Why Ozempic doesn't work for one in ten — and now we can predict who

TL;DR

• A Stanford-led analysis published 12 April 2026 across three GLP-1 clinical trials (n = 1,119) identified specific PAM gene variants that predict reduced response to semaglutide and tirzepatide.
• Carriers showed lower HbA1c reductions and lower body-weight loss compared with non-carriers — effect sizes large enough to be clinically meaningful, not just statistically detectable.
• This is the first replicated, multi-trial, genotype-mappable signal of GLP-1 non-response. It turns "the 10% who don't respond" from a clinical impression into a diagnostic phenotype.
• The decision-useful implication: payer prior-authorisation logic, prescriber triage protocols, and patient counselling all become more precise. Genetic screening before initiation is now scientifically defensible — and within twelve months, will be commercially viable.
• This is also a wedge into a larger truth about GLP-1s: they are not equally effective drugs across populations. We have been treating a heterogeneous responder distribution as if it were a single category.

What the study actually shows

The analysis pulled data from three independent semaglutide and tirzepatide clinical trials, totalling 1,119 participants with available genotype data. The investigators looked for genetic variants associated with differential drug response across HbA1c reduction (the diabetes endpoint) and body-weight reduction (the obesity endpoint).

The signal localised to variants in PAM (peptidylglycine alpha-amidating monooxygenase), a gene encoding an enzyme involved in the post-translational processing of multiple peptide hormones. Carriers of specific PAM variants showed:

• Lower mean HbA1c reduction at 26 weeks
• Lower mean body-weight reduction at 26 weeks
• A consistent direction of effect across all three trials

The effect sizes were not subtle. Variant carriers' weight-loss outcomes were materially lower than non-carriers — large enough that, applied at population scale, the variant explains a meaningful share of the clinical observation that some patients just do not respond. Across the trial population, the variant frequency was consistent with allele frequencies that translate to roughly 5–10% of typical Western patient populations carrying the relevant genotype.

The mechanism is plausible. PAM processes a class of peptide hormones that includes, among others, components of the gut-brain axis modulated by GLP-1 signalling. A variant that alters peptide processing efficiency could plausibly reduce the downstream effect of pharmacological GLP-1 receptor activation. The study identifies the association; it does not yet prove the causal mechanism.

What it actually means

Three things change.

The clinical narrative changes. "Non-responders" stops being an unexplained category and becomes a phenotype. Clinicians who currently counsel patients with some patients respond differently and we cannot predict who now have, for the first time, a defensible answer to can we predict? The honest version of that answer is partially, yes — for one specific genetic factor that explains a meaningful share of variance.

The prior-authorisation conversation changes. Payers cover GLP-1s under prior authorisation that includes BMI thresholds, comorbidity criteria, and, increasingly, weight-loss-at-3-months continuation criteria. Genetic non-response data introduces a new variable: should genotyping be a covered pre-treatment screen? In some commercial plans, this conversation is already happening. The cost-effectiveness math is favourable for the payer at scale, less obviously so for the patient.

The drug-development conversation changes. Sponsors developing GLP-1 successors (oral GLP-1s, GLP-1/GIP combinations, GLP-1/glucagon triple agonists) now have a new positioning opportunity: our compound retains efficacy in PAM-variant carriers. If true, that becomes a population-stratified marketing claim that targets a real and reasonably large patient sub-population.

The broader scientific implication is more important still. We have been treating GLP-1 efficacy as a population-wide property when it is in fact a distribution. Other genetic and physiological factors will turn out to predict response variance too. The 12 April paper is the first; it will not be the last.

Hype deconstruction

It is not a "single gene determines whether Ozempic works." PAM variants explain a meaningful but partial share of response variance. Other genetic, physiological, and behavioural factors contribute. The accurate framing is one identified factor among several, with effect size large enough to be clinically actionable.

It does not mean carriers should not try GLP-1s. Reduced response is not zero response. Variant carriers in the trials still showed meaningful weight loss on average — just less than non-carriers. The clinical decision is one of expectation-setting and triage, not exclusion.

It is not yet a clinical genetic test you can order tomorrow. The variants are research-identified. Translating that into a CLIA-validated, clinically reportable assay takes 6–12 months for a commercial laboratory. The first such tests will probably emerge by the end of 2026.

The "one in ten" framing is approximately right but jurisdictionally variable. Variant frequency varies by ancestry. The trials' demographic composition was predominantly European and East Asian; the published prevalence figures generalise less well to populations underrepresented in the trial cohorts. This is the standard genetic-research caveat and the field is improving at flagging it, but the headline number is a rough guide, not a universal truth.

Stakeholder landscape

Patients on GLP-1s. Particularly those who feel they are not responding as expected. The cohort with subjective "I am doing everything right and the drug is not working" experiences will be the most interested in genotyping access.

Endocrinologists, primary care, and weight-management specialists. The triage decision becomes more sophisticated. We have an explanation for why this might not be working for you is a meaningful conversation upgrade.

Commercial payers. Major plans (Aetna, UnitedHealthcare, Cigna, Anthem) will look at this data through the prior-authorisation lens. The actuarial question is whether the cost of universal pre-treatment genotyping is offset by the cost of failed treatment courses. The math favours genotyping at scale; whether plans codify that in 2026 is policy, not science.

