On this page
Updated and methods-checked: 6 October 2026.
The headline percentage is never enough to understand a weight-loss drug trial. A result depends on who entered the study, what the comparator was, how long participants were followed, what happened after treatment stopped, how missing measurements were handled and which adverse events caused discontinuation.
This guide provides a repeatable way to read obesity and metabolic-treatment trials without treating unlike studies as if they were a league table. It applies to trials of GLP-1 medicines, dual and triple agonists, amylin combinations and other investigational approaches.
The one-minute result decoder
| Headline phrase | Question to ask immediately | Why it matters |
|---|---|---|
| “Lost X% of body weight” | Change from baseline or placebo-adjusted difference? | Those are different effect sizes. |
| “At week 72” | How much time was spent titrating, and how many reached the endpoint? | Duration and exposure shape the result. |
| “Regardless of discontinuation” | Was this a treatment-policy estimand? | It estimates assignment to treatment, not uninterrupted use. |
| “If treatment was continued” | Was this a hypothetical or efficacy estimand? | It may not describe real-world discontinuation. |
| “Most common adverse events” | What were serious events and events causing discontinuation? | Frequency alone does not show severity or tolerability. |
| “Drug A beats Drug B” | Was this a randomised head-to-head trial? | Separate trials often use incompatible populations and methods. |
Read left to right. If one link is unclear, the percentage alone is incomplete.
Who entered, and who was excluded?
Placebo, active treatment or usual care?
How were stopping and rescue treatment handled?
Mean change, threshold response or another measure?
What was observed, imputed or excluded?
Text equivalent: verify population, comparator, estimand, endpoint and missing-data method before interpreting a weight-loss result.
The eight-question trial check
| Question | Why it changes the result | What to record |
|---|---|---|
| 1. Who was studied? | Diabetes status, baseline weight, age and comorbidities affect expected outcomes. | Eligibility criteria, sample size and baseline characteristics. |
| 2. What was the comparison? | Placebo, lifestyle support and an active medicine answer different questions. | Comparator and co-interventions. |
| 3. What effect was estimated? | A result assuming continued treatment differs from a result regardless of discontinuation. | Estimand, analysis population and treatment of intercurrent events. |
| 4. Which endpoint was primary? | Mean percentage change and responder thresholds describe different aspects of benefit. | Endpoint definition, time point and multiplicity plan. |
| 5. How long did the study run? | Titration, plateau, maintenance and regain occur on different timelines. | Titration and treatment periods, follow-up after treatment. |
| 6. Who stopped treatment or was lost? | Missing outcomes and discontinuation can move the estimate. | Participant flow, reasons for stopping and missing-data method. |
| 7. How was safety counted? | “Any adverse event”, serious events and events causing discontinuation are not equivalent. | Denominators, exposure time, severity and discontinuations. |
| 8. Can it be compared with another trial? | Different populations, doses, estimands and durations can make a numerical comparison misleading. | Direct head-to-head evidence or an explicit comparability warning. |
1. Start with the population, not the drug name
A trial in adults with obesity but without diabetes does not answer exactly the same question as a trial in people with type 2 diabetes. Baseline weight, prior therapy, cardiovascular or kidney disease, sex distribution, age and the intensity of lifestyle support can all change the observed average.
Before reading the result, record:
- the number randomised and the number included in each analysis;
- whether participants had type 2 diabetes;
- baseline body mass index and body weight;
- major inclusion and exclusion criteria;
- relevant cardiovascular, kidney, liver or sleep-apnoea conditions;
- the demographic and geographic mix of participants.
A large percentage reduction in one population cannot be transplanted into another population without evidence.
2. Identify the comparator and the background programme
“Placebo-controlled” rarely means that one group received a medicine and the other received nothing. Many obesity trials provide both groups with structured lifestyle counselling. The intensity and adherence to that programme matter.
An active-comparator trial is more useful for comparative efficacy because the treatments are randomised within the same protocol. Even then, readers should compare the studied regimens and populations, not every possible use of the two drugs.
3. Find the estimand: what treatment effect is the trial trying to estimate?
The International Council for Harmonisation’s ICH E9(R1) addendum distinguishes treatment discontinuation from study withdrawal. Discontinuation is an intercurrent event: something after randomisation that affects how the outcome should be interpreted. Study withdrawal can create missing data.
Different strategies answer different questions about events after randomization:
- Treatment-policy strategy: estimate the treatment effect regardless of the specified intercurrent event, such as treatment discontinuation. Continued outcome collection remains important.
- Hypothetical strategy: estimate an outcome under a stated scenario in which the intercurrent event would not occur. The assumptions defining that scenario matter.
- While-on-treatment strategy: focus on responses before the specified intercurrent event. Restricting the observation period changes the question; it is not the same strategy as a hypothetical scenario.
Labels such as “efficacy” or “on treatment” must be checked against the paper’s actual definition. A trial can use different strategies for different intercurrent events. A completers-only analysis is not automatically a valid implementation of any of these strategies. Clarification checked 26 October 2026 against ICH E9(R1), section A.3.2.
