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

When you hear that a 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. The real question isn’t just how many people had an adverse event-it’s how long they were exposed to the drug. This is where percentages alone fail, and why regulators like the FDA now demand more precise ways to measure risk.

Why Simple Percentages Mislead

The most common way to report adverse events is the Incidence Rate (IR): just divide the number of people who had an event by the total number of people in the study. If 15 out of 100 patients got a rash, you say 15%. Easy. But this method ignores time. In a trial where one group took the drug for 6 months and another for 24 months, the longer-exposed group will naturally have more events-not because the drug is more dangerous, but because people were exposed longer.

A 2010 analysis by Andrade found that using simple percentages in trials with uneven exposure times could underestimate true event rates by 18% to 37%. Imagine a drug that causes liver injury in 2% of patients over 6 months. If you give it to someone for 5 years, the risk isn’t still 2%. It’s much higher. But without adjusting for time, you’d miss that entirely.

Enter Patient-Years: The EIR Method

To fix this, statisticians use 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 6 months each = 5 patient-years. Then you divide the number of events by total patient-years.

For example: 12 liver injuries occurred across 1,200 patient-years of exposure. That’s 12 ÷ 1,200 = 0.01, or 1 event per 100 patient-years. This tells you: for every 100 people taking the drug for a full year, you’d expect 1 liver injury. Now you can compare groups with different treatment lengths fairly.

Tools like JMP Clinical automate this using start and end dates of treatment (TRTSDTM and TRTEDTM). But even here, problems arise. If a patient stops the drug for 3 weeks due to side effects, should that time count? Most systems don’t automatically exclude it. That’s why 31% of initial analyses have errors in exposure time calculation, according to PharmaSUG forums.

The FDA’s Push for EAIR

EIR helps, but it still has a flaw: it counts events, not people. If one patient has five separate episodes of nausea, EIR treats that as five events. But for a doctor deciding whether to prescribe, knowing that five episodes happened to one person is different from five different people each having one episode.

That’s where Exposure-Adjusted Incidence Rate (EAIR) comes in. EAIR adjusts for both time and recurrence. It answers: how many people had at least one event, and how long were they exposed? It’s more complex, but more clinically meaningful.

In 2023, the FDA requested EAIR in a supplemental biologics license application (sBLA) for the first time. It wasn’t a suggestion-it was a requirement. This signaled a major shift. The agency no longer accepts simple percentages as sufficient for safety evaluation, especially in long-term studies.

Relative Risk and Confidence Intervals

Comparing two drugs? You don’t just look at their individual rates-you calculate the ratio. That’s the Incidence Rate Ratio (IRR): divide the rate in Group A by the rate in Group B. If Group A has 2.5 events per 100 patient-years and Group B has 1.0, the IRR is 2.5. That means Group A has 2.5 times the rate of events.

But numbers alone aren’t enough. You need to know if that difference is real or just random noise. That’s where confidence intervals come in. For incidence rates, the Wilson score method with continuity correction is the gold standard. For IRR, the Wald method is used. In R, statisticians use prop.test() for rates and riskratio() for ratios.

If the 95% confidence interval for an IRR includes 1.0, the difference isn’t statistically significant. If it’s 1.8 to 3.2, then yes-there’s a real difference.

Competing Risks and Why Kaplan-Meier Fails

Not all patients stay in the study. Some die. Some quit. Some switch drugs. This creates what statisticians call “competing risks.” If a patient dies before having a heart attack, you never get to observe the heart attack. But if you use the Kaplan-Meier estimator (common in survival analysis), you treat death as a “censored” event, which distorts the risk of the event you’re studying.

A 2025 study in Frontiers in Applied Mathematics and Statistics showed that using Kaplan-Meier in these cases leads to misleading results. Instead, they recommend cumulative hazard ratio estimation, which separates the risk of death from the risk of the adverse event. This gives a clearer picture: is the drug increasing heart attack risk, or are people just dying faster from other causes?

Real-World Impact: What Happens When You Get It Right

MSD’s safety team found that switching to EAIR uncovered previously missed safety signals in 12% of their chronic therapy programs. One drug looked safe in traditional reports-until they realized patients on the placebo arm dropped out early due to side effects, making the comparison unfair. EAIR corrected that.

But adoption isn’t easy. A 2024 PhUSE survey found that writing EAIR code in SAS takes 3.2 times longer than IR. It requires deeper programming skills and more data cleaning. And even when done right, medical reviewers often don’t understand it. Roche reported that 35% of reviewers initially misinterpreted EAIR results, leading to unnecessary delays.

That’s why PhUSE released standardized SAS macros in March 2023. They’ve been downloaded over 1,800 times. Teams using them report an 83% drop in programming errors.

What You Need to Know Now

The FDA, EMA, and ICH are all moving toward exposure-adjusted methods. ICH E9(R1), effective since 2020, requires that safety analyses account for treatment discontinuation and exposure time. CDISC’s 2023 Oncology guide now mandates both IR and EAIR for serious adverse events.

If you’re reviewing a clinical trial report:

• Check if they used IR, EIR, or EAIR.
• Look for the denominator: number of patients, or patient-years?
• See if they adjusted for treatment interruptions.
• Ask: does the method match the clinical question?

No single method answers every question. IR is fine for short-term trials with uniform exposure. EIR works for recurrent events. EAIR is best for long-term, variable-exposure studies.

The Future Is Exposure-Adjusted

The global clinical trial safety software market hit $1.84 billion in 2023, growing 22.7% year-over-year. Why? Because regulators are demanding better data. By 2027, experts predict 92% of Phase 3 submissions will include EAIR.

The FDA’s 2024 draft guidance on exposure-adjusted analysis is now open for public comment. The Sentinel Initiative is training AI to detect safety signals using EAIR-early tests show 38% better detection than old methods.

This isn’t just statistics. It’s about patient safety. Getting the numbers right means doctors can make better decisions. Patients get the right treatment, with the right level of risk understood.

Forget the simple 15%. The future of drug safety is measured in time, recurrence, and context.