Direct-to-consumer testing companies. 23andMe, Genomic Health, Ancestry, and the obesity-genetics-focused players (Phenomix Sciences, MyToolbox Genomics) all have a clear product line extension. Expect the first commercial PAM variant assays in the second half of 2026.

Pharmaceutical sponsors. The current GLP-1 leaders (Novo Nordisk, Eli Lilly) face the question of whether to incorporate PAM genotyping into Phase 4 evidence generation and clinical positioning. The successor sponsors (Pfizer, AstraZeneca, Boehringer Ingelheim) face the question of whether to position next-generation compounds explicitly against the variant population.

Health technology assessment bodies. NICE in the UK, IQWiG in Germany, the PBAC in Australia. Each will need a position on whether genetic stratification belongs in the GLP-1 reimbursement framework. Expect formal guidance in 2027.

Cross-layer implications

• Precision-obesity medicine. Until April 2026, "precision medicine" in obesity has been mostly aspirational. The PAM variant finding is the first concrete instantiation. Expect a wave of similar genotype-stratified analyses across other GLP-1 trials and adjacent metabolic drug classes.
• Diagnostic infrastructure. A new clinical-grade peptide-hormone-pharmacogenomics test category is being created. The companies that ship first will define the reimbursement framework.
• Pharmaceutical labelling. Pharmacogenomic labels for GLP-1s are not yet on the FDA's near-term horizon, but the data is now sufficient to begin the conversation. Expect post-marketing label discussion in the next 18–24 months.
• Health equity. The dominant trial populations were European and East Asian. Generalising the finding to African, South Asian, Middle Eastern, and Latin American populations requires additional studies. Without that work, PAM genotyping risks being a precision tool that works less well for under-studied populations — a familiar and avoidable failure mode in pharmacogenomics.
• Insurance underwriting. Pharmacogenomic data, like all genetic data, is regulated under GINA in the US. The legal framework forbids discrimination in employment and health insurance, but life and disability insurance carve-outs remain. Patients should understand the disclosure implications before genotyping.

What this means for you

Patients on or considering a GLP-1. The right move today is not yet. Wait for CLIA-validated commercial assays, which are likely 6–12 months out. If you are already on a GLP-1 and not responding, the conversation with your prescriber is the same as it was last month — adjust dose, allow time, consider switching agent or adding behavioural support — but the why layer has gained a new candidate explanation.

Clinicians. Begin updating your patient-counselling script. The accurate version is now: response to GLP-1s varies, and we are starting to be able to predict some of that variance from genetic factors. Tests are not yet clinically available, but if your response is below expected, this is one of several explanations we will be able to investigate within the next year.

Health system pharmacy and benefits leaders. Consider building a position on pharmacogenomic-screened GLP-1 prior authorisation now. The clinical evidence will be there; the operational question is whether your plan is ready to operationalise it. The decisions being made in 2026 will set the precedent for the rest of the decade.

Investors. Three trades. (1) Pharmacogenomics testing companies with CLIA-validated peptide-hormone-pharmacogenomics capability — the first commercial PAM assay is a clean win. (2) Next-generation GLP-1 sponsors who can credibly position against the variant population. (3) The diagnostic-payer-coding cycle: what gets covered and what does not in 2027 prior-authorisation decisions will move pharmacy-benefit-management economics.

Researchers. Replicate. Extend to underrepresented populations. The field needs a multi-ancestry replication of the PAM finding and a systematic search for other variants in adjacent peptide-processing genes. The methodological template is set; the work is now mechanical.

Uncertainty ledger

• Whether the PAM variant association replicates in independent, non-trial cohort data. Real-world EHR-based replication is the next milestone.
• The causal mechanism — does PAM variation alter pharmacokinetics, pharmacodynamics, or both?
• How variant frequency and effect size translate to non-European, non-East-Asian populations. Likely meaningful differences; absence of data is the issue.
• Whether genotyping will demonstrate cost-effectiveness in formal HTA assessment. Plausibly yes for high-cost GLP-1s; the math is sensitive to test cost and clinical decision-rule design.
• Whether other GLP-1-class compounds (oral semaglutide, orforglipron, retatrutide, the next-generation triple agonists) show the same variant-response association.

Bottom Line

The first replicated, clinically actionable genetic predictor of GLP-1 non-response is now in the literature, and within twelve months it will be in the diagnostics market. That is a real shift in how the most consequential drug class of the decade gets prescribed, paid for, and explained to patients. The clinicians, payers, and sponsors who treat PAM as the start of pharmacogenomic stratification in obesity medicine — not the end — will be the ones whose protocols, formularies, and pipelines hold up when the next variant is identified, and the one after that.

Written in the tradition of — P.

Sources

• Tier 1. Stanford-led multi-trial PAM variant analysis, published 12 April 2026
• Tier 1. US Food and Drug Administration — semaglutide and tirzepatide trial protocols and prescribing information
• Tier 2. Medical Xpress — peer commentary on findings
• Tier 2. News-Medical — clinical implications coverage
• Tier 1. National Institutes of Health — GINA pharmacogenomic data framework
• Tier 2. BioSpace — sponsor and diagnostic industry response