GLP-1 Microdosing Finally Has Real-World Data: 2.2% Weight Loss in a Year

Microdosing GLP-1s is one of the most confidently discussed practices in metabolic health and one of the least evidenced. The pitch is appealing: stay at 0.25 mg of semaglutide indefinitely, keep the metabolic benefits, skip the nausea, spend a quarter of the money. Entire forums are built on it. Until now there was no clinical data on it at all.

There is some now. The study: Murugadoss, Venkatakrishnan & Soundararajan, Comparison of Cardiometabolic Benefits and Tolerability in Patients on Sustained Low Doses of Semaglutide or Tirzepatide, Biology Methods & Protocols, 2026 — accepted manuscript.

A team at nference searched a federated system of 29 million de-identified patient records for people whose prescriptions show they never escalated past the starting dose, and compared them head to head.

It is the first attempt at this question, and the answer is more deflating than either camp expected.

Almost nobody actually does it

The first finding is in the recruitment numbers, and it is easy to skip past.

Of 490,072 semaglutide-treated patients, the number who met a sustained-lowest-dose definition — 0.25 mg only, at least three separate prescriptions, spanning at least six months, with no other GLP-1 in the window — was 814. That is 0.17%, about one in 600.

For tirzepatide it was 1,016 of 322,442, or 0.31%. After matching, the head-to-head compared 534 patients per arm.

Whatever the internet suggests, durable non-escalation is rare in recorded clinical practice. It is worth sitting with that before reading anything else in the study.

The weight loss number

This is what people actually want, so here it is plainly.

For an average 100 kg starting weight, a year of microdosed semaglutide is a little over two kilograms — and the 15-month figure is smaller than the 12-month one, meaning the curve has flattened and started to drift back.

The authors also compared these patients against matched people who escalated normally. High-dose semaglutide at the same timepoints: −5.08%, −8.85%, −11.45%, −12.03%, −12.44%.

So the honest comparison at 12 months is 2.2% against 12%. Microdosing is not a way to get most of the effect for less. On these numbers it delivers roughly one sixth of it.

"Low dose" means two completely different things

The second table column deserves its own paragraph, because it quietly undermines how people talk about this.

Tirzepatide's starting dose produced −5.5% at a year — more than double what semaglutide's starting dose managed, and comparable to what a full course of some older weight-loss drugs achieved. Both are labelled initiation doses. They are not equivalent interventions.

If someone tells you they are microdosing, the drug matters more than the word.

The part almost everyone will get wrong

Here is where this study is going to be misreported, and it is worth being precise.

The abstract lists a series of adverse events with p-values under 0.05, higher on low-dose tirzepatide: acute kidney injury 2.68% vs 0.43%, constipation 32.58% vs 22.41%, muscle cramps 8.34% vs 4.00%, lumbar disc disease 3.30% vs 0.29%, actinic keratosis, lentigo, dyspnea on exertion. Semaglutide came out higher on otitis, diaphoresis, ankle swelling and migraine.

A six-fold difference in kidney injury is the kind of number that travels. But the researchers tested 461 separate conditions. When you run that many comparisons, some will clear p<0.05 by chance alone — and the authors applied the standard correction for exactly this. Their own words:

After Benjamini-Hochberg correction for multiple comparisons across the 461 adverse-event conditions, none of these signals remained significant (all q>0.5), so the individual adverse-event associations should be interpreted as hypothesis-generating rather than confirmatory.

None of them survived. Not the kidney injury, not the constipation, none.

Separately, the structured analysis of diagnosis codes tested 104 outcomes across six disease-burden groups. Not one reached statistical significance either.

So the accurate summary of the safety comparison is: no difference was demonstrated between these two groups on anything. Any headline pulling the 2.68%-versus-0.43% figure out of the abstract and calling it a finding is reporting noise that the authors themselves discarded.

The one signal with a second witness

One thread is more interesting than the rest, not because it is stronger statistically but because it was corroborated by a different kind of data.

Prescriptions for renal-support and hyperkalemia therapies in the tirzepatide arm rose from 5.63% to 9.51% by 12 months (p=0.016). In the semaglutide arm they moved from 8.20% to 6.64% and did not reach significance (p=0.382).

That is a prescribing pattern changing in the same direction as the kidney signal from the clinical notes. Two independent data sources pointing the same way is worth more than one p-value. It is still not proof — but it is the finding most deserving of a proper prospective study.

The kidney question, properly

This is the thread most worth pulling, so it deserves more than the abstract's one line. The study produced four separate renal readouts, and they do not agree.

1. Coded diagnoses: nothing. The structured analysis included a dedicated renal pre-existing-burden group. Across all 104 outcomes in six burden groups, no comparison reached significance. Incident coded kidney disease did not differ between the arms.

