How to Read a GLP-1 Trial on ClinicalTrials.gov: A STEP 1 Walkthrough

Every major GLP-1 trial has a public record on ClinicalTrials.gov, and reading it takes about 15 minutes once you know where to look. Find the NCT number, check what the primary outcome was before the trial started, see which estimand the headline number uses, and compare the posted results with the paper and press release. This guide walks through each step using STEP 1, one of the pivotal trials behind Wegovy's weight-loss approval (NCT03548935).

Step 1: Find the NCT number

A ClinicalTrials.gov identifier is "NCT" followed by eight digits. Look for it in a paper's abstract or methods, usually near the funding statement, or in the "about the trial" section of a press release. Lilly's December 2025 release on ATTAIN-MAINTAIN, for example, lists NCT06584916 for that trial and NCT05822830 for SURMOUNT-5.

Search the number on ClinicalTrials.gov, or go straight to the record at clinicaltrials.gov/study/ followed by the number. Our example is STEP 1, NCT03548935, Novo Nordisk's 68-week trial of semaglutide 2.4 mg in adults with overweight or obesity.

Step 2: Read the dates before anything else

The dates tell you whether the trial was registered before it began, and how long results took to appear. Here are STEP 1's, from its registry record.

Source: ClinicalTrials.gov results record, NCT03548935.

Registration before the start date is what you want to see. A trial registered after enrollment began, or after primary completion, gave its sponsor a chance to choose outcomes after seeing early data.

Step 3: Read the primary outcome, word for word

The Outcome Measures section lists the primary, secondary, and other prespecified measures, each with a time frame. According to the registry's own help page, these are "the measurements planned in the study protocol that are used to determine the effects" of the intervention.

STEP 1 had two primary outcomes:

1. Change in body weight (%), baseline (week 0) to week 68.
2. Participants achieving at least 5% body weight reduction, at week 68.

Note three things for any GLP-1 trial: the unit (percent or kilograms), the time point (week 68, 72, or 52), and whether the comparison is against placebo or an active drug. A press release that leads with a different unit or time point from the registered primary outcome is a flag worth chasing.

STEP 1's secondary outcomes included the share of people reaching 10%, 15%, and 20% weight loss, waist circumference, blood pressure, quality of life scores, and HbA1c. Secondary outcomes are useful, but the trial was sized and tested around the primary ones.

Step 4: Figure out which estimand you're looking at

This is where most confusion about GLP-1 results comes from. An estimand is a precise statement of what a trial is estimating, including how it handles "intercurrent events" such as stopping the drug or starting another treatment. The framework comes from the ICH E9(R1) addendum, which ICH regulators adopted in November 2019.

Two strategies show up in nearly every obesity trial:

• Treatment policy counts everyone as randomized, whether or not they kept taking the drug. It answers "what happens if you're prescribed this?"
• Hypothetical (called "trial product" or "efficacy" estimand by some sponsors) estimates what would have happened if everyone had stayed on treatment. It answers "what does the drug do when taken?"

STEP 1's statistical analysis plan defines its primary estimand as the effect in all randomized participants "regardless of adherence to treatment." The secondary estimand covers participants "had they remained on their randomised treatment for the entire planned duration." Missing week-68 weights for the primary estimand were filled in by multiple imputation using participants who left treatment but came back for measurement, across 1,000 imputed datasets.

The hypothetical estimand is almost always the bigger number. Company press releases sometimes lead with it, while journal abstracts typically lead with the treatment-policy figure. Neither is wrong, but you need to know which one you're reading.

Step 5: Open the results and compare them with the paper

When results are posted, the record includes participant flow, baseline characteristics, outcome measures with statistical analyses, and adverse events. STEP 1's participant flow shows 1,306 people started on semaglutide and 655 on placebo; 1,240 and 609 completed.

Here is where it gets instructive. The registry and the published paper report the primary outcome differently.

Sources: ClinicalTrials.gov results, NCT03548935; published treatment-policy figures from Wilding et al., NEJM 2021; trial-product figures from Novo Nordisk, June 4, 2020.

