclusions are wrongheaded. Firstly, and most importantly, it is impossible to account for the underlying indication for the steroids, and there will be clear confounding by indication. Despite the apparent matching, clearly patients prescribed steroids for URTI are much more likely to have underlying COPD or asthma, and so this is not a true comparison. It is also clear from the demographic group that these patients are older, and more comorbid.
Secondly, there doesn't appear to be a clear mechanism for short term steroids causing any of these conditions outside sepsis, which the authors accept, although there is a clear mechanism from the underlying condition (i.e. fall due to pain from arthritis, confusion from LRTI). Again, it would seem silly to attribute this to the steroids without any clear mechanism.
Thirdly, if we really felt steroids were the cause of these adverse events, we would expect a dose-response relationship, which is clearly not evident from these results.
Finally, even if we feel the above hypotheses to be true, the numbers needed to harm are very high: using the web appendix to calculate numbers needed to harm (NNH): I calculate NNHs for most adverse events to be in the high hundreds or thousands. Steroid use is clearly common, so even high NNHs are important, but given the huge confounding by indication, the NNH are likely to be even higher in the real world.
In summary, I'm not convinced there is any strong theoretical evidence supporting steroid use causing any of these adverse events except sepsis, where the NNH is 5000. There is no dose response relationship evident, and there is clearly strong confounding by indication in this study, which makes the results difficult to interpret with any certainty.
I'd struggle to not prescribe steroids to patients with an exacerbation of COPD (NNT 9 to prevent treatment failure, from a Cochrane review (1)) based on this very limited evidence.
Competing interests: No competing interests