How to Read Obesity Drug Trials Without Getting Fooled

This article is about a narrower skill than spotting a misleading marketing claim: it's about opening the actual trial publication behind an obesity drug and reading it the way a reviewer would, rather than the way a press release wants it read. A phase 3 obesity trial costs tens of millions of dollars to run, and once it's done, everyone involved, the sponsor, the investigators, the journal, has some incentive to lead with the most favorable framing available. That isn't necessarily dishonest. It's how publication and promotion work. If you can't read the tables yourself, you're consuming someone else's editorial choices about what to show you.

Start with what was actually measured, and how. The primary endpoint in most obesity trials is something like 'percent change in body weight from baseline at week X,' reported as either a treatment policy estimand (everyone randomized, analyzed regardless of whether they stayed on the drug) or a treatment effect estimand (only people who completed treatment as assigned). The first number is smaller and describes what would happen to a real population offered the drug; the second describes what happens to the subset who tolerate and stick with it. In the pivotal SURMOUNT-1 trial of tirzepatide, the trial-product estimand at the 15 mg dose showed 20.9% mean weight loss at 72 weeks, versus 3.1% with placebo, and the paper reports both estimands explicitly in its results tables [1]. Look for which one a summary is quoting, and prefer the treatment policy number when comparing across sources.

Placebo response is the second thing worth checking before trusting a headline number. In obesity trials, placebo groups routinely lose several percent of body weight, partly from the behavioral effects of trial participation itself, more contact with clinicians, self-monitoring, awareness of being observed. In the STEP 1 trial of semaglutide 2.4 mg, the placebo arm lost 2.4% of body weight over 68 weeks, while the semaglutide arm lost 14.9%; the placebo-subtracted difference, about 12.4 percentage points, is the more honest estimate of the drug's specific contribution than the semaglutide number viewed in isolation [2]. A reported result that never mentions what the comparator group did is missing a necessary piece of context.

Discontinuation rates and reasons matter as much as the headline efficacy number, because a drug that works well but that a large share of people cannot tolerate is a different proposition from one with a smaller effect and high adherence. Trial publications report discontinuation due to adverse events as a specific line item, generally in the range of 4-10% for approved GLP-1-class obesity drugs in registration trials, and it is worth checking whether the primary efficacy analysis is 'intention to treat' (keeping dropouts in the denominator) or 'completers only' (removing them). Completer-only analyses systematically look better because the people who left, often those with the worst side effects or least benefit, are excluded from the picture.

Duration changes what a result means. Weight loss with GLP-1-class drugs typically continues for roughly the first 12 months of treatment before slowing into a plateau; a 16- or 24-week early trial captures the steep part of that curve, not the eventual result. Longer trials matter for a second reason too: durability after treatment stops. The STEP 1 trial extension followed participants for a year after semaglutide was withdrawn and found body weight rose steadily back toward baseline, with roughly two-thirds of the weight lost during treatment regained by 52 weeks off-drug [3]. That single finding reframes how a short trial's endpoint should be read: an early, dramatic number describes response during active treatment, not a durable outcome.

Who was enrolled shapes the result more than most summaries acknowledge, and the eligibility criteria section of a study record, the same section ClinicalTrials.gov's guidance on reading a study record recommends checking first, spells this out directly [5]. Registration trials for obesity drugs commonly exclude people with poorly controlled diabetes, significant psychiatric history, recent major cardiovascular events, or active gastrointestinal disease, in part because comorbidities like diabetes are well documented to blunt the magnitude of GLP-1-associated weight loss compared with metabolically healthier participants. A result from a trial of a relatively healthy population should be read as a description of that population, not a direct prediction for someone with a substantially different health profile.

Endpoint switching is a subtler trap. If a trial's registration on ClinicalTrials.gov lists one primary outcome but a summary or press release foregrounds a different, more favorable finding, that favorable finding was very likely a secondary or exploratory endpoint. Reporting guidance from the CONSORT group, the standard framework for how randomized trials should be written up, specifically calls for pre-specifying primary versus secondary outcomes and reporting all of them, precisely because post hoc emphasis on a flattering secondary result is one of the more common ways trial findings get overstated in translation to the public [4].

Funding source and conflicts of interest, disclosed in the methods or a dedicated statement near the end of most publications, are worth noting without treating them as disqualifying. The large majority of pivotal obesity drug trials are funded by the manufacturer, since these trials are expensive and manufacturers have the strongest incentive to run them; that alone doesn't make the data false, and these trials go through peer review and regulatory audit before a drug is approved. But sponsorship is a documented factor associated with more favorable framing of results in the broader clinical trial literature, which is one more reason to read the outcome tables directly rather than relying solely on the sponsor's own summary of what those tables show.

Adverse event tables reward a careful read rather than a skim. Distinguish serious adverse events (hospitalization, disability, death) from the much larger category of any adverse event (which includes mild, transient nausea). Look specifically at events of special interest that are pre-specified because of a plausible mechanism, pancreatitis and gallbladder disease for GLP-1-class drugs, and note what a trial of a few hundred to a few thousand participants can and cannot detect: it is adequately powered to characterize common effects but has essentially no power to detect a genuinely rare event occurring in 1 in 2,000 or 1 in 10,000 people. That is the specific job of postmarketing surveillance after approval, which is one reason regulators and clinicians treat newly approved drugs, and especially unapproved or investigational compounds, with more caution than long-marketed ones.

Confidence intervals and subgroup analyses are worth a specific mention, because they're where a lot of Phase 2 excitement quietly outruns what the data can support. A mean weight loss figure with a wide confidence interval, common in smaller trials, means the true population effect could plausibly be meaningfully higher or lower than the headline number; a narrower interval, more typical of large Phase 3 trials, indicates a more precisely estimated effect. Subgroup analyses, showing that a drug worked especially well in, say, women, or people under 50, or a particular baseline weight range, are exploratory by nature unless the trial was specifically designed and powered to test that subgroup comparison in advance. A subgroup finding highlighted in a press release without that pre-specification is a hypothesis worth testing in a future trial, not a confirmed result for that subgroup.

Trial registries themselves are a resource worth using directly rather than relying entirely on someone else's summary. ClinicalTrials.gov entries show the originally registered primary and secondary outcomes, the planned sample size, and, once available, a results section with the actual numbers reported directly to the registry, independent of how a journal article or press release chooses to frame them. Comparing what a trial's registry entry says it planned to measure against what a later publication emphasizes is one of the more direct ways to catch endpoint switching described above, and it doesn't require any specialized statistical training, only the patience to look up the NCT number and read the entry.

None of this is a substitute for a clinician's interpretation of how a specific trial's findings might or might not apply to an individual person's situation, and reading a trial well does not turn a lay reader into a substitute for peer review. But a practical habit is worth building regardless: open the publication, find the baseline characteristics table, the primary and secondary outcomes table, and the adverse events table, and read those three before reading the abstract or discussion section. The abstract and discussion exist partly to interpret and frame the results for a reader; the tables report what was actually observed with much less room for framing.