Drug safety
Proportional reporting ratio (PRR): how drug safety signals are found
The proportional reporting ratio (PRR) compares how often a reaction is reported for one drug against all other drugs. Formula, worked example and limits.
By the All In Biotech editorial team · Published November 18, 2026 · 7 min read
A proportional reporting ratio (PRR) measures whether a reaction shows up more often in one drug's adverse event reports than in reports for all other drugs. It is the share of the drug's reports that mention the reaction, divided by the same share among every other report in the database. Safety reviewers use it to decide which drug and reaction pairs to examine more closely; it does not show that a drug causes anything.
Why disproportionality analysis exists
Spontaneous reporting systems such as FDA's Adverse Event Reporting System (FAERS) collect reports from patients, clinicians and manufacturers. They hold millions of records, but they do not record how many people took each drug. Without that denominator you cannot calculate an incidence rate.
Disproportionality analysis works around the missing denominator by comparing reports with other reports. If a reaction makes up a much larger slice of one drug's reports than of the database as a whole, that imbalance is a possible signal. The PRR is the simplest and most widely used of these measures. For a primer on the database itself, see how to search FAERS.
The 2x2 table behind every PRR
Every PRR starts from four counts of reports:
| Reaction of interest | All other reactions | Total | |
|---|---|---|---|
| Drug of interest | a | b | a + b |
| All other drugs | c | d | c + d |
- a is the number of reports that name the drug and the reaction.
- b is reports that name the drug but not the reaction.
- c is reports that name the reaction but not the drug.
- d is reports that name neither.
The PRR is then:
PRR = (a / (a + b)) / (c / (c + d))
The top half of the fraction is the proportion of the drug's reports that mention the reaction. The bottom half is the same proportion in the rest of the database. A PRR of 1 means no imbalance. A PRR of 2 means the reaction is twice as common, proportionally, in that drug's reports.
A worked example: Ozempic and impaired gastric emptying
This site's FAERS tool flags impaired gastric emptying as a disproportionate reaction for Ozempic (semaglutide). As of November 2026, the site's openFDA data show:
- 70,721 FAERS reports naming Ozempic (both Ozempic NDAs plus the brand name).
- 4,024 of those reports list impaired gastric emptying.
- About 21 million FAERS reports in total.
- A PRR of about 124.
The one number not listed above is the total count of impaired gastric emptying reports across all drugs. For this example it is worked backwards from the PRR of 124, which gives about 13,600 reports in all, so roughly 9,600 reports that name the reaction without naming Ozempic. Treat that figure as illustrative; the live count on the tool may differ slightly.
| Impaired gastric emptying | Other reactions | Total | |
|---|---|---|---|
| Ozempic | a = 4,024 | b = 66,697 | 70,721 |
| All other drugs | c = 9,600 | d = 20,919,679 | 20,929,279 |
| Total | 13,624 | 20,986,376 | 21,000,000 |
Now the arithmetic:
- Share of Ozempic reports with the reaction: 4,024 / 70,721 = 0.0569, or about 5.7%.
- Share of all other reports with the reaction: 9,600 / 20,929,279 = 0.000459, or about 0.046%.
- PRR = 0.0569 / 0.000459 = about 124.
In plain terms, impaired gastric emptying appears in roughly one Ozempic report in 18, but in roughly one other report in 2,200. If the reaction were reported for Ozempic at the background rate, you would expect about 46 such reports rather than 4,024.
That result is not surprising. Semaglutide slows gastric emptying as part of how it works, and delayed gastric emptying has drawn public attention, which itself raises reporting. That is a useful reminder of what a PRR can and cannot tell you, covered below. The drug's own label is the place to see what FDA has already added; the Ozempic record on this site links to it.
The confidence interval and chi-squared
A PRR from small counts can be large by chance. Two extra statistics help separate noise from a consistent pattern.
95% confidence interval
The standard error of the natural log of the PRR is:
SE = square root of (1/a - 1/(a + b) + 1/c - 1/(c + d))
The lower bound of the 95% confidence interval is:
Lower bound = exp(ln(PRR) - 1.96 x SE)
For the example: SE = square root of (1/4,024 - 1/70,721 + 1/9,600 - 1/20,929,279) = about 0.0184. The interval runs from about 119.7 to about 128.6. Because both counts are large, the interval is narrow. With only 3 or 4 reports, the same PRR would come with a very wide interval.
