Clinical Trial Safety: AEs, Black Box Warnings, CRLs

Every efficacy guide on this site teaches you to read the numbers that make a stock go up. This one teaches you the numbers that make it go down. You read a press release, see “primary endpoint met with a p-value of 0.01,” and you’re ready to buy. Then the FDA issues a Complete Response Letter (CRL) because of liver toxicity you never looked at. Safety is the hidden half of every readout, and it’s where most CRLs are born: across the 2018-2022 PDUFA cycle, 37% of new drug and biologics applications received a CRL, and safety is the single most common reason.

The solution is simple. Read the safety tables first, not last. Before you care whether the drug worked, find out whether the patients who took it are still alive and still on the drug. Three numbers tell you most of the story: the Grade 3+ adverse event rate, the discontinuation rate, and the death imbalance.

Step 1: Learn the AE grading scale (CTCAE)

Every adverse event in a trial is graded on the CTCAE scale, the standard severity yardstick the FDA and every sponsor use. Grade 1 is mild, Grade 2 is moderate, Grade 3 is severe or medically significant (hospitalization, or limiting self-care), Grade 4 is life-threatening, and Grade 5 is death related to the drug. When a press release says a drug was “well tolerated,” it almost always means the Grade 1-2 rate was low. The numbers that matter are Grade 3 and above. A drug where 40% of patients had a Grade 3+ event is a completely different investment from one where 5% did, even if both “met their primary endpoint.”

Step 2: Separate TEAEs, SAEs, and AESIs

Not all side effects are equal, and companies exploit the difference. Treatment-emergent adverse events (TEAEs) are any new or worsening side effect that appears after the patient starts the drug, a broad bucket that includes mild nausea and headache. Serious adverse events (SAEs) are the subset that are life-threatening, cause hospitalization, or cause death. Adverse events of special interest (AESIs) are the specific events the trial is watching for, like liver enzyme elevations or cardiac arrhythmias, pre-specified because the drug class is known to cause them. A low overall TEAE rate can hide a dangerous signal: a drug with mostly Grade 1 fatigue but a 5% SAE rate for infections is riskier than one with a 30% Grade 2 nausea rate and zero SAEs. Find the SAE table and the AESI section specifically.

Step 3: Read the discontinuation rate

The percentage of patients who dropped out of the trial because of side effects is often more informative than the raw AE list. If 20% of patients stopped taking the drug because they couldn’t tolerate it, that drug won’t sell even if it works, because real-world patients aren’t monitored the way trial patients are. A high discontinuation rate is the market’s verdict on tolerability, and the FDA reads it exactly that way. A drug patients can’t stay on is a drug that won’t sell.

Step 4: Know the four classic CRL triggers

Safety rejections cluster around four signals. Liver toxicity is the big one: elevated ALT/AST liver enzymes, and specifically Hy’s Law, the combination of ALT or AST more than 3x the upper limit of normal plus bilirubin more than 2x the upper limit, which predicts severe drug-induced liver injury. Gene therapy has made this concrete: Sarepta’s DMD gene therapy Elevidys was linked to two patient deaths from acute liver failure in 2025, and the FDA suspended distribution. Cardiac signals are next, led by QTc prolongation, a measure of the heart’s electrical rhythm that can trigger arrhythmias. Severe infections, especially opportunistic infections from immune-suppressing drugs, are the third. Safety deaths, an imbalance of deaths in the drug arm versus the control arm, are the fourth and most obvious. Any one of these in the safety tables is a reason to slow down before the PDUFA date.

Step 5: Read the black box warning and the REMS

A black box warning is the FDA’s most severe safety warning, boxed text at the top of the label. It doesn’t block approval, but it permanently narrows the commercial opportunity because doctors hesitate and payers demand prior authorization. A REMS (Risk Evaluation and Mitigation Strategy) goes a step further: a required safety program, like mandatory patient monitoring or restricted distribution, that adds cost and friction to every prescription. Both are disclosed in the drug label, and both are exactly what the FDA deploys when it approves a drug despite a concerning safety profile. Read the label, not just the approval headline, which is where the guide to what the FDA label hides comes in.

Step 6: Find where the safety data is buried

Press releases bury bad news. The safety table usually sits on the last page, or gets compressed into one line: “safety was consistent with the known profile.” That sentence, with no numbers attached, is a red flag. Cross-check three places: the full press release tables, the drug label on DailyMed, and the trial’s record on ClinicalTrials.gov. The Amgen Tavneos withdrawal is the cautionary tale: the drug’s data was later found to have been manipulated, and investors who trusted the headline numbers paid for it (the full Tavneos withdrawal story).

Step 7: Weigh benefit against risk the way the FDA does

The FDA doesn’t demand zero side effects; it demands a favorable benefit-risk trade-off. A cancer drug with a 30% Grade 3+ AE rate can still win approval if it extends survival, while the same safety profile in a chronic, non-fatal condition like atopic dermatitis will get rejected. Strong efficacy plus poor safety can produce a CRL or a narrow label. That’s why you read safety alongside efficacy, not after it: the efficacy half is covered in the guide to hazard ratios, confidence intervals, and Kaplan-Meier curves, and the mechanics of a rejection are explained in what a CRL actually is.

Common mistakes

Buying the efficacy story without opening the safety tables. Most investors read “met its primary endpoint” and stop. The safety tables are where the rejection lives.

Assuming “well tolerated” means safe. It usually means the mild events were mild. Find the Grade 3+ rate, the SAE rate, and the deaths.

Ignoring the discontinuation rate. A drug patients can’t stay on is a drug that won’t sell, regardless of how good the efficacy data looked.

Trusting a one-line safety summary. “Consistent with the known profile” with no numbers attached means go find the numbers.

Forgetting that safety is indication-specific. The same side effect profile that’s acceptable in cancer is a rejection in a chronic disease. Compare the safety profile to the severity of the disease, not to an absolute standard.

Final checklist

Before you buy ahead of a PDUFA date or a readout, confirm five things. The Grade 3+ AE rate: is it single digits, or is a third of patients getting severe side effects? The SAE and death rates: is there an imbalance versus the control arm? The discontinuation rate: can patients actually stay on the drug? The four CRL triggers: any liver, cardiac, infection, or death signals? The label status: does it carry a black box warning or a REMS? If the safety data passes all five, then, and only then, does the efficacy story matter.