4. Understand the primary weight endpoints
The US Food and Drug Administration’s January 2025 draft guidance for weight-reduction drug development discusses long-term body-weight reduction and maintenance. Modern obesity trials commonly use endpoints such as:
- mean percentage change in body weight from baseline;
- the proportion reaching at least a specified percentage reduction;
- change at a prespecified week, often after a titration period;
- maintenance of prior weight reduction;
- disease-specific outcomes such as cardiovascular events, sleep-apnoea measures or functional symptoms.
The mean shows the average response, not what every participant experienced. Responder thresholds show how many crossed a line, but not the full distribution. Absolute kilograms and percentage change answer related but different questions, especially across different starting weights.
The primary endpoint and time point should be prespecified. Secondary endpoints can be valuable, but multiple testing increases the chance of a nominally positive result unless the analysis controls for it.
FDA draft guidance: obesity and overweight drug development
Percentages, percentage points and relative risk
Start by identifying the outcome. Percentage body-weight change is a continuous outcome. The proportion of people who experience a defined event is an event risk. A relative difference between average weight losses is not a relative reduction in event risk.
Assume 1,000 people in each group, the same fixed follow-up, and 20 comparator-group people versus 16 treatment-group people experiencing the event. These are illustrative counts, not findings from an obesity-drug trial.
The relative reduction uses the comparator risk as its denominator. Reporting “20% lower risk” alone hides the baseline risk and follow-up period. In this illustration, it does not mean 20 fewer events per 100 people. The calculations show arithmetic only; they do not supply a trial’s uncertainty estimates.
CONSORT 2025, item 26, calls for group results, denominators and precision, and both absolute and relative effect sizes for binary outcomes. For a real result, read the confidence interval for the between-group effect as well as the result in each group.
A hazard ratio comes from a time-to-event analysis and should not be treated as interchangeable with a ratio of cumulative event risks. Use the measure actually reported, together with its population and time horizon. Likewise, subtracting two mean percentage weight changes yields a difference in percentage points; it does not estimate a clinical-event risk reduction.
For a real head-to-head weight-change example, see the SURMOUNT-5 estimates and their confidence intervals. Source check for this effect-size explanation: 26 October 2026. The guide’s historical full methods-review dates remain unchanged.
5. Put duration and titration on the same timeline
A 24-week result and a 72-week result are not directly comparable. Many incretin trials gradually increase treatment during an initial titration period. The time at a maintenance regimen, the point at which weight change begins to plateau and the duration of follow-up all affect the endpoint.
Record three dates separately:
- randomisation;
- the end of titration or escalation;
- the primary endpoint and any post-treatment follow-up.
A study of initial loss does not establish long-term maintenance. A withdrawal study answers a different question from a continuous-treatment study.
For worked examples of those designs, read weight regain after stopping GLP-1 medicines: what withdrawal trials show. Keep the randomized-withdrawal and off-treatment follow-up denominators separate.
6. Read the participant-flow diagram and missing-data method
A trial may randomise a large group but have fewer measured outcomes at the final visit. The reasons matter: adverse events, lack of efficacy, withdrawal of consent, loss to follow-up and administrative closure are not interchangeable.
Ask:
- How many participants discontinued the assigned treatment?
- How many remained in follow-up after stopping treatment?
- Were post-discontinuation measurements collected?
- How many final outcomes were observed rather than imputed?
- What assumptions were used for missing measurements?
- Did sensitivity analyses reach similar conclusions?
A completers-only calculation can overstate benefit if participants who stop treatment have poorer outcomes. Conversely, a highly conservative imputation can answer a deliberately different question. The statistical method must match the estimand.
7. Separate tolerability, serious harm and discontinuation
Safety tables use several denominators and categories. “Participants with any adverse event” can be high in both treatment and placebo groups and does not by itself describe severity. Readers should distinguish:
- any treatment-emergent adverse event;
- common events by type and severity;
- serious adverse events;
- events leading to treatment discontinuation;
- events of special interest identified before analysis;
- deaths, with cause and investigator assessment;
- laboratory or physiological changes that may not be reported as symptoms.
Exposure time matters when follow-up differs. A higher discontinuation rate can also reduce later exposure and complicate a simple count comparison.
8. Prefer head-to-head trials over cross-trial ranking
The strongest comparison randomises both treatments in the same protocol. In the 72-week SURMOUNT-5 trial, tirzepatide and semaglutide were compared directly in adults with obesity without diabetes. The peer-reviewed report found a greater mean percentage weight reduction with tirzepatide and reported treatment discontinuation because of adverse events in 6.1% and 8.0% of participants, respectively.
That conclusion belongs to the studied population, regimens and 72-week protocol. It does not make every tirzepatide trial comparable with every semaglutide trial, and it does not establish an individual treatment choice.
