Can you use your own patient outcomes in your marketing?
Yes, and you should. Your outcomes are the one asset a competitor cannot copy, buy, or outbid you for. A funded rival can clone your landing page this afternoon and outspend you in the auction tomorrow. It cannot clone what happened to your patients.
The way you will ruin it is the denominator.
Not with a fake number. With a true one, describing a group you quietly narrowed. This is the most common way honest companies publish misleading results, and it is worth understanding in detail, because the people who do it are usually not lying and often do not notice.
Why is your outcomes data worth more than another ad?
Traditional healthcare cannot really do this. Patient volumes at a single practice are too small for a number to mean anything, data lives in fragments across systems that do not speak, and the relationship with the patient is mediated by an insurer.
Telehealth has the inverse problem, which is an advantage. Volume is high, capture is digital by default, and the relationship runs from intake through refill inside one system. If you treat a few thousand patients on a protocol, you are sitting on something that a marketer with a budget simply cannot manufacture.
This is the differentiation lever in its most concrete form. Differentiation is one of only two things that make a patient cheap to acquire, and "our patients actually get results, here is the evidence" is the version of it that survives contact with a skeptic. Most operators in this category are sitting on that asset and not using it.
The ones who do use it tend to break it in the same place.
What does the FTC actually require of a health claim?
More than most operators expect, and the standard is not a technicality.
The FTC's health products compliance guidance sets the bar plainly: "As a general matter, substantiation of health-related benefits will need to be in the form of randomized, controlled human clinical testing to meet the competent and reliable scientific standard." And it closes the door on the easy substitute: "Anecdotal evidence about the individual experiences of consumers, including surveys of consumer experiences, are never sufficient to substantiate claims about the effects of a health product."
So testimonials are not evidence. Neither is a survey of happy patients.
Two further lines matter more than the headline standard, and almost nobody quotes them.
The first is about who was in the study: "Inclusion and exclusion criteria for subjects should be clearly stated in the protocol and relevant to the population to which the product is marketed." The group you measured has to resemble the group you are selling to.
The second is the one to write on the wall:
"Subject dropout rates, non-compliance, or concurrent changes in diet or other health-related behaviors should be carefully assessed to ensure they don't undermine any findings. For example, researchers should conduct an 'intent-to-treat' analysis that includes data from every subject initially assigned to the treatment and control groups, including subjects who dropped out during the course of the study or did not fully comply with the study protocol."
That is the FTC, in its own words, describing the error this page is about. Intent-to-treat means you count the people who quit. If your headline number excludes them, it is a completer number, and a completer number answers a different question than the one your reader is asking.
One more, because "peer-reviewed" gets used as a shield: "The mere fact that a study is published, however, isn't a guarantee of quality or proof that the product is effective for the advertised benefit."
What does the denominator problem look like in real life?
The clearest recent example is public, peer-reviewed, and worth studying precisely because nobody involved did anything dishonest.
In December 2025, Ro published a study in Obesity, the journal of The Obesity Society, on semaglutide outcomes through its telehealth platform. Start with what the paper says, from its own abstract:
"Data were obtained from deidentified EHRs for a random sample of 4500 patients who initiated semaglutide treatment via telehealth for overweight and obesity between December 1, 2022, and June 1, 2023."
Four thousand five hundred patients started. Now the result:
"Of 655 patients (n = 445 female, n = 210 male), mean body weight reduction was −16.6% (SD 7.5%, 95% CI: −17.1% to −16.0%)."
Six hundred and fifty-five of four thousand five hundred. The 16.6% describes about one patient in seven.
Before the marketing question, there is a clinical one: who are the other 3,845, and why did they go?
From Pranay Parikh, MD, founder of Off-Label:
"The unfortunate thing with GLP-1s is that everyone reacts differently. Some people will have amazing weight loss, and that is what gets marketed. Others will be on the max dose and have no results. Some will have serious side effects while others barely feel anything. There are not other medications with such varied responses. So a patient may have a lot of side effects and no positive effect, or no effect either way, and both of them leave, for different reasons."
Sit with what that does to the number.
The patients who left are not one group who lacked resolve. They are at least two groups leaving for opposite reasons: the ones it hurt without helping, and the ones it did nothing to at all. Neither is in the 16.6%.
And it means persistence is not really measuring persistence. In a drug class where response varies this widely, the people who stay are disproportionately the people it worked for. So a persister cohort is close to a responder cohort, and "what happened to the patients who stayed 68 weeks" is close to "what happens to the patients this drug works on." That is a genuinely useful number. It is not the number a reader thinks they are being given.
The paper is careful about this. Its own summary box reads: "Patients who persisted with semaglutide treatment via a nationwide telehealth platform achieved 68-week weight loss outcomes comparable to clinical trial completers, with no major safety concerns observed." Read the two qualifiers the authors chose. Patients who persisted. Comparable to trial completers. Both are precise, and both are doing real work.
Now read the press release. The subhead, which is also the meta description, the thing that appears in search results and social previews:
"Peer-reviewed study demonstrates patients treated with semaglutide through Ro achieve an average body weight loss of 16.6% over 68 weeks"
"Patients treated with semaglutide through Ro." Not patients who persisted. Not completers. And the number 4,500 does not appear anywhere in the release.
To be fair to Ro, because the fairness is the point: the qualifier is not hidden. Their own methodology paragraph says the sample was "patients treated with branded semaglutide (Wegovy or Ozempic) through Ro's platform who persisted with treatment and reported their weight at 68 weeks (±14 days)." It is disclosed. It is just disclosed in paragraph nine, while the unqualified version is in the headline, the subhead, and the preview text.
