When the Outcome Measure Doesn't Match the Compound
A mid-size CRO running a Phase 1b extension for a multi-receptor peptide agonist recently hit a design wall that has become increasingly common in the GLP research space: the sponsor's statistical analysis plan specified percent body weight change at week 24 as the primary endpoint, borrowed almost verbatim from the STEP 1 semaglutide protocol (Wilding et al., NEJM 2021;384:989-1002, PMID 33567185). The compound in question, an early-stage molecule described in preclinical literature as acting on an expanded receptor set beyond canonical GLP-1 signaling — a class some researchers now informally label "GLP-3" activity — did not behave like semaglutide on the timeline the endpoint assumed. Weight change plateaued earlier, but a secondary marker (fasting insulin sensitivity) kept improving through week 36.
The trial was not underpowered. It was mis-specified. The endpoint described a phenomenon the compound produced on a different curve than the reference drug, and the data monitoring committee had no pre-specified way to interpret a dissociation between the primary and secondary signal. This is the recurring failure mode in outcome measure selection for emerging peptide classes: endpoints get inherited from the last successful trial design rather than derived from the pharmacology of the compound being tested. This article lays out a systematic approach to selecting outcome measures for GLP-3-class and related multi-receptor peptide trials, using the endpoint architecture of established GLP-1 and dual-agonist programs (semaglutide, tirzepatide, retatrutide) as the evidentiary baseline, since no GLP-3-specific pivotal trial has yet reached publication.
What "GLP-3" Refers to in the Current Research Literature
There is no FDA-recognized receptor designated GLP-3, and no approved therapeutic carries that label. The term circulates in early discovery-stage literature as informal shorthand for peptide constructs that extend beyond single GLP-1 receptor (GLP-1R) agonism into additional incretin or glucagon-family receptor activity — conceptually adjacent to how retatrutide is described as a GIP/GLP-1/glucagon triple receptor agonist (Jastreboff et al., NEJM 2023;389:514-526, PMID 37366315; ClinicalTrials.gov NCT04867785). Some preclinical groups use "GLP-3-like activity" to describe compounds engaging a third, less-characterized receptor target alongside GLP-1R and GIPR, but binding data supporting a distinct receptor of that name has not cleared peer review at pivotal-trial scale.
This matters directly for outcome measure selection: a trial designed around an incompletely characterized receptor profile cannot borrow a primary endpoint wholesale from a drug with a fully mapped mechanism. Sponsors moving compounds in this class into Phase 1/2 work are, in effect, running endpoint discovery and efficacy testing simultaneously, which raises the bar on how outcome measures are chosen and pre-specified. Where this article uses "GLP-3 peptide therapy," it refers to this early-stage, multi-receptor peptide category — not an approved drug class — and every recommendation below assumes the corresponding regulatory uncertainty. Readers evaluating a specific compound should confirm its actual receptor pharmacology and trial phase directly from the sponsor's ClinicalTrials.gov registration rather than assuming GLP-1 precedent applies.
Why Outcome Measure Selection Is the Central Design Problem
Outcome measure selection is not a downstream administrative step after the "real" scientific decisions are made — it is the decision that determines whether a trial can detect the effect the compound actually produces. The FDA's 2007 Guidance for Industry on developing products for weight management specifies that obesity-indication trials should demonstrate either mean placebo-subtracted weight loss of at least 5% with statistical significance, or a categorical responder rate showing at least 35% of treated subjects losing 5% or more body weight versus placebo (FDA, Guidance for Industry: Developing Products for Weight Management, 2007). Both thresholds assume weight loss is the dominant, first-order effect — reasonable for GLP-1R monoagonists, less safe for a compound whose expanded receptor engagement might produce its clearest signal in glycemic variability, lipid partitioning, or lean mass preservation before weight change catches up statistically.
In SURMOUNT-1, tirzepatide's primary endpoints were percent change in body weight and proportion of participants achieving ≥5% weight reduction at 72 weeks, with the highest dose (15 mg) producing a mean reduction of 20.9% versus 3.1% for placebo (n=2,539; Jastreboff et al., NEJM 2022;387:205-216, PMID 35658024). That endpoint worked because dual GIP/GLP-1R agonism produces weight loss as a proximate, dose-linear effect. A compound with a slower-acting or non-adipose-first mechanism tested against the same endpoint at the same timepoint risks a false negative — not because the drug lacks efficacy, but because the outcome measure targets the wrong physiological readout.
The practical rule that follows: outcome measures should be derived from the compound's demonstrated PK/PD curve in Phase 1, not imported from the most recent successful trial in an adjacent drug class.
Primary Endpoint Architecture: Lessons from STEP and SURMOUNT
STEP 1 (n=1,961) established the now-standard primary endpoint pairing for GLP-1R agonist obesity trials: percent change in body weight from baseline to week 68, and achievement of ≥5% weight reduction, both versus placebo with intention-to-treat analysis. Semaglutide 2.4 mg produced a mean change of -14.9% versus -2.4% for placebo (P<0.001), with 86.4% of treated participants reaching the 5% threshold versus 31.5% on placebo. That two-endpoint structure — a continuous measure plus a categorical responder threshold — has become close to a regulatory default because it satisfies FDA's dual requirement (mean effect size and responder proportion) in a single co-primary structure.