The arm averages don't match, but the treatment differences line up: −12.4 under treatment policy in both the registry and the paper, and about −14.4 to −14.5 under the hypothetical (trial-product) estimand. The registry's observed means describe people who were actually measured. The published estimates come from statistical models that account for missing data. When the differences agree like this, the gap in arm averages is a method difference, not a discrepancy.

The responder outcome lines up too. In the registry's in-trial counts, 1,047 of 1,212 semaglutide participants reached 5% weight loss, which works out to 86.4%. The published treatment-policy figure is also 86.4%, versus 31.5% for placebo (NEJM).

Step 6: Check the record history for outcome switching

Outcome switching means the outcomes reported in a paper differ from the ones registered: a primary outcome is dropped, demoted, or retimed, or new outcomes appear without being declared as new. It matters because switching after seeing data can make a weak result look strong.

It is not rare. The COMPare project checked 67 trials published in NEJM, JAMA, The Lancet, BMJ, and Annals of Internal Medicine. Discrepancies between registered and published outcomes turned up in 58 of them (87%), with a mean of 5.4 undeclared additional outcomes per trial, as published in Trials in 2019. The team's interim tally had counted 301 prespecified outcomes left unreported.

To check a GLP-1 trial yourself:

1. Open the record's Record History tab. NLM reorganized it in the 2023 site redesign so users can view earlier versions of a record. NLM also says registered records are not removed once posted.
2. Open the version current on the study start date and note the primary outcome, its unit, and its time frame.
3. Compare that with the latest version and with the paper's stated primary endpoint.
4. Look for changes made after the primary completion date in particular.

Framework based on the outcome checks used by COMPare (Goldacre et al., Trials 2019).

Step 7: Read the protocol and analysis plan

Many industry trials attach the protocol and statistical analysis plan (SAP) as PDFs in the record's documents section. STEP 1's SAP is version 2.0, dated May 6, 2020, and spells out the primary analysis, every sensitivity analysis, and even the random seed used for imputation.

One honest caveat: that SAP date falls after STEP 1's primary completion date of March 30, 2020. Finalizing an SAP after the last visit but before the data are unblinded is common practice. The registry doesn't record the unblinding date, though, so the dates alone can't settle the question. That is a reason to read carefully, not a sign of a problem.

Putting it to work on a new GLP-1 result

ATTAIN-MAINTAIN shows why this matters. Lilly's December 2025 press release led with efficacy-estimand results. The Nature Medicine paper led with the percent of weight loss maintained under a treatment-regimen estimand: 79.3% versus 37.6% on placebo for people switching from semaglutide. Both are accurate. Knowing which one you're reading, and what the registry promised, is how you compare one GLP-1 headline with the next.

Sources

1. ClinicalTrials.gov: STEP 1, NCT03548935
2. ClinicalTrials.gov results record, NCT03548935
3. STEP 1 statistical analysis plan, version 2.0
4. Wilding JPH et al. Once-weekly semaglutide in adults with overweight or obesity (STEP 1). NEJM 2021
5. Novo Nordisk: STEP 1 topline results (June 4, 2020)
6. ClinicalTrials.gov: How to Read a Study Record
7. NLM Technical Bulletin: Modernized ClinicalTrials.gov website (2023)
8. NLM Support: ClinicalTrials.gov record history
9. Quanticate: Estimands in clinical trials (ICH E9(R1))
10. Goldacre B et al. COMPare: a prospective cohort study correcting and monitoring 58 misreported trials in real time. Trials 2019
11. Retraction Watch: Top journals respond to COMPare findings
12. Eli Lilly: ATTAIN-MAINTAIN topline results (Dec. 18, 2025)
13. Aronne LJ et al. ATTAIN-MAINTAIN. Nature Medicine 2026

Medical disclaimer: This article is general information, not medical advice, and does not replace a licensed clinician who knows your history. Approvals, labels, coverage and prices change; check the linked sources and talk with your prescriber or pharmacist before making decisions about any medication.