Inside the STEP-2 Methodology: What Most Summaries Skip

At a glance

| Parameter | Detail | |-----------|--------| | N | 1,210 randomized (1:1:1 sema 2.4 mg : sema 1.0 mg : placebo) | | Intervention | Subcutaneous semaglutide 2.4 mg once weekly | | Comparator | Semaglutide 1.0 mg once weekly; matched placebo | | Duration | 68 weeks (16-week dose escalation + 52-week maintenance) | | Primary endpoint | Percent change in body weight from baseline to week 68 | | Key result | -9.6% (sema 2.4 mg) vs -3.4% (placebo); ETD -6.2 percentage points (p < 0.0001) | | Population | Adults with BMI 27+ and type 2 diabetes (HbA1c 7-10%) |

Why STEP-2 Exists Separately from STEP-1

STEP-1 enrolled adults with obesity but excluded type 2 diabetes (T2D). This matters because patients with T2D consistently lose less weight on any intervention, likely due to insulin resistance, concomitant medications that promote weight gain (sulfonylureas, insulin), and altered incretin physiology. The STEP-2 investigators designed a standalone trial in this harder-to-treat population rather than running a subgroup analysis. That choice raises the statistical bar: a dedicated confirmatory trial with its own sample-size calculation, its own alpha spending, and its own regulatory submission pathway.

The Three-Arm Design and Randomization Details

Participants were randomized 1:1:1 to semaglutide 2.4 mg, semaglutide 1.0 mg, or placebo. All groups received identical-appearing injections and the same lifestyle intervention (500 kcal/day deficit counseling plus 150 min/week physical activity). The 1.0 mg arm served a dual purpose: it functioned as an active comparator already approved for glycemic control (Ozempic FDA label), and it allowed regulators to see the dose-response curve within a single trial rather than across separate studies.

Randomization used an interactive web-response system stratified by:

• Background diabetes medication (metformin alone vs. metformin + another oral agent)
• HbA1c category (<8.5% vs. 8.5-10%)
• Country

These strata matter because sulfonylurea or SGLT2 inhibitor co-administration could independently affect weight trajectory and hypoglycemia risk. Stratification ensures balance on these prognostic factors across arms.

Blinding Mechanics

The trial was double-blind at the participant, investigator, and sponsor-analysis level. Semaglutide and placebo pens were visually identical. The 16-week dose-escalation schedule (starting at 0.25 mg, stepping through 0.5, 1.0, and 1.7 mg before reaching 2.4 mg) applied equally to both active arms and placebo, preserving the blind during a period when GI side effects might otherwise unmask allocation.

One subtlety: the 1.0 mg arm stopped escalation at week 8 and remained on 1.0 mg, while the 2.4 mg arm continued escalation through week 16. Both placebo sub-groups mirrored the respective escalation schedules. This means two "flavors" of placebo existed within the single placebo arm, a design detail the primary publication notes in its supplementary appendix but that most summaries skip.

The Estimand Framework: Two Parallel Analyses

STEP-2 adopted the ICH E9(R1) estimand framework before most obesity trials did. Two estimands were pre-specified:

Treatment-policy estimand. Includes all data regardless of treatment discontinuation or rescue medication. This answers: "What happens when you prescribe semaglutide 2.4 mg in clinical practice, including patients who stop early?" Data after intercurrent events (discontinuation, rescue) are still included. Missing data are handled via multiple imputation under a missing-at-random assumption drawn from the same treatment group.

Trial-product estimand. Censors data after participants discontinue study drug or receive rescue medication. This answers: "What is the pharmacologic effect of semaglutide 2.4 mg in patients who remain on treatment?" Missing data here are imputed from same-arm on-treatment participants using a mixed model for repeated measures (MMRM).

The headline -9.6% comes from the treatment-policy estimand. Under the trial-product estimand, the loss was -10.6% vs. -3.7% for placebo, a wider separation that reflects the pharmacologic ceiling when adherence is perfect. Clinicians reading only the abstract get the first number; prescribers counseling patients about what to expect if they stick with the injection for a full year should reference the second.

Primary Endpoint Definition

Percent change in body weight from baseline to week 68, confirmed by the treatment-policy estimand as the regulatory primary. The co-primary endpoint was the proportion of participants achieving 5% or greater weight loss. Both had to be significant for the trial to succeed, with a hierarchical testing procedure to control the family-wise type I error rate at 5%.

Results on the co-primary:

| Arm | Achieved 5%+ loss | Achieved 10%+ loss | Achieved 15%+ loss | |-----|-------------------|--------------------|--------------------| | Semaglutide 2.4 mg | 68.8% | 45.6% | 25.8% | | Semaglutide 1.0 mg | 57.1% | 29.3% | 13.0% | | Placebo | 28.5% | 8.2% | 3.2% |

The 10% and 15% thresholds were confirmatory secondary endpoints tested in a pre-specified hierarchical order. All passed. These categorical thresholds matter clinically because ADA Standards of Care tie surgical referral discussions to the 15% mark and metabolic benefit inflection points to 5-10%.

Statistical Analysis Plan Specifics

The primary analysis used an ANCOVA model with treatment, stratification factors, and baseline body weight as covariates. For the treatment-policy estimand, a pattern-mixture model handled missing data. The sample size of 1,210 provided greater than 90% power to detect a 4-percentage-point difference between semaglutide 2.4 mg and placebo, assuming 15% dropout and an SD of 8%.

