A bit of a shaggy dog story for those not statistically inclined - which includes ME - or perhaps for those who ARE statistically inclined - you know who you are:
I got into a situation at w*rk many years ago. My lab produced data for two different internal customers. Each customer looked either 'good' or 'bad' based on their particular chunk of data (one group produced a chemical, the other group refined the material - for want of a better description of the 'issue') My lab determined each customer's "contribution" to the final result. The "ratio" of the two numbers was always thrown up to me as "proving" we were wrong - it just depended on which customer looked bad as to which one used this tired attack on me (my lab.)
So one customer always looked bad when the other looked good. Therefore, they each blamed me (my lab) when they looked bad. Management knew the little game they were playing but they had a dog in the hunt as well. SOMEBODY had to be the bad guy - and no one wanted it to be them. (You know what they say about what runs down hill.)
I finally got tired of it and brought in a statistician to go through several thousand data points (dating back 10+ years.) She confirmed what I had been saying all along. Comparing two numbers which each had a fair amount of variability meant that the manipulation of these two numbers together led to an even more perverse comparison. IOW the data manipulation my customers were doing was bogus.
By the way, no one was happy with this outcome - except me. Now no one could point fingers at my lab or anyone else. An entire bi-monthly meeting of peons and management was permanently canceled after I published the statistical findings. I was no longer "it" or "piñata" in front of management every couple of weeks.
"Why bring up this dreary subject of a w*rk situation?" I hear you ask. My point is that the IFR is a ratio of two numbers which arguably have some "slop" in them. Even with the very purist of motives, there is error in every calculated number (either cases or deaths). Put them together in a ratio and the errors can become perverse. I'm not saying that has happened but I know it is possible - even when using state-of-the-art technology to produce each individual number (numerator or denominator.)
I cast no stones but only suggest a deeper dive into the data and elucidation of any error-bar information NOT shown in this treatise. YMMV