Three Studies, Three Numbers, No Contradiction

The arithmetic, stated plainly

A spontaneous reporting database contains reports, not patients. For any drug-event pair you can build a two-by-two table: reports of this event for this drug, reports of other events for this drug, reports of this event for everything else, reports of other events for everything else. The reporting odds ratio is the cross-product of that table. The proportional reporting ratio is a closely related quantity built from proportions rather than odds; in practice the two usually agree closely, and where they diverge it is a sign the counts are small. Neither is a rate. There is no denominator of people exposed anywhere in the calculation, so no incidence can be recovered from it.

The same question, three comparators, three answers

Consider acute kidney injury and semaglutide. Gandhi and colleagues compared tirzepatide against semaglutide directly in FAERS reports from January 2022 to September 2025, and found a reporting odds ratio for tirzepatide versus semaglutide of 0.44 (95% CI 0.38 to 0.50). Shokr and colleagues compared each drug against the rest of the database and reported semaglutide with a PRR of 1.25 and ROR of 1.25 for acute kidney injury. Makhmutov and Qureshi restricted to non-diabetic patients and looked at ketoacidosis, finding semaglutide at ROR 3.15 (95% CI 2.78 to 3.56). These are not three estimates of one quantity that happen to disagree. They are three different quantities.

Why the drug-versus-drug comparison is the sharpest and the most fragile

Comparing one drug against a single named drug rather than against the whole database removes a lot of noise: two agents used for the same indication in similar populations share many of the biases that distort spontaneous reporting. That is why Gandhi's design is the informative one for the question 'do these two differ'. It is also the most fragile, because it inherits every difference between the two drugs' reporting environments - how long each has been marketed, how much media attention each has had, which countries report them, and which one is likelier to be named as primary suspect when a patient takes both. A ROR of 0.44 says tirzepatide reports were less likely to mention acute kidney injury than semaglutide reports. It does not say a patient's kidneys are safer on one than the other.

The absolute numbers are the ones worth quoting

Gandhi's analysis covered 133,872 reports - 92,807 for tirzepatide, 41,065 for semaglutide - and acute kidney injury appeared in 432 tirzepatide reports (0.47%) and 440 semaglutide reports (1.07%). Those percentages are the proportion of reports mentioning the event, not the proportion of treated people experiencing it, and the distinction is not pedantic: the reporting fraction for serious events is generally low and unknown, so the true frequency in treated people is unmeasured in either direction. What the numbers do establish is that acute kidney injury is an uncommon thing to find in a report about either drug, and the authors say as much.

What a signal is actually for

A disproportionality signal is a prompt to look, not a finding. It says that a drug-event pair appears together more often than the background pattern would predict, which is a reason to design a study with a denominator - a cohort, a case-control, a self-controlled series - and answer the question properly. Treating the signal itself as the answer inverts the purpose of the system. Every one of the three papers here says so in its own limitations section, and every piece of coverage that quotes the ratio without the comparator strips that out.

The one-line test

If someone quotes a reporting odds ratio to you and you cannot immediately say what it was compared against, over what period, and in which subpopulation, the number has told you nothing. That is not a criticism of the method. It is the method: the ratio is defined relative to a reference group, and the reference group is a choice.

Quick reference