Proportional Reporting Ratio (PRR)

The PRR is a simple 2x2 contingency table calculation. It divides the proportion of a specific adverse event for a drug of interest by the proportion of that same event for all other drugs in the database.

• Formula: PRR = [a / (a + b)] / [c / (c + d)]
• a: Cases with the drug and the event of interest
• b: Cases with the drug but without the event
• c: Cases without the drug but with the event
• d: Cases without the drug and without the event

A PRR of 2.0 indicates the event is reported twice as frequently for the drug compared to the background rate.

A drug-event combination is traditionally flagged as a statistical signal of disproportionate reporting (SDR) when it meets three specific criteria simultaneously, as established by Evans et al. (2001).

• PRR Value: The calculated PRR must be ≥ 2
• Chi-Squared Statistic: The Pearson chi-squared test for independence must be ≥ 4, ensuring the association is unlikely due to chance
• Case Count: There must be a minimum of 3 individual cases to reduce the risk of flagging spurious associations from single reports

These thresholds balance sensitivity with the need to filter statistical noise.

As a frequentist method, PRR is highly sensitive, meaning it is effective at detecting true signals early. However, it suffers from low specificity, generating a high volume of false-positive associations.

• Volatility with Small Counts: PRR values are extremely unstable when the number of observed cases (cell 'a') is small. A single additional report can cause the PRR to spike dramatically.
• No Shrinkage: Unlike Bayesian methods like EBGM, PRR does not apply a statistical penalty to shrink estimates toward the null for low-count pairs, making it prone to flagging noise in sparse data.

This characteristic makes PRR an excellent triage tool but requires rigorous clinical review to filter false positives.

PRR and the Reporting Odds Ratio (ROR) are the two primary frequentist disproportionality measures and are mathematically related. They often produce concordant results in large databases.

• PRR: A ratio of proportions (risk ratio)
• ROR: A ratio of odds
• Convergence: When the adverse event is rare (low background incidence), the PRR and ROR values converge and become nearly identical
• Divergence: For very common events, the ROR will exaggerate the measure of association compared to the PRR

Both methods share the same vulnerability to instability with low cell counts.

PRR is not a causal inference tool but a hypothesis-generating triage mechanism. It rapidly sifts through millions of drug-event combinations in databases like FAERS and EudraVigilance to prioritize pairs for in-depth clinical review.

• Automated First Pass: Computationally inexpensive to run across entire databases
• Prioritization Metric: Generates a ranked list of drug-event pairs sorted by PRR magnitude
• Input to Signal Validation: A flagged PRR signal triggers the more resource-intensive process of signal validation, which includes causality assessment, literature review, and analysis of dechallenge/rechallenge data

Its primary value is in automating the initial filtering of massive spontaneous reporting datasets.

PRR is a crude measure that does not inherently control for confounding variables. Confounding by indication is a major interpretive challenge where the treated disease, not the drug, causes the event.

• Simpson's Paradox: Stratifying data by a confounder (e.g., age) can reverse the direction of a PRR signal observed in the aggregate data.
• Masking: A true safety signal can be hidden or 'masked' if a drug is overwhelmingly associated with another, highly reported event that inflates the background rate.
• Stratification: Analysts often calculate stratified PRRs within subgroups (e.g., age bands, sex) to mitigate confounding, though this reduces the case counts and increases statistical instability.

Clinical context is mandatory to interpret whether a high PRR reflects a true drug effect or a confounding bias.