The short answer
Disproportionality measures such as PRR and ROR describe how often an adverse event appears in a spontaneous reporting database relative to other events, not how often it occurs in people. They detect signals for investigation. They cannot estimate risk, because the database has no denominator.
Key facts
- PRR
- Proportional reporting ratio
- ROR
- Reporting odds ratio
- Source
- Spontaneous reporting databases (FAERS, Yellow Card)
- What they measure
- Disproportionate reporting
- What they cannot measure
- Incidence or risk
- Missing element
- A denominator (how many people took the drug)
- Correct use
- Signal detection, prompting further study
What a spontaneous reporting database is
FAERS in the United States and the Yellow Card scheme in the UK collect reports of suspected adverse reactions submitted voluntarily by clinicians, patients and manufacturers. They are enormously valuable for detecting rare events that trials are too small to find. They are also, by construction, a collection of reports rather than a study. Nobody was enrolled, nothing was randomised, and no one counted how many people took the drug without incident.
What PRR and ROR actually compute
Both compare how frequently a particular event is reported for a particular drug against how frequently it is reported for everything else in the database. A high value means this event is disproportionately represented among reports for this drug. That is a statement about the composition of a report collection. It is not a statement about how many people experienced the event.
Research material referenced
Retatrutide 10mg, third-party HPLC tested
The missing denominator
To calculate a risk you need to know how many people were exposed. A spontaneous database does not know this. If a drug is taken by ten million people and generates two hundred reports of an event, and another is taken by ten thousand and generates twenty, the crude report counts are wildly misleading, and the disproportionality measures do not fix that. They were never designed to.
Reporting is not random
Several biases push in the same direction. Newly launched drugs attract more reporting than established ones. Media attention to a possible association increases reporting of that association, sometimes dramatically. This is known as notoriety bias, and it can manufacture a signal from publicity alone. Serious events are reported more than mild ones. None of these are flaws in the system so much as properties of it, and all of them affect disproportionality values.
A worked example
A disproportionality analysis reporting a PRR of 18.9 and ROR of 19.4 for liraglutide and acute pancreatitis is frequently repeated as though it meant a nineteen-fold risk. It does not. It means reports of pancreatitis were disproportionately represented among liraglutide reports relative to the rest of the database. Whether that reflects a real elevated risk, and of what magnitude, requires a cohort study with a defined population, which is what such a signal should prompt.
What signals are for
They are the first stage of a working pharmacovigilance system, not its conclusion. A signal triggers investigation; a cohort or case-control study with a real denominator estimates risk; a regulator weighs that against benefit and acts. The MHRA's February 2026 semaglutide update is the visible end of that pipeline. Note that it came with a frequency estimate, up to 1 in 10,000, which is the sort of figure disproportionality analysis can never produce.
Quick reference
| Disproportionality (PRR/ROR) | Cohort study | |
|---|---|---|
| Data source | Spontaneous reports | Defined population |
| Denominator | None | Known |
| Can estimate incidence | No | Yes |
| Can estimate risk | No | Yes |
| Vulnerable to reporting bias | Yes, strongly | Much less so |
| Correct role | Detect signals | Quantify them |
Frequently asked questions
Does a ROR of 19 mean a nineteen-fold risk?
No. It means the event was disproportionately represented among reports for that drug. Without a denominator, no risk can be calculated from it.
Are these databases useless then?
Far from it. They detect rare events trials are too small to find. The error is treating a detection tool as a measurement tool.
What is notoriety bias?
Publicity about a possible drug-event association increases reporting of that association, which can inflate disproportionality measures independently of any real change in risk.
Continue researching
Related guides and reference material.
Primary sources & clinical trials
Published papers, trial registrations, regulator pages and company announcements. Links open the original.
Written and published by UK Peptides, the trading name of AURYXRESEARCH LTD (company 17388727, registered in England and Wales).
Questions or a correction? Email hello@ukpeptides.org or call +44 7426 987848.
More GLP-1 & Incretin Science articles
- Danuglipron: The Oral GLP-1 Pfizer StoppedPfizer halted danuglipron in April 2025 after one asymptomatic liver injury case. Why a single event ended a programme that had met its objectives.
- Why Hepatic Safety Became the Oral GLP-1 QuestionInjectable peptide GLP-1s did not raise this question. Small molecules do, because of how they are metabolised - and one programme has already ended over it.
- Hy's Law: How Trials Decide a Drug Hurt the LiverThree measurements and an exclusion. What Hy's Law is, why it predicts severe outcomes, and why raised enzymes alone mean very little.
- The Obesity Drug That Is Not an IncretinSetmelanotide targets the melanocortin-4 receptor, not GLP-1. An approved medicine for rare genetic obesity, and a different mechanism entirely.
- The Survodutide Comparison, Read ProperlyAn open-label semaglutide arm in a type 2 diabetes dose-finding trial. Why that is weaker evidence than a head-to-head, and what the numbers mean.