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

When a new drug enters clinical trials, one of the most important questions isn't just whether it works - it's whether it's safe. But how do we really know? Saying "15% of patients had a headache" sounds simple. But what if some patients took the drug for 3 months and others for 3 years? That 15% doesn't tell the full story. This is where understanding adverse event rates becomes critical - not just as percentages, but as real measures of risk over time.

Why Simple Percentages Can Mislead

For years, the go-to way to report adverse events was the Incidence Rate (IR): just count how many people had an event, divide by the total number of people in the trial, and call it a day. If 30 out of 200 patients had nausea, you say "15% experienced nausea." Easy. But this method ignores something vital: time.

Imagine two groups in a trial. Group A takes the drug for 6 months. Group B takes it for 2 years. If 10 people in each group report nausea, the IR is identical: 10/100 = 10%. But Group B was exposed to the drug 4 times longer. That means each person in Group B had far more opportunity to experience nausea. The simple percentage hides that difference. It’s like saying two cars have the same fuel efficiency because they both used 5 gallons - even if one drove 50 miles and the other drove 200.

The FDA noticed this problem. In 2023, they requested that a biotech company resubmit its safety data using a different method. That wasn’t a routine request. It was a signal: the old way isn’t good enough anymore.

The Shift to Exposure-Adjusted Measures

The industry is moving toward Exposure-Adjusted Incidence Rate (EAIR). Instead of just counting people, EAIR counts patient-years. That’s the total time all patients spent on the drug.

Here’s how it works: If 100 patients each took the drug for 1 year, that’s 100 patient-years. If 50 patients took it for 2 years, that’s 100 patient-years too. Now, if 15 people had liver enzyme changes over that time, EAIR says: "15 events per 100 patient-years." That’s a rate - not just a percentage.

This matters because it accounts for real-world usage. In long-term studies - say, for diabetes or arthritis drugs - patients don’t all stay on treatment the same length of time. Some quit due to side effects. Others switch. EAIR captures that. It doesn’t just tell you how many people had a problem. It tells you how often the problem happened per unit of exposure.

A 2023 analysis by MSD’s safety team found that switching to EAIR uncovered safety signals in 12% of their drug programs that had been missed using old methods. These weren’t minor issues. They were patterns that only became clear when you looked at time on drug.

Comparing Methods: IR, EIR, and EAIR

There are three main ways to measure adverse events today:

• Incidence Rate (IR): Simple percentage. (# of affected patients / total patients). Easy to calculate, but ignores time.
• Event Incidence Rate (EIR): Events per 100 patient-years. Better for recurrent events like diarrhea or rashes, but counts each event separately - so one patient with 5 episodes counts as 5 events.
• Exposure-Adjusted Incidence Rate (EAIR): Accounts for both how many people had events and how long they were exposed. Also considers if someone had multiple events. This is now the gold standard for regulatory submissions.

About the three methods for calculating adverse event rates

Method	What It Measures	Best For	Limitations
Incidence Rate (IR)	Percentage of patients affected	Quick, early-phase trials	Underestimates risk when exposure times differ
Event Incidence Rate (EIR)	Events per 100 patient-years	Recurrent events (e.g., nausea, rash)	Overstates risk if one patient has multiple events
Exposure-Adjusted Incidence Rate (EAIR)	Events per patient-year, adjusted for recurrence and duration	Long-term studies, regulatory submissions	Harder to calculate; requires detailed date tracking

The European Medicines Agency (EMA) still accepts IR, but only if the sponsor explains why they didn’t use EAIR. The FDA now expects EAIR in most new drug applications. And it’s not just big pharma - smaller biotechs are adopting it too, because regulators won’t approve without it.

Why Time Matters More Than You Think

A 2025 study in Frontiers in Applied Mathematics and Statistics showed that when death is a competing risk - meaning a patient dies before an event like liver failure can be observed - traditional methods like Kaplan-Meier can give misleading results. That’s why experts now recommend using cumulative hazard ratios instead. It’s a mouthful, but the idea is simple: don’t just count events. Model the risk over time.

Think of it like this: If a drug causes heart palpitations, but 20% of patients die from other causes in the first year, you can’t just say "20% had palpitations." Some of those patients never lived long enough to have them. EAIR and cumulative hazard models account for this. They don’t pretend everyone was at risk for the whole time.

Dr. Gary Koch, a former FDA advisor, put it bluntly: "Failure to account for exposure time constitutes a fundamental statistical error." He wasn’t exaggerating. In one case, a drug looked safer than its competitor because the comparison group was on placebo for 6 months, while the drug group was on therapy for 18 months. The IR showed fewer events in the drug group - but EAIR revealed the opposite. The drug actually caused more events per unit of exposure.

What’s Changing in the Industry

The tools are catching up. CDISC, the global standard for clinical data, now requires both IR and EAIR for serious adverse events in oncology trials (v3.0, Sept 2023). SAS and R have new macros to automate EAIR calculations. The PhUSE team released open-source code in 2023 that’s been downloaded over 1,800 times. Companies that used it saw an 83% drop in calculation errors.

But adoption isn’t smooth. A 2024 survey of 75 clinical programmers found that EAIR took 3.2 times longer to code than IR. Common mistakes? Wrong start/end dates, not accounting for treatment breaks, or mixing up event counts with patient counts. One company reported that 35% of their medical reviewers initially misread EAIR results because they’d never seen it before.

That’s why training matters. The Society for Clinical Data Management saw enrollment in their Advanced Safety Analysis course jump 148% since 2021. Regulatory agencies are updating their review templates too. The FDA now includes checklists for exposure time calculations. If you can’t prove how you calculated patient-years, your submission gets sent back.

What You Need to Know Now

If you’re reviewing clinical trial data, here’s what to ask:

• Was EAIR calculated alongside IR?
• How was exposure time defined? Did they use treatment start and end dates?
• Were treatment interruptions accounted for? (e.g., if a patient stopped the drug for 2 weeks, was that time excluded?)
• Did they use validated software? (PhUSE macros, JMP Clinical, or similar)
• Are the results presented as rates per 100 patient-years - not just percentages?

Don’t settle for "X% had side effects." That number might be technically correct - but dangerously incomplete. The goal isn’t just to report events. It’s to understand risk. And risk isn’t just about who had an event. It’s about how long they were exposed to the drug.

The future of drug safety isn’t about bigger trials. It’s about smarter math. The FDA, EMA, and leading pharmaceutical companies are all moving toward methods that reflect real-world use. If you’re still using simple percentages for long-term studies, you’re not just outdated - you’re risking misinterpretation.

Adverse events aren’t just numbers. They’re signals. And signals only make sense when you know how long people were listening.