Research · 10 min read
What a Side-Effect Percentage on a Label Actually Counts
Every one of these labels opens its adverse reactions section by forbidding the comparison that comparison pages are built on. The placebo columns show exactly why.
Key takeaways
- Every one of these labels opens its adverse reactions section by stating that rates cannot be directly compared between drugs — the exact comparison most pages make.
- The placebo columns prove the point: nausea at 8, 10 and 16 percent and overall gastrointestinal reactions at 30, 37 and 47 percent, in groups receiving nothing.
- The inclusion threshold differs between these labels — at least 2 percent on two of them, at least 5 percent on the third — so a shorter table can mean a higher bar.
- An absent row means a reaction failed the frequency threshold or was not more common than placebo; it is not a zero.
- Single rows are buckets: one abdominal pain row folds in seven separately reported terms.
- Discontinuation because of a side effect ran 3 to 3.4 percent on placebo against 4.8 to 10 percent on treatment, rose with the amount studied, and was mostly gastrointestinal and early.
- Safety pools contained 25, 19 and 31 percent of patients with type 2 diabetes, while several headline weight trials on the same labels excluded it.
Answer first: the labels say these numbers cannot be compared across drugs, in the first sentence
Open the Clinical Trials Experience section of any of these three labels and the first sentence is the same on all of them. Because clinical trials are conducted under widely varying conditions, adverse reaction rates observed in the clinical trials of a drug cannot be directly compared to rates in the clinical trials of another drug, and may not reflect the rates observed in practice.
That is not boilerplate anybody should skim. It is the manufacturers and the regulator jointly stating that the tables you are about to read do not support a side-by-side ranking, which is the single most common thing done with them.
The proof is in the labels themselves. Compare not the treated columns but the placebo columns — the people who got nothing. Nausea was reported by 8 percent of the placebo group in one program, 10 percent in another and 16 percent in a third. Diarrhea: 8, 11 and 16 percent. Constipation: 5, 9 and 11. Gastrointestinal reactions overall were reported by 30 percent of one placebo group, 37 percent of another and 47 percent of a third.
Those are all placebo arms. If groups receiving nothing report the same symptom at rates differing by roughly a factor of two, then the treated columns sitting beside them cannot be lined up either. The difference is in the trials, not in the drugs.
Why placebo groups differ that much
Several reasons, all mundane and all documented on these pages. The populations differ — the safety pools on these three labels contain 25, 19 and 31 percent of patients with type 2 diabetes, and mean baseline body mass indexes of 37.4, around 38, and 36.5.
The way symptoms are solicited and recorded differs between trial programs, and the labels do not publish those instruments. Trial length differs, and so does the window after the drug stops during which reactions are still counted: 4 weeks on one label, 7 weeks on another, 2 weeks on the third. A longer tail collects more events.
And what counts as one row differs. On one of these labels, the row labeled abdominal pain includes abdominal pain, upper abdominal pain, lower abdominal pain, gastrointestinal pain, abdominal tenderness, abdominal discomfort and epigastric discomfort — seven reported terms folded into one number. On another, the injection site reactions row folds in bruising, redness, itching, pain, rash and general reaction. A row is a bucket, and the buckets are not the same size on every label.
A shorter table can mean a higher bar, not a safer drug
This one is worth checking every time, because it is invisible unless you read the table caption. Two of these labels print reactions that occurred in at least 2 percent of treated patients and more often than placebo. The third prints reactions that occurred in at least 5 percent.
So a reaction affecting 3 percent of patients would appear on two of these labels and be absent from the third, having occurred at more than twice the rate the other labels consider worth printing. The table is shorter because the bar is higher.
The same logic governs any row you do not see. Under a rule that requires both a frequency threshold and a rate above placebo, an absent row means one of those two conditions failed. Headache appears on two of these three tables and not the third — which tells you it did not clear both conditions there, and does not tell you that nobody had one.
The placebo column is the number that changes what a table means
Take one label's table on its own terms, which is the only way these are meant to be read. Nausea in the treated group at 44 percent looks alarming until you see 16 percent in the placebo group beside it. Diarrhea, 30 against 16. Constipation, 24 against 11. Headache, 14 against 10. Fatigue, 11 against 5.