For a GLP-3-class compound, the same architecture can serve as a starting template, but three modifications are typically warranted based on what Phase 1b PK/PD data shows. First, the assessment window may need to extend or contract relative to the 68-week STEP timeline depending on terminal half-life — semaglutide's is approximately 7 days, tirzepatide's approximately 5 days — since a peptide with different depot kinetics may plateau earlier, meaning a fixed 68-week readout could bury an earlier peak effect inside a stable maintenance phase. Second, the responder threshold itself (5%, 10%, 15% categories) should be chosen based on where the Phase 1b dose-response curve showed the steepest separation from placebo, not assumed from precedent. Third, when a compound's mechanism plausibly affects a non-weight parameter first (HbA1c or lean-to-fat mass ratio, for instance), a true co-primary — not a secondary — endpoint structure should be pre-specified, since FDA has accepted composite or dual co-primary endpoints in diabetes and obesity programs when justified by mechanism (FDA, Guidance for Industry: Diabetes Mellitus — Developing Drugs and Therapeutic Biologics for Treatment and Prevention, 2008).
Secondary Endpoints and Composite Outcome Design
Secondary endpoints in GLP-1-class trials typically cluster into four domains: cardiometabolic markers (waist circumference, blood pressure, lipid panel, HbA1c), body composition (DXA-measured fat mass and lean mass), patient-reported outcomes (IWQOL-Lite score, SF-36), and safety-tolerability composites (discontinuation rate, gastrointestinal event frequency). SURMOUNT-1 tracked all four domains and reported a mean waist circumference reduction of 18.4 cm at the highest tirzepatide dose versus 4.6 cm for placebo — a magnitude large enough to function as supportive evidence even where the primary endpoint alone might not fully characterize metabolic benefit.
For an emerging multi-receptor peptide, the secondary endpoint set needs to be built around the specific receptors the compound is thought to engage rather than copied wholesale from the GLP-1R monoagonist template. A compound with plausible glucagon receptor activity, for instance, warrants endpoints tracking resting energy expenditure and hepatic fat fraction (MRI-PDFF), since glucagon receptor agonism has a documented thermogenic and lipolytic signature in mechanistic studies that a pure GLP-1R agonist would not be expected to match.
Composite endpoints deserve particular caution. A composite that averages together a weight metric and a glycemic metric can mask a dissociation between the two — exactly the failure pattern described in the opening scenario, where the CRO's insulin sensitivity signal kept improving after weight change plateaued. When mechanism suggests a compound may decouple these two physiological processes, the pre-specified analysis plan should treat them as separate secondary endpoints with independent statistical testing, not folded into a single composite score, so a differential effect remains visible rather than averaged away.
Population Heterogeneity and Sub-Group Signal Detection
Trial populations for weight-management and metabolic peptide studies are rarely as homogeneous as the topline number suggests. STEP and SURMOUNT both pre-specified subgroup analyses by baseline BMI category, sex, age, race, and diabetes status, and the subgroup effect size in several cases diverged meaningfully from the overall population estimate — a signal worth building into outcome measure selection from the start rather than treating as a post-hoc curiosity.
For a GLP-3-class compound in early-phase testing, three subgroup considerations are worth pre-specifying rather than discovering after unblinding. Baseline insulin resistance status (HOMA-IR) may predict differential response if the mechanism includes a glucagon or GIP-adjacent pathway, since these receptors affect hepatic glucose handling differently across insulin-sensitive and insulin-resistant strata. Prior GLP-1R agonist exposure matters too — patients who discontinued semaglutide or liraglutide for tolerability reasons may show a different adverse-event profile than treatment-naive participants, and pooling these groups without stratification can obscure a safety signal. Sex-based differences in body composition response have also appeared repeatedly across the GLP-1 literature; enrollment targets should be set with subgroup power calculations, not just overall sample size, when heterogeneity is plausible.
Adaptive Designs and Interim Analysis Triggers
Because early-phase multi-receptor peptide compounds carry more mechanistic uncertainty than a fourth-generation GLP-1R monoagonist, adaptive trial designs offer a practical way to correct outcome measure mis-specification before it sinks an entire program. A group sequential design with a pre-specified interim analysis at 20-30% of planned enrollment, using an O'Brien-Fleming alpha-spending boundary, allows a data monitoring committee to evaluate whether the primary endpoint is tracking the expected effect size without inflating the overall type I error rate beyond the pre-specified 0.05 threshold.
Response-adaptive randomization is a further option worth pre-specifying for dose-finding studies in this compound class, since it shifts allocation toward doses showing the clearest separation from placebo as data accumulates, reducing exposure to a dose later shown subtherapeutic. This was not part of the STEP or SURMOUNT designs (both used fixed-dose parallel arms), but it has precedent in oncology and is increasingly discussed for early metabolic peptide development given wider dose-response uncertainty at this stage.