Yates-corrected chi-squared
The chi-squared statistic tests whether the counts in the table depart from what you would expect if the drug and the reaction were unrelated. With Yates' continuity correction, and n = a + b + c + d:
Chi-squared = n x D x D / ((a + b)(c + d)(a + c)(b + d)), where D is the absolute difference between ad and bc, minus n/2
For the example, this comes to about 346,000. With one degree of freedom, a value above 3.84 corresponds to p below 0.05, so the usual threshold of 4 sits close to that conventional cut-off. This pair clears it by a wide margin.
When a PRR counts as a signal: the Evans criteria
A widely used screening rule comes from Evans and colleagues, who described the PRR method in 2001 for the UK's spontaneous reporting data. A drug and reaction pair is flagged when all three conditions hold:
- PRR of at least 2.
- Chi-squared of at least 4.
- At least 3 reports (a of 3 or more).
The report threshold stops single anecdotes from being flagged. The chi-squared threshold filters out ratios driven by small numbers. Different organizations use variations, such as requiring the lower confidence bound to exceed 1, so check the method behind any signal list you read.
How this site calculates PRR
The FAERS tool applies the method above to each drug's most reported reactions. For each reaction it:
- Takes a, the number of reports naming the drug and the reaction, from openFDA.
- Sets b as the drug's total reports minus a.
- Looks up the total number of FAERS reports naming that reaction (by exact MedDRA preferred term) and subtracts a to get c.
- Sets d as all FAERS reports minus the drug's reports minus c.
- Calculates the PRR, the 95% lower confidence bound and the Yates-corrected chi-squared.
- Flags the reaction when the PRR is at least 2, the chi-squared is at least 4 and there are 3 or more reports.
Results are sorted by PRR. For Ozempic, the flagged list includes impaired gastric emptying (PRR about 124), constipation (about 6.4) and decreased appetite (about 6.1), as of November 2026. Nausea is the most reported reaction overall, at 10,175 reports, but a high count alone does not make a reaction disproportionate. The methods page describes the data source and caching.
Why a high PRR does not prove harm
FDA states plainly that FAERS reports do not establish that a drug caused an event. Several features of spontaneous data inflate or hide PRRs:
- No exposure data. A PRR compares reports with reports, not patients with patients. A drug with many reports overall can show a lower PRR for a real effect.
- Stimulated reporting. Media coverage, lawsuits and safety communications raise reporting for specific pairs.
- Confounding by indication. Reactions linked to the disease being treated show up as disproportionate. Drugs for diabetes will over-report events common in people with diabetes.
- Duplicates and incomplete reports. The same case can arrive from a patient, a doctor and a manufacturer. Many reports lack doses, dates or outcomes.
- Masking. When one drug dominates reports for a reaction, it raises the background rate and can hide signals for other drugs.
- Multiple comparisons. Screening thousands of pairs will produce some flags by chance.
A flagged pair is a prompt for case review: reading individual reports, checking timing and dose, looking at the label and the literature, and, where warranted, studying the question in other data such as claims or trial results. The drug label guide explains where confirmed risks end up.
Other disproportionality measures
The PRR is one of several measures built on the same 2x2 table:
- Reporting odds ratio (ROR): (a/b) / (c/d). Close to the PRR when the reaction is rare.
- Information component (IC): a Bayesian measure used by the WHO's Uppsala Monitoring Centre, which shrinks estimates from small counts toward no effect.
- Empirical Bayes geometric mean (EBGM): another shrinkage method, used in FDA's data mining of FAERS.
Bayesian methods give steadier results for rare pairs, at the cost of being harder to calculate by hand. The PRR's advantage is that anyone can check it with four numbers and a calculator.
For context on the drug in this example, including Medicare spending on semaglutide, see every FDA-approved GLP-1 drug.
Frequently asked questions
- What is a proportional reporting ratio?
- It is the share of a drug's adverse event reports that mention a given reaction, divided by the share of all other reports that mention the same reaction. A PRR of 3 means the reaction appears three times as often, proportionally, in that drug's reports.
- What PRR counts as a signal?
- A common screening rule, the Evans criteria, flags a drug and reaction pair when the PRR is at least 2, the chi-squared statistic is at least 4 and there are at least 3 reports. A flag means the pair deserves a closer look, not that the drug causes the reaction.
- Does a high PRR prove a drug causes a side effect?
- No. Spontaneous reports have no denominator of exposed patients and are shaped by publicity, reporting habits and the disease being treated, so a PRR only points to where reviewers should look next.
- How is PRR different from the reporting odds ratio?
- Both use the same 2x2 table of report counts. The PRR compares proportions of reports, while the reporting odds ratio compares odds, and the two give similar values when the reaction is rare.
For information only, not medical, legal or investment advice. Facts about FDA and CMS programs are current as of the publication date; check the linked official source for changes.