Aronne et al., tirzepatide compared with semaglutide for obesity
Why common cross-trial comparisons fail
A concrete example is why REDEFINE 1 and REDEFINE 2 should not be directly subtracted: each trial randomized within a different population, so subtracting their average weight changes does not establish a causal effect of diabetes.
| Apparent comparison | Hidden mismatch | Better approach |
|---|---|---|
| “Drug A lost 20%; Drug B lost 15%” | Different populations, durations, comparators or missing-data assumptions. | Use a direct randomised comparison or label the figures as non-comparable. |
| “More side effects” | Different reporting windows, definitions and exposure. | Compare specific event categories and discontinuation within the same protocol. |
| “The higher dose worked best” | Participants may not have reached or tolerated the assigned maintenance regimen. | Read achieved exposure, titration and estimand. |
| “Everyone lost the average amount” | A mean hides the response distribution. | Read responder thresholds, variability and individual-level plots when available. |
| “The press release proves the trial” | Topline releases omit methods and complete safety data. | Wait for a registry update, regulator review or peer-reviewed report. |
A worked reading sequence
- Open the trial-registry record and note the original primary endpoint, population and completion date.
- Read the peer-reviewed methods before the abstract conclusion.
- Find the participant-flow diagram and the number with observed endpoint data.
- Identify the primary estimand and the missing-data method.
- Read the effect estimate with its confidence interval, not only the percentage in the headline.
- Check serious events and events leading to discontinuation.
- Compare the paper with the registry and regulator documents for outcome switching or status differences.
- Write the limitation beside the result, not several paragraphs later.
Blank extraction worksheet
Use this compact record for every trial. If a field is missing from the abstract, check the full paper, supplement, registry and regulator documents before drawing a conclusion.
| Field | What to extract |
|---|---|
| Trial identity | Registry ID, sponsor, phase, countries and completion date |
| Population | Number randomised, diabetes status, baseline weight/BMI, major exclusions |
| Comparison | Placebo or active comparator, background lifestyle programme |
| Primary endpoint | Exact measure, time point and whether it was prespecified |
| Estimand | Treatment-policy, hypothetical, on-treatment or another stated strategy |
| Exposure | Titration period, achieved regimen and follow-up duration |
| Participant flow | Stopped treatment, withdrew from study, lost to follow-up, observed outcomes |
| Missing data | Imputation method and sensitivity analyses |
| Effect | Estimate, comparator-adjusted difference, confidence interval and responder distribution |
| Safety | Common, serious and discontinuation-causing events with denominators |
| Transfer limits | Which people, time horizon or regimens the conclusion does not cover |
| Source status | Registry, press release, conference report, peer-reviewed paper or regulator review |
Questions to ask before sharing a result
- Am I describing assignment to treatment, continued treatment or completers?
- Is the result placebo-adjusted or change from baseline?
- Was the study designed for this endpoint?
- How many participants stopped treatment, and why?
- Are the safety figures based on all treated participants?
- Is this a direct comparison or a comparison across separate trials?
- Is the source a registry, press release, conference presentation, peer-reviewed paper or regulator review?
- Does the conclusion stay within the studied population and time horizon?
Frequently asked questions
Is average percentage weight loss the amount everyone lost?
No. A mean is a group average. Read responder thresholds, variability and individual-level plots when available.
What is an estimand in a weight-loss trial?
An estimand precisely defines the treatment effect being estimated, including how events such as treatment discontinuation or rescue therapy are handled.
Why can two papers report different percentages for the same trial?
They may use different estimands, analysis populations, missing-data assumptions, time points or comparator adjustments.
Can semaglutide and tirzepatide percentages be compared across separate trials?
Only cautiously. A direct randomised head-to-head trial is stronger because the population, duration, comparator and analysis are shared.
Does a positive Phase 3 result mean a medicine is approved?
No. Trial results and regulatory authorisation are separate milestones. Approval and Australian availability must be checked with the relevant regulator.
Related Elite Peps resources
Follow current readouts: our Obesity Drug Clinical Trial Results Tracker separates fresh Phase 2/3 results, company topline announcements and regulatory status for retatrutide, CagriSema, survodutide and other major programs.
- Phase 1, 2, 3 and 4 trials: what each stage can and cannot establish
- How to read a clinical-trial registry record
- Press release versus peer-reviewed paper
- Tirzepatide versus semaglutide: the SURMOUNT-5 head-to-head trial
- Retatrutide trial tracker
Primary references
- ICH E9(R1): Estimands and Sensitivity Analysis in Clinical Trials
- FDA draft guidance on obesity and overweight drug development, January 2025
- SURMOUNT-5: tirzepatide compared with semaglutide
- STEP 1: once-weekly semaglutide in adults with overweight or obesity
- Phase 2 retatrutide trial in obesity
Methods checked 29 September 2026. This guide will be updated when major trial-reporting or regulatory guidance changes.
How to cite this article
How to Read a Weight-Loss Drug Trial: Endpoints, Dropouts, Safety and Comparisons. Elite Peps Australia. Published 29 September 2026; updated 26 October 2026. https://elitepepsaustralia.com/how-to-read-weight-loss-drug-clinical-trial/
Source checks: Consult the primary studies, trial records and regulator documents linked in this article for current evidence and status. Publication and update dates do not by themselves indicate an evidence review.