That is not fraud. That is gravity. A qualifier is heavy and a headline is short, and the qualifier falls out. It happens at good companies, written by honest people, about real data. Which is exactly why it is worth naming: you will not avoid this by having integrity. You will avoid it by having a rule.
And credit where it is due, because almost nobody in this category has earned it: Ro published. Peer-reviewed, in a real journal, with the attrition in the abstract and the conflicts disclosed in full. Most operators sitting on outcomes data publish nothing at all, which means there is nothing to critique and nothing to trust either. Publishing and being scrutinised is the better failure mode.
Why is "consistent with clinical trials" the interesting claim?
Because of what a trial actually reports, and it reports two numbers.
The STEP 1 trial of once-weekly semaglutide is the reference point everyone reaches for. Read what it says:
"For the treatment policy estimand (showing the effect regardless of treatment discontinuation or rescue intervention), the estimated mean weight change at week 68 was −14.9% with 2.4-mg semaglutide, as compared with −2.4% with placebo"
And then:
"For the trial product estimand (showing the effect if the drug or placebo was taken as intended), the corresponding changes were −16.9% and −2.4%"
Two numbers, from the same trial, at the same week. 14.9% counts everyone, including the people who stopped. 16.9% counts the people who took it as intended. The trial reports both, on purpose, because they answer different questions.
A real-world persister cohort belongs next to the second number. And the honest observation is that Ro's 16.6% and the trial's 16.9% are close, which is roughly what you would expect: two groups of people who stayed on the drug lost about the same amount of weight.
The thing worth noticing is what is missing. There is no Ro equivalent of 14.9%, because the 3,845 patients who are not in the analysis cannot be counted. So "consistent with clinical trials" is a claim you can check against the completer number and cannot check at all against the intent-to-treat number. It is not that the figure is wrong. It is that half the comparison does not exist.
Two more things you would need before running that comparison at all, and both cut the same way. The trial's patients started at a mean BMI of 37.9; Ro's cohort had a median of 32.3. Every trial patient was on 2.4 mg; only 52.1% of Ro's were at week 68. And the trial had a placebo arm that lost 2.4% on its own, so the drug's effect was the gap between the two. A single-arm cohort has no such gap, and quietly credits the platform with all of it.
How do you report outcomes honestly and still have a good number?
Start with the denominator and never move it.
Publish the intent-to-treat figure, or say plainly that you have not. If 4,500 started and 655 finished, the number your reader wants is what happened to 4,500. If you cannot compute it, that sentence itself is the disclosure, and it is a more credible thing to say than a headline that quietly means something narrower.
Put the qualifier where the number is. If the group is people who stayed on treatment for 68 weeks and reported a weight, that phrase belongs in the same sentence as the percentage, not nine paragraphs below it. The test is simple: if a reader quotes your headline and drops your methodology, is the quote still true? If not, the headline is doing the misleading, whatever the methodology says.
Say why the leavers left. This is the one that separates a real outcomes report from a dressed-up one, and in this drug class it is not optional. If patients quit because of side effects with no benefit, and other patients quit because nothing happened at all, those are two different findings and both belong to you. An operator who publishes "of 4,500 starters, X% stopped for side effects, Y% saw no response by week 12, and the Z% who continued to 68 weeks averaged this" has told the truth and given the reader something no competitor can copy. An operator who publishes only the last clause has published a testimonial with a p-value.
Name what you cannot know. No control arm means you cannot separate your platform's effect from the drug's, or from the fact that people who seek treatment and stick with it are different from people who do not. Say so. It costs you nothing with a serious reader and it is the difference between evidence and advertising.
Do not lean on "peer-reviewed" as a shield. The FTC already said publication is not proof. It is a real signal about the paper. It is not a claim about your product.
And do not reach for "results not typical." In the testimonials section of the same guidance, the FTC dismisses that phrase outright, and while that passage is scoped to testimonials rather than study populations, the logic transfers exactly: a disclaimer does not repair a headline that means something the data does not support. The fix is a truthful headline, not a footnote.
The prize for doing this properly is real. An outcomes claim that survives a hostile read is the most defensible marketing asset in telehealth, and it becomes more valuable the more the category fills with people who cannot back anything up. The reason to keep the denominator is not only that the FTC says so. It is that the number is only worth having if it is true of the people you are selling to.
Frequently asked questions
Can you advertise your own patient results in telehealth?
Yes, with real constraints. The FTC's guidance says substantiation of health benefits generally requires "randomized, controlled human clinical testing," and that anecdotal evidence and consumer surveys "are never sufficient." Real outcomes data can support a claim if the population you measured resembles the population you market to, and if the reported figure counts the patients who dropped out rather than only those who finished.
What is the denominator problem in outcomes marketing?
Reporting the results of the patients who stayed as if they described everyone who started. It uses a true number to answer a question the reader did not ask. The FTC's guidance calls for an "intent-to-treat" analysis that includes "every subject initially assigned to the treatment and control groups, including subjects who dropped out." A completer-only figure is the opposite of that.
Does a peer-reviewed study make a marketing claim safe?
No. The FTC states directly that "the mere fact that a study is published... isn't a guarantee of quality or proof that the product is effective for the advertised benefit." Publication tells you the paper cleared review. Whether your advertised claim matches what the study actually measured, in the population it measured, is a separate question.
Why does a control group matter for a marketing claim?
Because without one you cannot tell your effect from the drug's, or from self-selection. In STEP 1, the placebo arm lost 2.4%, so the drug's effect was the difference, not the raw figure. A single-arm real-world cohort has no comparison, which means attributing the whole result to your platform is an assumption, not a finding.