Multiplicity was managed through a pre-specified hierarchical testing sequence:

1. Percent weight change (2.4 mg vs. placebo)
2. Proportion achieving 5%+ loss (2.4 mg vs. placebo)
3. Percent weight change (2.4 mg vs. 1.0 mg)
4. Proportion achieving 10%+ loss (2.4 mg vs. placebo)
5. Proportion achieving 15%+ loss (2.4 mg vs. placebo)

Each step had to pass at the 5% two-sided level before the next could be formally tested. All five passed. This rigid gate-keeping procedure is conservative but protects against inflated false-positive claims on secondary endpoints.

Inclusion and Exclusion Criteria That Shaped the Population

Key inclusion criteria:

• Age 18+ years
• BMI 27 kg/m² or greater
• HbA1c 7.0-10.0% (managed with diet/exercise alone, metformin, or metformin + one additional oral agent)
• Stable body weight (self-reported <5 kg change in 90 days)

Key exclusions:

• Injectable diabetes medications (GLP-1 RAs, insulin) within 90 days
• History of bariatric surgery
• eGFR <30 mL/min/1.73 m²
• History of pancreatitis
• Proliferative retinopathy or maculopathy requiring acute treatment

The HbA1c ceiling of 10% excluded poorly controlled patients who would likely need insulin initiation during the trial (confounding weight outcomes). The floor of 7% excluded well-controlled patients who might have had less metabolic derangement and thus potentially different weight-loss trajectories. This creates a defined "moderate T2D" population. Clinicians extrapolating these results to patients on basal insulin or with HbA1c above 10% should do so cautiously.

How Background Medications Were Handled

Participants continued their baseline diabetes medications throughout. If hypoglycemia occurred on a sulfonylurea, investigators could reduce its dose per protocol guidelines. This is realistic for clinical practice but introduces a minor confounder: dose reductions of sulfonylureas could independently reduce weight (sulfonylureas are weight-promoting). The protocol addressed this by pre-specifying that any such changes would be captured but not censored in the treatment-policy estimand.

Limitations the Authors Acknowledged

The trial ran 68 weeks, so durability beyond 16 months is unknown from this dataset alone. The STEP-5 extension (2 years) addressed duration but in a non-diabetic cohort. The population was 55% White and conducted primarily in North America, Europe, and East Asia; applicability to other groups requires assumption. The 16-week dose escalation means the true maintenance-dose exposure was 52 weeks, not 68. Dropout was 12.9% in the 2.4 mg arm, with GI adverse events (nausea, diarrhea, vomiting) as the most common reason for early discontinuation.

The exclusion of injectable therapies means STEP-2 does not answer whether adding semaglutide 2.4 mg to basal insulin produces comparable weight loss. SUSTAIN 6 showed cardiovascular benefit with semaglutide 1.0 mg in patients on insulin, but weight loss was more modest in that insulin-treated subgroup.

Interpreting the Comparator Choice

Including semaglutide 1.0 mg as an active comparator was strategically important. It allowed a within-trial demonstration that the higher dose produces clinically meaningful incremental benefit (-9.6% vs. -7.0%). The 2.6-percentage-point difference between 2.4 mg and 1.0 mg was statistically significant (p = 0.0003). For a clinician already prescribing Ozempic 1.0 mg for glycemia, this quantifies what the patient gains by escalating to the obesity-indication dose (Wegovy FDA label).

Glycemic Outcomes as Secondary Endpoints

Though not the primary endpoint, HbA1c change was a key secondary: -1.6 percentage points with semaglutide 2.4 mg vs. -0.4 with placebo. The proportion achieving HbA1c <6.5% was 47% vs. 12%. These glycemic effects complicate the interpretation of "pure" weight-loss efficacy because improved insulin sensitivity from weight loss and direct GLP-1 receptor activation on beta cells are pharmacologically entangled. The trial cannot separate these two mechanisms, and the estimand framework does not attempt to.

What This Means for Evidence-Based Prescribing

The dual-estimand structure gives prescribers two clinically distinct numbers to reference. When counseling a newly started patient about expected weight loss if they tolerate the drug for a year: cite the trial-product estimand (-10.6%). When modeling population-level outcomes in a health system formulary review: cite the treatment-policy estimand (-9.6%). The gap between these numbers is smaller than in many cardiovascular trials, reflecting relatively high adherence in STEP-2 (87% completed treatment in the 2.4 mg arm). In real-world practice, where adherence is typically lower, the gap would widen.

Frequently asked questions

›

›

›

›

›

›

›

›

›

›

References

1. Davies M, Færch L, Jeppesen OK, et al. Semaglutide 2.4 mg once a week in adults with overweight or obesity, and type 2 diabetes (STEP 2): a randomised, double-blind, double-dummy, placebo-controlled, phase 3 trial. Lancet. 2021;397(10278):971-984. PubMed
2. Wegovy (semaglutide) prescribing information. FDA. 2021. Label
3. Ozempic (semaglutide) prescribing information. FDA. 2020. Label
4. Marso SP, Bain SC, Consoli A, et al. Semaglutide and cardiovascular outcomes in patients with type 2 diabetes (SUSTAIN 6). N Engl J Med. 2016;375:1834-1844. PubMed
5. American Diabetes Association Professional Practice Committee. Standards of Care in Diabetes, 2023. Diabetes Care. 2023;46(Suppl 1). PubMed