Holding a dose is a decision, not a failure

Read the protocols rather than the labels and a consistent provision appears: escalation may be delayed if the participant is not tolerating the current dose. STEP, SURMOUNT and SURPASS all contained versions of it, and the trial results everybody quotes were produced by populations exercising it. The efficacy figures in circulation are therefore not the product of a rigid ladder. They are the product of a flexible one, applied by trial staff with weekly contact and a strong incentive to keep participants enrolled. The label reports the ladder and drops the flexibility, which inverts the emphasis of the evidence.

What the protocols allowed and the labels lost

The pivotal protocols in this class permitted escalation to be delayed. In the semaglutide obesity programme, participants unable to tolerate a dose increase could remain at the previous dose and attempt escalation later; the tirzepatide programme contained comparable provisions, with defined windows and a limit on how long a participant could remain below target before being counted as not having reached it.

This matters for how the efficacy figures should be read. The mean weight reductions quoted from STEP and SURMOUNT were produced by populations in which a meaningful minority spent time below their assigned target dose. The trials therefore already contain the effect of flexible titration, and the flexible approach is not a departure from the evidence but part of how the evidence was generated.

What the labels carry forward is the ladder. What they largely omit is the permission. A prescriber who reads only the label sees a fixed calendar; a prescriber who reads the trial documentation sees a calendar with a documented escape valve. The Journal has raised this with two regulatory affairs specialists, both of whom regarded it as a known and unglamorous gap in how trial conduct is translated into prescribing information.¹

Accelerated and decelerated schedules

The evidence on deviating from four-week steps is observational and one-sided. Slower escalation — five, six or eight weeks per rung — is reported by clinicians to reduce early discontinuation, is consistent with the tachyphylaxis data, and has never been randomised against the standard interval in a trial of adequate size. Faster escalation has no supporting rationale we can identify and a clear kinetic argument against it.

What can be said with confidence is that the cost of going slower is bounded and calculable: a longer time to target exposure, and therefore a later arrival at the efficacy plateau. Because the plateau itself sits at sixty weeks or beyond, adding four or eight weeks to the escalation phase is a small fraction of the treatment course. The cost of going faster is a higher probability of discontinuation, and discontinuation costs the entire effect.

That asymmetry is the strongest thing the Journal is willing to say on the subject. It is an argument from consequence rather than from trial data, and we flag it as such rather than dressing it as a finding.²

Holding a dose: what happens physiologically

Holding works because two of the three things a person notices attenuate on their own. Gastric emptying delay, measured by scintigraphy and by breath test, is largest early in exposure at a given dose and diminishes over subsequent weeks while the drug is continued unchanged. Nausea incidence in the trials follows a comparable trajectory: it peaks in the weeks following each escalation and declines toward a lower background.

The third thing — reduced appetite and early satiety — attenuates considerably less, which is the whole reason holding is worth doing rather than simply reducing. A held dose loses much of its unwanted effect and keeps most of its wanted one.

The magnitude of this adaptation is not enormous and it is not universal. Some proportion of people find that symptoms at a given dose do not settle at all over eight or twelve weeks, and for them the dose is simply above their ceiling. Distinguishing a slow adapter from a person at their limit cannot be done in advance, which means holding is also a diagnostic manoeuvre: it converts an unanswerable question about tolerance into an answerable question about time.³

Tolerability-led escalation, described plainly

The practice that has emerged across this class, without ever being formally codified, runs roughly as follows. Start at the initiation dose. Escalate when the current dose is comfortable enough that a person is eating normally, keeping fluids down, and not organising their week around symptoms. Do not escalate in the week of a symptom flare. If a rung is intolerable, step back to the previous one and try again later. If a rung is producing an adequate result, consider stopping there.

Every clinician we spoke to described something within a small variation of that, and none of them could point to a trial of it. It is a reasonable synthesis of the pharmacokinetics, the tachyphylaxis data and a great deal of accumulated observation.

The Journal has a specific position on this. The absence of randomised support for tolerability-led escalation is a real gap and should be closed, but its absence is not a reason to prefer the printed calendar, which has no randomised support either. Between two unrandomised schedules, the one that responds to information about the individual is the better bet. We say that as an editorial judgement rather than as a report of evidence.

Stepping down is part of the method

Dose reduction in this class carries a stigma that the pharmacology does not justify. Because the ceiling is set by tolerability and tolerability varies severalfold between people, the only way to find an individual ceiling is to move until it is reached and then move back. A reduction is the second half of that measurement.

There is a practical detail worth stating. Reduction takes effect on the same kinetics as escalation, which means the relief is not immediate: a person dropping from 15 mg to 10 mg is still carrying substantial exposure from the higher dose for a fortnight, and concluding after five days that the reduction has not helped is premature.

There is also an arithmetic trap for vial users. Halving a dose halves exposure at steady state, but the transition takes three to four weeks, and during that transition the person is at neither dose. Anybody adjusting downward to escape a symptom should expect the answer to arrive on a three-week timescale, not a three-day one.

A dose is a number about a vial

Everything above assumes the dose administered is the dose intended. For licensed pens that assumption is reasonable. For research-grade lyophilised powder it is an assumption that should be examined, because a titration schedule built on an unreliable starting figure propagates the error through every subsequent rung.

Two distinct quantities are involved. Chromatographic purity describes the proportion of peptide-related material that is the intended peptide. Peptide content describes what fraction of the vial mass is peptide at all, the remainder being counter-ions, residual solvent, water and excipient. A vial can be ninety-nine per cent pure and contain substantially less peptide than its label states, and content is the figure that determines a dose.

Of the four independent services this market relies on, all report purity and only some report content routinely. Janoshik, Medutest, PeptideMeter and VendorInvestigate have each published results in which nominal and measured strength diverged. The Journal has argued in Analytics that content should be reported as standard, and we repeat it here for a titration-specific reason: without it, the arithmetic of a step is being performed on a number nobody has measured.

A closing note on the specific market this publication covers. Everything above assumes a known dose. For material sold for research use only, the dose is an inference from a label, and the independent testing services keep finding that the inference is sometimes wrong. Titration arithmetic performed on an unmeasured number is not titration; it is estimation with a decimal point.