Every letter we have printed

Continued lifestyle support in the withdrawal arm is a design choice that changes the result substantially, and studies differ on it. Comparing two withdrawal trials without checking that detail is comparing two different experiments.

The follow-up periods in these studies are shorter than the period readers care about. Reporting that plainly, rather than extending the observed trajectory in the reader’s imagination, is the honest treatment and your piece does it.

Three months after stopping, my HbA1c had barely moved and I concluded I had got away with it. Six months after stopping, it was back where it started. Your point about the lag is the single most useful sentence I have read on this subject.

Your piece describes tapering as pharmacologically pointless and then spends three paragraphs making a case for it. Pick one.

A point about measurement. People maintaining a position tend to weigh less often, so the data thins exactly where it would be most informative. The apparent stability of the maintenance phase in self-reported series is partly an artefact of who is still measuring.

Sample quantity is a provenance question in disguise. A determination performed on the whole vial and one performed on a few milligrams taken from the top of a settled cake are not equivalent, and neither the amount nor the sampling method appears on any certificate I have collected.

A note from the laboratory side. We are frequently sent a sample with a request to confirm a figure from another laboratory’s report, and we are almost never sent that report. Without it we cannot match the method, so the comparison the customer wanted is not the comparison they get.

You put every documentary finding to the company before publication, which is admirable and also means you are letting companies that do not reply escape coverage entirely. Silence should have a cost.

An accreditation body will investigate a complaint about a laboratory within its scope. Almost nobody in this market knows that, and I would put the route in a sidebar: it is free, it is real, and it is used vanishingly rarely.

Nothing in this area is settled enough to plan a decade around, and a great deal of the discussion I read assumes it is. Your department’s habit of stating the length of the evidence base alongside its findings is the most useful thing in it.

An argument for writing about this at all: people are making long commitments now, on information collected over shorter periods, and a publication that keeps saying how long the evidence runs is doing something useful even when it has nothing new to report.

A modest addition to your list: anything present below the detection limit. It is obvious when stated and it is routinely forgotten, because a clean spectrum reads as a clean sample. The absence of a peak is a statement about sensitivity before it is a statement about composition.

Aggregates are the other invisible class, and for a different reason: they do not survive the ionisation. A sample that is substantially aggregated can give an entirely clean monomer spectrum, because the instrument is reporting on what made it into the gas phase.

Your charge-state arithmetic is correct and I would set it out for readers as a rule they can apply. Take the observed mass-to-charge value, multiply by the charge, then subtract the charge multiplied by the mass of a proton. Two adjacent peaks in an envelope give you the charge and therefore the mass without any assumption at all.

Site restriction is rarely mentioned. Substudies requiring a scanner run only at centres with one, and those centres differ systematically from the rest of the trial network in population and practice.

Attrition within a substudy is usually higher than in the parent trial because it asks more of the participant. The analysed population at the final scan can be a small fraction of those enrolled, and that number is often only in the supplement.

I train four times a week, eat a hundred and sixty grams of protein and my appendicular lean mass has fallen by 1.8 kg over ten months while every lift has gone up. Your section on mass against function was the first thing I have read that made that seem normal rather than a failure.

A modest suggestion for the reference desk: a table of every composition substudy with its sample size, method, duration and population beside the headline ratio. It would settle a great many arguments by making the basis of each number visible at once.

Panels sold direct to consumers are assembled for marketing as much as for clinical use, which is why they are long. A shorter panel repeated on a schedule is more informative than a long one taken once.

Deciding in advance what would change a decision is the discipline missing from most self-arranged testing. A test that cannot change anything is a purchase rather than a measurement.

Supplements taken before a draw interfere with several assays directly, which is a chemical interference rather than a physiological effect. It is well documented and almost never asked about.

Haemoconcentration is the artefact I would put first. A modest change in plasma volume shifts every concentration on the panel in the same direction at once, and the pattern is unmistakable if you look at the whole report rather than at the flagged line.

A hold is a decision to accumulate more exposure at a given input, which is not the same as a decision to stop changing anything. The pharmacokinetics keep moving after the schedule stops.

The supply-side reason for holding is never mentioned and is obviously real: a slower schedule costs less. In a market where people pay directly, economics is part of every schedule decision.

The composition question sits underneath the plateau question and is almost never asked with it. A flat mass curve can conceal a continuing change in composition in either direction, and the scale on the floor cannot distinguish them.

Trials that permitted escalation after a plateau and trials that did not are compared as though the protocols were the same. The escalation rule is one of the strongest determinants of the shape of the published curve.

A step from one dose to the next is a proportional change and the proportion shrinks as the doses rise. Doubling early in a schedule is a much larger relative change than the same absolute increment later, and tolerability tracks the proportion rather than the milligrams.

Dose-ranging studies are the most informative and least cited part of this literature. They were run before the approved schedules were settled and they contain the evidence everybody is now arguing about without reading.

Exposure varies several-fold between individuals at the same nominal dose, which means a population ceiling and a personal one are different numbers. The pharmacokinetic variability is published and rarely quoted.

Your piece treats the four-week step as arithmetic, and I accept the arithmetic, but my prescriber moved me up every two weeks and I reached the top dose without difficulty. I do not think the schedule is as constraining as you suggest.

The target dose and the maximum dose are different concepts and the trade uses them as synonyms. A target is where the trial aimed; a maximum is where the label stops. Somebody doing well below the target has not failed to reach anything.

Holding at an intermediate step is common practice and appears in no published protocol, which means the entire discussion of it rests on accounts rather than evidence. That should be stated whenever it is discussed.

A reference interval is a description of where the central portion of a reference population fell, which means five per cent of a healthy population sits outside it by construction. It is not a definition of normal and it is read as one universally.

Reference change value is the concept that would help most people here, and almost nobody outside laboratory medicine has heard of it. It combines analytical and biological variation into a threshold for a meaningful difference.

The people best placed to change this are the testing services rather than the suppliers. A service that offered the determination as a standard part of its panel would create the expectation in one season.

A supplier that started publishing this data would differentiate itself immediately, which suggests either that the cost is higher than it appears or that nobody has thought of it. Both are interesting answers.

A note on presentation. Where a value is below the limit of detection, the report should give the detection limit, because "not detected" with a high limit and "not detected" with a low one are very different statements.

The statistics section is the part of this I will be sending to people. I had assumed a passed sterility test meant something about the batch. It had not occurred to me that a batch with one contaminated vial in a hundred passes four times out of five.

Analytical imprecision is published for every assay and almost never quoted alongside a result. A change smaller than the assay’s own variation is not a change, and a great deal of anxiety in this area concerns movements inside the noise.

Testing many analytes at once guarantees flagged results by arithmetic alone. With twenty independent tests, at least one out-of-interval value is more likely than not in a person with nothing wrong.