Understanding Adverse Event Rates: Percentages and Relative Risk in Clinical Trials

When a new drug is tested in clinical trials, regulators like the FDA don’t just look at whether it works-they need to know how safe it is. But not all safety data is created equal. Two drugs might both cause headaches in 15% of patients, but if one group took the drug for 6 months and the other for 2 years, that 15% doesn’t tell the full story. This is where adverse event rates and relative risk become critical. Understanding the difference between simple percentages and exposure-adjusted measures can mean the difference between a misleading safety profile and an accurate one.

Why Simple Percentages Can Mislead

The most common way to report adverse events is the Incidence Rate (IR): the number of people who had an event divided by the total number of people exposed. If 15 out of 100 patients got a rash, you say 15% experienced it. Simple. Clean. But here’s the problem: it ignores how long each person was actually on the drug.

Imagine a trial where one group gets a new medication for 30 days, and another gets a placebo for 180 days. If 5 people in the drug group get a headache and 8 in the placebo group do, the IR says the drug causes fewer headaches. But what if the placebo group was exposed five times longer? The real risk might be the same-or even higher. That’s why the FDA started pushing back in 2023, asking companies to stop relying on IR alone in safety submissions.

Enter Patient-Years: The EIR Method

To fix this, statisticians started using Event Incidence Rate adjusted by Patient-Years (EIR). Instead of counting people, you count time. One patient taking a drug for 1 year = 1 patient-year. Ten patients taking it for 3 months each = 2.5 patient-years.

The formula? Total number of events divided by total patient-years of exposure. Then you multiply by 100 to get events per 100 patient-years. If 12 headaches happened across 600 patient-years of exposure, the EIR is 2.0 per 100 patient-years.

This method works well when the same event happens more than once in a person-like recurring nausea or dizziness. It gives you a sense of how often the event occurs over time, not just how many people experienced it once. JMP Clinical and other regulatory-grade tools now calculate EIR automatically using treatment start and end dates (TRTSDTM and TRTEDTM).

But EIR has a blind spot: it counts events, not people. If one patient gets 10 headaches, that’s 10 events. Another gets none. The EIR goes up-even though only one person is affected. That’s why regulators now prefer something even more precise.

The FDA’s New Standard: Exposure-Adjusted Incidence Rate (EAIR)

In 2023, the FDA formally requested EAIR in a supplemental biologics license application. This wasn’t a suggestion-it was a requirement. EAIR doesn’t just count events or time. It counts unique patients who had an event, adjusted for how long they were exposed.

EAIR = Number of patients with at least one event / Total patient-years of exposure.

This removes the inflation problem of EIR. If 10 patients had a rash and they were exposed for a combined 500 patient-years, EAIR = 2.0 per 100 patient-years. It tells you: “Out of everyone on the drug, how many had the event, and how long were they on it?”

MSD’s safety team found that switching to EAIR revealed hidden safety signals in 12% of their chronic therapy programs. In one case, a drug looked safe by IR because most patients dropped out early. But EAIR showed that those who stayed on long-term had a much higher rate of liver enzyme spikes. That signal would’ve been missed without exposure adjustment.

Relative Risk and Confidence Intervals: What the Numbers Really Mean

Comparing two groups? You don’t just look at their individual rates-you compare them. That’s relative risk.

If Drug A has an EAIR of 3.5 per 100 patient-years and Drug B has 1.8, the relative risk is 3.5 ÷ 1.8 = 1.94. That means patients on Drug A are nearly twice as likely to experience the event per unit of exposure.

But is that difference real? Or just noise? That’s where confidence intervals come in. The FDA expects statisticians to report 95% confidence intervals using the Wald method for incidence rate ratios. In R, this is done with the riskratio function. In SAS, it’s built into PROC FREQ with the RISKDIFF option.

If the confidence interval crosses 1.0, the difference isn’t statistically significant. A relative risk of 1.94 with a 95% CI of 0.98 to 3.89? That’s not enough to say Drug A is riskier. The data is too uncertain.

Competing Risks: When Death Changes the Story

Here’s a twist most people miss: if a patient dies, they can’t have another adverse event. That’s called a competing risk. In cancer trials, for example, death often happens before a slow-developing side effect like neuropathy can occur.

Traditional methods like Kaplan-Meier estimators assume everyone stays at risk until the end. But if someone dies, they’re no longer at risk for the event you’re studying. Using Kaplan-Meier here overestimates the risk of that event.

A 2025 study in Frontiers in Applied Mathematics and Statistics showed that using cumulative hazard ratio estimation improved accuracy by 22% in trials where competing events (like death) made up more than 15% of outcomes. The FDA hasn’t mandated this yet, but it’s being actively tested in their Sentinel Initiative with machine learning models that automatically flag these distortions.

Industry Challenges: Why This Isn’t Easy

Switching from IR to EAIR sounds simple. But in practice, it’s messy.

Pharmaceutical programmers report that EAIR takes 3.2 times longer to code than IR. A median of 14.7 hours per analysis versus 4.5. Common errors? Incorrect handling of treatment interruptions (19% of cases), wrong date formats (28%), and inconsistent patient-year calculations (23%).

Even worse, medical reviewers often don’t understand EAIR. Roche’s internal report found that 35% of reviewers misinterpreted the numbers at first. They saw “2.0 per 100 patient-years” and thought it meant 2% of patients had the event-ignoring the time component entirely. Companies had to create training videos and glossaries just to get their own teams on the same page.

The PhUSE team built a free SAS macro for EAIR that’s been downloaded over 1,800 times. Their GitHub repo now includes 37 validation checks-like making sure no patient’s exposure time exceeds the study duration, or that event counts match the number of patients reported.

What This Means for You

Whether you’re a patient reading a drug label, a clinician reviewing trial data, or a researcher analyzing safety outcomes, understanding these metrics matters.

- If a drug says “10% of patients had nausea,” ask: “Over what time?”

- If two drugs have similar IRs but different exposure times, the one with longer exposure might be safer.

- EAIR is becoming the gold standard-not because it’s perfect, but because it’s harder to game.

The FDA isn’t just asking for more data. They’re asking for smarter data. And the industry is responding. In 2020, only 12% of regulatory submissions included exposure-adjusted metrics. By 2023, that jumped to 47%. CDISC now mandates EAIR reporting for serious adverse events in oncology trials. Training enrollment in advanced safety analysis courses has grown 148% since 2021.

This isn’t just statistical pedantry. It’s about making sure the benefits of a drug aren’t outweighed by hidden risks. And it’s about giving patients and doctors the truth-not just a number that looks good on paper.

What’s Next?

The FDA’s 2024 draft guidance on exposure-adjusted analysis is open for public comment. The PhUSE team is finalizing an R-based reference implementation for Q1 2025. And by 2027, experts predict 92% of Phase 3 submissions will include EAIR alongside traditional IR.

The message is clear: if you’re evaluating drug safety, you can’t rely on percentages alone. Time matters. Exposure matters. And the way you calculate risk? It matters even more.