2. Clinical notes: a signal that fails its own test. Acute kidney injury appeared at 2.68% versus 0.43% over 24 months (p=0.007) — but with q=0.6 after correction across the 461 conditions. The authors do not treat it as a finding.

3. Prescriptions: a directional change. Renal-support and hyperkalemia therapies rose from 5.63% to 9.51% in the tirzepatide arm by 12 months (p=0.016), while semaglutide's moved 8.20% to 6.64% and did not reach significance. Note carefully what this test is: a within-arm before-and-after comparison, not a head-to-head. It shows prescribing changed inside one group. It does not establish that one drug was worse than the other.

4. Dose-response: absent. In the separate comparison of low-dose against high-dose semaglutide, acute kidney injury came out at log-rank p=0.072 — not significant. Hyaline casts, p=0.172. More semaglutide did not produce more coded kidney injury, which is what you would expect to see if the drug were driving renal harm.

Two pieces of context matter before anyone draws a conclusion.

These cohorts were not renally healthy to begin with. At baseline, 7.7% of the semaglutide arm and 9.7% of the tirzepatide arm already carried a chronic kidney disease diagnosis, and 4.1% and 3.7% respectively already had a history of acute kidney injury. In a population like that, incident renal events accumulate for reasons that have nothing to do with which incretin was prescribed.

The randomised evidence points the other way. In FLOW — 3,533 patients with type 2 diabetes and chronic kidney disease — semaglutide at 1.0 mg reduced major kidney disease events by 24% (HR 0.76). That is a hard renal endpoint in a randomised controlled trial, and it is in this paper's own reference list. A hypothesis-generating signal at a sub-therapeutic dose does not outweigh it.

There is also a mechanistic wrinkle. GLP-1-associated kidney injury is classically pre-renal — dehydration driven by vomiting and diarrhoea. The dominant gastrointestinal signal in this cohort was constipation, which is a much less direct route to volume depletion.

The honest summary: this study contains no long-term kidney data at all. Maximum follow-up is 24 months. There are no eGFR trajectories, no creatinine or albuminuria measurements, no CKD-progression endpoints — only diagnosis codes and language-model extractions from notes. What long-term low-dose incretin exposure does to kidneys is, at this point, simply unmeasured.

What the notes actually say

The team also ran a language model over clinical notes to find explicit mentions of microdosing. Of 451 patients whose notes mentioned it, chart review confirmed genuine microdosing in 41 — just 9.1%.

Among those 41: 22 (53.7%) had documented weight loss or weight control benefit, 16 (39.0%) had no clear documented benefit, and 10 (24.4%) had documented adverse effects or poor tolerability.

Roughly half saw something. That is the most direct evidence available about people deliberately doing this, and it is 41 people.

What this study cannot tell you

The authors are unusually candid, and the limitations matter more than usual here.

Prescription records are not doses taken. The cohort is defined by what was prescribed, not what was injected. The median patient had three prescriptions across about 371 days — a sparse fill pattern. As the authors write, they "cannot distinguish deliberate low-intensity or supply-extending use from incomplete adherence." Some of these people were microdosing on purpose. Others were struggling to fill scripts, or stopped and restarted. The data cannot separate them.

The pen problem. Semaglutide comes in multi-dose pens — a starter pen titrating from 0.25 mg to 0.5 mg. The recorded dose does not always reflect the delivered dose.

Compounded and telehealth sources are invisible, which is precisely where a lot of real microdosing happens.

The cohorts are small — 534 per arm — which limits power to detect anything modest.

And this is an accepted manuscript, not yet the final typeset version. The authors are employees of nference, the company whose federated data platform the study runs on.

The bottom line

Microdosing GLP-1s is not a hack that preserves the benefit and drops the cost. In the only real-world data that exists, a year of microdosed semaglutide produced about 2% weight loss where standard dosing produced about 12%, and the effect had already begun to fade by month 15.

Tirzepatide's starting dose is a genuinely different proposition from semaglutide's, at −5.5%, and the two should stop being discussed as one practice.

On safety, nothing was demonstrated in either direction. The striking numbers in the abstract are the ones the authors' own correction threw out — which is a useful reminder that the most quotable figure in a paper is often the one the researchers trust least.

This is general information, not medical advice. Dose changes belong with your prescriber.

Sources

• Murugadoss K, Venkatakrishnan AJ, Soundararajan V. Comparison of Cardiometabolic Benefits and Tolerability in Patients on Sustained Low Doses of Semaglutide or Tirzepatide. Biology Methods & Protocols. 2026 (accepted manuscript). doi:10.1093/biomethods/bpag048
• Perkovic V, et al. Effects of Semaglutide on Chronic Kidney Disease in Patients with Type 2 Diabetes (FLOW). New England Journal of Medicine. 2024;391:109-121. doi:10.1056/NEJMoa2403347