The gap is what the drug added. For headache on that label the gap is four points; for nausea it is 28. A page that prints the treated number alone has thrown away the comparison the trial existed to make.
The same pattern holds on the other two. On one, nausea ran 25 to 29 percent across the amounts studied against 8 percent on placebo. On the other, 26 to 35 percent against 10 percent. Read within its own label, each of those is a clear signal. Read across labels, the placebo columns alone would mislead you.
Two figures that are more useful than any single row
The first is the proportion who stopped because of a side effect, because that combines frequency and severity into one number a person can act on. On these three labels: 4.8, 6.3 and 6.7 percent across the lowest, middle and highest amounts studied of one product against 3.4 percent on placebo; 6.8 percent against 3.2 percent on another; and 8 percent overall on the third — 6, 9 and 10 percent across its three amounts — against 3 percent on placebo.
Two things are visible there. Discontinuation rises with the amount studied on both labels that break it out. And on all three, the majority of those stops were gastrointestinal and happened early. One label states it directly: most patients who discontinued because of adverse reactions did so during the first few months, because of gastrointestinal reactions.
The second useful figure is the severity split, which one of these labels publishes and the others do not. Among treated patients reporting gastrointestinal reactions, 60 percent were mild, 36 percent moderate and 4 percent severe. Another label reports severe gastrointestinal reactions in 4.1 percent of treated patients against 0.9 percent on placebo. That is a different and better question than how many people had nausea.
The safety population is usually not the efficacy population
This trips up almost everyone, including people who read the labels carefully. The percentages in Section 6 come from a pooled safety population, and the percentages in Section 14 come from individual efficacy trials. They are different groups.
On one of these labels the safety pool contains 2,519 patients with obesity or overweight, with or without type 2 diabetes, and 25 percent of them had type 2 diabetes — while the headline weight percentage on the same label came from a trial that excluded it. On another, the safety pool is 2,116 patients, 19 percent with type 2 diabetes. On the third, 3,155 patients, 31 percent with type 2 diabetes.
So the side-effect rates and the weight-loss percentages on a single label frequently describe different mixes of people. Neither is wrong. They are just not two facts about the same population, and a summary that presents them as a matched pair is quietly wrong.
One of these labels adds a further wrinkle in plain sight: its safety section states that the data come from trials of an investigational formulation of the drug, presented as equivalent dosages of the marketed product. That is disclosed, and it is the kind of thing that only appears if you read the paragraph above the table.
How to read one of these tables in two minutes
Read the caption first. It tells you the inclusion threshold, and a table with a 5 percent bar is not comparable to one with a 2 percent bar even for the same symptom.
Read the paragraph above the table next. It tells you how many patients, for how long, how long after stopping events were still counted, and what the population looked like — including how many had type 2 diabetes, which is usually different from the efficacy trial on the same label.
Then read every row as a pair, treated against placebo, and look at the gap rather than the number. Check the footnotes for what got folded into each row. Find the discontinuation percentage, which is often in prose rather than in the table. And do not carry a number from this label to a sentence about a different drug, because the first sentence of the section says that comparison is not supported.
What that leaves you with is honest and useful: for this drug, in this trial program, this is roughly how often people reported this, and this is how often people on nothing reported it too.
Sources
- ZEPBOUND (tirzepatide) injection — full prescribing information, Section 6.1The opening sentence forbidding cross-drug comparison; the pooled safety population of 2,519 patients with a mean body mass index of 37.4 and 25 percent with type 2 diabetes; the 4-week off-drug follow-up period; the Table 1 caption threshold of at least 2 percent and greater than placebo; the placebo column figures of 8 percent nausea, 8 percent diarrhea and 5 percent constipation; the treated nausea range of 25 to 29 percent; the absence of a headache row; the footnote grouping bruising, redness, itching, pain and rash into one injection site reactions row; the discontinuation figures of 4.8, 6.3 and 6.7 percent against 3.4 percent on placebo; the statement that most such discontinuations occurred in the first few months and were gastrointestinal; and the overall gastrointestinal rate of 56 percent against 30 percent on placebo.
