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

When you hear that a new drug causes headaches in 15% of patients, it sounds simple. But what if one group took the drug for 3 months and another took it for 2 years? That 15% doesn’t tell the whole story. Adverse event rates aren’t just about how many people got sick-they’re about how long they were exposed, how often events happened, and whether the data accounts for real-world complexity. The FDA is shifting away from outdated methods, and if you’re reading clinical trial results, you need to know why.

Why Simple Percentages Mislead

For decades, the go-to way to report adverse events was the Incidence Rate (IR): divide the number of people who had an event by the total number of people in the study. If 30 out of 200 patients got nausea, you say 15%. Easy. But here’s the problem: it ignores time.

Imagine two groups in a diabetes drug trial. Group A gets the drug for 6 months. Group B stays on it for 3 years. In Group A, 10 people report nausea. In Group B, 45 people do. The IR says Group B has a 45% nausea rate versus Group A’s 10%. That looks scary. But if you dig deeper, Group B had 18 times more exposure time. That 45% isn’t a higher risk-it’s just more time for nausea to show up.

The FDA noticed this. In 2023, they asked a biotech company to resubmit safety data using a different method because the original IR made a drug look riskier than it was. That’s not a technicality-it’s a safety issue. Misleading numbers can scare patients, delay approvals, or even block useful treatments.

What Is Exposure-Adjusted Incidence Rate (EAIR)?

The solution? Exposure-adjusted incidence rate, or EAIR. Instead of counting people, EAIR counts patient-years. One patient taking a drug for one year = one patient-year. Ten patients taking it for six months = five patient-years.

The formula is simple: total number of adverse events divided by total patient-years of exposure. Then you scale it to per 100 patient-years for readability. If 45 nausea events happened over 1,200 patient-years, the EAIR is 3.75 per 100 patient-years. Now you can compare that to another drug with 5.2 per 100 patient-years-even if one trial lasted 1 year and the other lasted 5.

EAIR doesn’t just fix time issues. It also handles recurrence. If one patient gets nausea 3 times in 2 years, EAIR counts all three events. That’s important because some side effects aren’t one-offs-they’re ongoing. A single person having 5 migraines on a drug tells you more than 5 different people having one each.

The FDA’s 2024 draft guidance on exposure-adjusted analysis says EAIR should be standard for serious adverse events. CDISC, the group that sets data standards for clinical trials, now requires EAIR reporting in oncology trials starting in 2023. Companies like MSD found that switching to EAIR uncovered safety signals they’d missed for years-especially in chronic disease trials where patients stay on drugs for decades.

When EIR Works-and When It Doesn’t

There’s another method called Event Incidence Rate (EIR), which also uses patient-years. It’s often confused with EAIR, but there’s a key difference: EIR counts events per patient-year, but doesn’t always adjust for multiple events per person. In practice, many use EIR and EAIR interchangeably, but regulators are starting to distinguish them.

EIR is great for recurrent events like rashes, diarrhea, or coughing fits. But if you’re studying something like death or hospitalization, counting multiple events per person doesn’t make sense. One person can only die once. That’s where EAIR’s structure-counting events while adjusting for exposure-fits better.

The real trouble with EIR comes when patients drop out early. If someone stops the drug after 2 weeks because of side effects, their short exposure still counts in the denominator. But if you don’t account for why they left, you might overestimate risk. EAIR handles this better by tracking exact start and end dates of treatment, including interruptions.

A 2023 PhUSE survey found that 23% of early EAIR calculations had errors because teams used the wrong dates or ignored treatment breaks. That’s why standardized tools like the PhUSE SAS macros-downloaded over 1,800 times-are becoming essential. Without them, even experienced statisticians make mistakes.

Relative Risk and Why It’s Not Enough

You might see phrases like “the drug doubles the risk of liver injury.” That’s relative risk-or more precisely, incidence rate ratio (IRR). It’s the ratio of the event rate in the treatment group versus the control group. If 5 per 100 patient-years get liver injury on the drug, and 2.5 on placebo, the IRR is 2.0.

But here’s the trap: relative risk looks dramatic, but absolute risk tells you what matters. A 2x increase sounds scary. But if the baseline risk is 0.1% and the drug raises it to 0.2%, you’re talking about one extra case per 1,000 people. That’s very different from a baseline of 10% jumping to 20%.

The FDA now requires both absolute and relative measures in safety summaries. They want you to see the full picture: the rate per 100 patient-years and the ratio compared to placebo. One without the other is incomplete.

Competing Risks and Why Kaplan-Meier Fails

Here’s a hidden problem: patients die. And if they die, they can’t have another adverse event. That’s called a competing risk.

In cancer trials, for example, many patients die from the disease before they ever get a rare heart side effect. If you use the old Kaplan-Meier method to estimate how long until a side effect happens, you’ll underestimate the true risk. Why? Because it treats death as if it’s just a lost observation-not a reason the patient stopped being at risk for the event.

A 2025 study in Frontiers in Applied Mathematics and Statistics showed that traditional methods underestimated adverse event rates by up to 22% in trials where death rates exceeded 15%. The fix? Cumulative hazard ratio estimation. It breaks down the risk into separate pathways: death, heart event, kidney event. Each has its own hazard function. You don’t pretend death didn’t happen-you model it as a competing force.

This isn’t theoretical. The FDA’s Sentinel Initiative is now testing machine learning models that use this approach to detect safety signals earlier. Early results show a 38% improvement in catching hidden risks.

What This Means for You

If you’re a patient reading a drug label: look for numbers like “per 100 patient-years,” not just percentages. Ask: “How long were people on this drug?” If it’s not stated, the data might be misleading.

If you’re a clinician reviewing trial results: don’t trust IR alone. Compare EAIR values between arms. Check if the study adjusted for treatment interruptions. Look for both absolute and relative risk.

If you’re in pharma or biotech: EAIR is no longer optional. The FDA expects it. CDISC requires it for certain trials. Your programming teams need standardized SAS or R scripts. Your medical reviewers need training-35% of them misinterpreted EAIR in early rollout.

The industry is changing fast. In 2020, only 12% of FDA submissions included exposure-adjusted metrics. By 2023, that jumped to 47%. By 2027, experts predict 92% of Phase 3 trials will report EAIR alongside traditional rates.

This isn’t about statistics for statistics’ sake. It’s about giving patients and doctors the truth-not a simplified number that hides real risk or falsely scares people away from life-saving treatments.

What’s Next?

The next wave is automation. Tools like JMP Clinical and R’s survival package are building built-in EAIR calculators. The PhUSE team is releasing a public R implementation in early 2025. The FDA’s draft guidance will likely become final by mid-2025, making EAIR mandatory for many submissions.

Until then, the best practice is simple: always ask, “Was exposure time accounted for?” If the answer is no, treat the numbers with caution. If the answer is yes-check the details. How were patient-years calculated? Were interruptions included? Were competing risks modeled?

The science of safety is evolving. The old way of counting people who got sick is outdated. The new way-measuring risk over time, accounting for real-life complexity-is the only way to know what a drug truly does to the body.