Interim analyses should also include a pre-specified endpoint-switching rule: if blinded aggregate data suggest the primary endpoint's assumed timeline does not match the observed pooled response curve, a protocol amendment — reviewed by the same data monitoring committee, with the sponsor and FDA notified per 21 CFR 312.30 — should be pre-agreed rather than improvised mid-trial, preventing a late-stage scramble that raises regulatory scrutiny about data integrity.
Adverse Event Monitoring as a Formal Outcome Domain
Safety outcome measures deserve the same pre-specification rigor as efficacy endpoints, not treatment as an afterthought MedDRA-coded table. In STEP 1, gastrointestinal adverse events (nausea, diarrhea, vomiting, constipation) occurred in 74.2% of semaglutide-treated participants versus 47.9% on placebo, and 7% of the semaglutide arm discontinued due to adverse events versus 3.1% on placebo — a difference large enough that GI tolerability functions as a de facto efficacy-limiting endpoint, since it directly affects real-world adherence outside the trial's controlled dosing schedule.
For a compound engaging additional receptor targets, the monitoring plan should be built around the organ systems those receptors are known to affect in mechanistic or animal-model literature, not the standard GLP-1R monoagonist checklist. Glucagon receptor engagement, for example, carries a theoretical signal for heart rate elevation that GLP-1R monoagonists do not consistently show, so frequent-interval heart rate monitoring becomes a relevant outcome measure rather than a routine check box. Laboratory intervals should likewise be pre-specified by mechanism: lipase and amylase for the pancreatitis signal tracked across the GLP-1R agonist class, calcitonin monitoring where thyroid C-cell proliferation has been a rodent-model concern, and liver enzyme panels timed to the compound's metabolism pathway. Each is itself an outcome measure with a pre-specified normal range, an escalation threshold, and a documented action plan if breached.
Regulatory Considerations for Endpoint Justification
FDA's guidance documents function less as a checklist and more as a starting negotiation position for a pre-IND or end-of-Phase-2 meeting. For a genuinely novel multi-receptor mechanism, sponsors should expect FDA to request additional justification for any endpoint that departs from the standard weight-loss primary/responder-rate co-primary structure described in the 2007 obesity guidance.
Historical precedent is instructive: dual and triple receptor agonists (tirzepatide, retatrutide) used largely the same endpoint architecture as GLP-1R monoagonists because their dominant clinical effect — weight loss — matched the assumption embedded in existing guidance, even though receptor pharmacology differed. A GLP-3-class compound whose primary signal diverges meaningfully from weight loss (glycemic stability as the dominant early signal, for example) would likely need a formal endpoint justification package, referencing mechanistic and Phase 1b PK/PD data, submitted at a Type B meeting before Phase 2 initiation. Research-grade compounds distributed outside an IND-approved clinical protocol carry no trial-design flexibility at all, since research-use-only labeling precludes human dosing and any associated clinical outcome measurement.
A Practical Framework for Outcome Measure Selection
Distilled into a working sequence, outcome measure selection for a GLP-3-class or comparable emerging peptide trial should follow five steps rather than defaulting to precedent. First, characterize the compound's receptor engagement profile and PK/PD curve from Phase 1 data before drafting any endpoint — terminal half-life, time to peak effect, dose-linearity. Second, map that pharmacology against the four established secondary domains (cardiometabolic, body composition, patient-reported outcome, safety-tolerability) to identify which is likely to show the earliest, cleanest signal, and consider that domain for co-primary status rather than defaulting to weight loss alone.
Third, pre-specify subgroup analyses tied to plausible mechanism-based heterogeneity with adequate power rather than treating findings as exploratory. Fourth, build interim analysis and endpoint-switching rules into the protocol before enrollment starts, using a group sequential design with a pre-specified alpha-spending function, so a mismatch between assumed and observed response curves triggers a formal amendment rather than an ad hoc scramble. Fifth, treat adverse event monitoring as mechanism-specific outcome measures with pre-set thresholds and escalation plans, not a generic safety table appended at the end. None of these steps requires abandoning the STEP and SURMOUNT precedent — they require using it as a reference point rather than a template to copy. The programs that run into the most friction with FDA are consistently the ones that imported an endpoint structure wholesale from the most recent successful trial in an adjacent drug class, without confirming the new compound's pharmacology matched the assumptions embedded in that structure.
Where the Field Stands and What to Do Next
The evidence base for GLP-3-class multi-receptor peptides remains almost entirely preclinical and early Phase 1, with the mature endpoint architecture built for GLP-1R monoagonists and dual/triple agonists (semaglutide, tirzepatide, retatrutide) serving as the best available reference rather than a validated template for this newer category. Sponsors and investigators evaluating trial designs in this space should treat every borrowed endpoint as a hypothesis to be tested against Phase 1b data, not an assumption to be inherited.
The concrete next step for any team drafting a protocol in this category: before finalizing the statistical analysis plan, cross-reference the proposed primary and secondary endpoints against the compound's own Phase 1 dose-response curve, terminal half-life, and receptor-binding data — not against the most recent competitor trial design — and document that justification explicitly in the protocol's endpoint rationale section for the eventual FDA meeting record.
This article summarizes research and does not constitute medical advice. Consult a licensed clinician for diagnosis, treatment, or any decisions about medications or supplements.