- WEGOVY (semaglutide) injection and tablet — full prescribing information, Section 6.1The opening sentence forbidding cross-drug comparison; the pooled safety population of 2,116 patients with 19 percent having type 2 diabetes; the 7-week off-drug follow-up period; the Table 3 caption threshold of at least 2 percent and greater than placebo; the paired figures for nausea 44 against 16, diarrhea 30 against 16, vomiting 24 against 6 in the table and 25 against 6 in the accompanying prose, constipation 24 against 11, headache 14 against 10 and fatigue 11 against 5; the footnote folding seven terms into the abdominal pain row; the discontinuation figure of 6.8 percent against 3.2 percent with nausea, vomiting and diarrhea named as the most common reasons; and the overall gastrointestinal rate of 73 percent against 47 percent on placebo with severe reactions in 4.1 percent against 0.9 percent.
- FOUNDAYO (orforglipron) tablet, film coated — full prescribing information, Section 6.1The opening sentence forbidding cross-drug comparison; the statement that the safety data come from trials of an investigational formulation presented as equivalent dosages of the marketed product; the pooled safety population of 3,155 patients with a mean body mass index of 36.5 and 31 percent with type 2 diabetes; the 2-week off-drug follow-up period; the Table 1 caption threshold of at least 5 percent; the placebo column figures of 10 percent nausea, 11 percent diarrhea and 9 percent constipation; the treated nausea range of 26 to 35 percent; the discontinuation figures of 8 percent overall and 6, 9 and 10 percent by amount against 3 percent on placebo; the overall gastrointestinal rates of 60, 68 and 69 percent against 37 percent on placebo; and the severity split of 60 percent mild, 36 percent moderate and 4 percent severe.
Frequently asked questions
Can I compare side-effect rates between two of these drugs using their labels?
The labels say no, in the first sentence of the section those tables appear in: rates observed in the clinical trials of one drug cannot be directly compared to rates in the clinical trials of another. The clearest evidence is in the placebo columns. Across three of these labels, placebo groups reported nausea at 8, 10 and 16 percent, diarrhea at 8, 11 and 16 percent, and gastrointestinal reactions overall at 30, 37 and 47 percent. Groups receiving nothing differ that much, so the treated columns beside them cannot be lined up.
Why do some labels list fewer side effects than others?
Often because the inclusion threshold differs. Two of these labels print reactions occurring in at least 2 percent of treated patients and more often than placebo; the third prints reactions occurring in at least 5 percent. A reaction affecting 3 percent of patients appears on two of them and is absent from the third. A shorter table can mean a higher bar rather than a safer drug.
If a symptom is not in the table, does that mean it does not happen?
No. The stated rule on these tables requires both a frequency threshold and a rate higher than placebo, so an absent row means one of those two conditions was not met. Headache appears on two of these three tables and not the third. That tells you it did not clear both conditions there. It does not tell you nobody had one.
What does the placebo column actually tell me?
It tells you how often people reported the same thing without the drug, which is the only way to see what the drug added. On one of these labels, nausea was 44 percent on treatment and 16 percent on placebo — a gap of 28 points. Headache on the same label was 14 against 10 — a gap of 4. The two rows read very differently once the placebo number is beside them, and a page printing only the treated figure has removed the comparison the trial was built to make.
Which number on these tables is the most useful?
Probably the proportion who stopped treatment because of a side effect, since it combines how common and how tolerable into one figure. Across these three labels it ran 4.8 to 6.7 percent, 6.8 percent, and 6 to 10 percent, against placebo rates of 3.4, 3.2 and 3 percent. It rose with the amount studied on both labels that break it down, and on all three the majority of those stops were gastrointestinal and happened early in treatment.
Do the side-effect percentages and the weight-loss percentages describe the same people?
Usually not. Side-effect figures come from a pooled safety population and weight figures from individual efficacy trials on the same label. The three safety pools here contained 25, 19 and 31 percent of patients with type 2 diabetes, while several of the headline weight trials on those same labels excluded type 2 diabetes entirely. Both sets of numbers are accurate; they are just not two facts about the same group.