Every letter we have printed

A randomised withdrawal design answers a narrower question than most readers take from it. It tells you what happens to people who reached a defined point and were then randomised, which is not the same as what happens to somebody who stops on their own terms.

The reasons literature would improve enormously if a single study published the free-text answers alongside the coded ones. The coding is where the information is lost, and the raw text usually still exists.

I lost access for eleven weeks during the shortage, restarted at the dose I had been on because nobody told me otherwise, and spent a fortnight unable to keep food down. I had been on that dose for seven months without difficulty. Reading your resumption section was the first time anybody explained it.

The number worth asking for is resolving power at the mass you care about, not the headline figure from the brochure. Instruments are specified at a mass convenient to the manufacturer, and a peptide at four thousand daltons is not that mass. Ask where the specification was measured.

A note on calibration intervals. Mass accuracy drifts with laboratory temperature, and a determination run eight hours after the last calibration is a different measurement from one run eight minutes after it. The interval is recorded in the instrument log and never in the report.

A caution on the word high-resolution as it appears in marketing. It is not a threshold; it is a continuum, and instruments separated by a factor of twenty in resolving power are sold under the same adjective. Ask for the figure and the mass at which it was obtained, and the adjective becomes unnecessary.

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.

Biological variation within one person is smaller than the variation between people for many analytes, which is why a series of results on one individual is far more informative than a single result against a population interval.

Where the seller ships directly to the laboratory, custody is clean and the sample is chosen. Where the buyer ships, custody is broken and the sample is representative. The trade is not acknowledged anywhere in the literature I have read.

An observation about scale. Custody discipline is easy at one sample a month and hard at fifty, which is why the programmes with the most data have the weakest records. The failure is a consequence of volume rather than of care.

Where a result is flagged high or low, the flag is generated by software against the interval and carries no clinical judgement whatsoever. Readers treat the flag as an opinion and it is a comparison.

Biological variation is the concept I would add. Each analyte has a characteristic within-person variability, and for some of them a change has to be substantial before it means anything at all. Without that figure a delta is uninterpretable, and it is available in published tables.

My ferritin fell from 118 to 34 across a year on treatment and I was told this was expected because inflammation had fallen. Six months later I was clearly iron deficient. I appreciate the section on this being ambiguous, but the ambiguity was resolved in one direction and nobody looked.

A sterile field is a concept and not a product. The bench, the hands, the air movement and the sequence all belong to it, and a person working carefully on a clean surface is doing more than one working carelessly with expensive supplies.

Surface disinfection requires a contact time that is printed on the product and almost never observed. Wiping and immediately proceeding achieves very little, and the wait is the active ingredient.

Your needle-length section says four millimetres is adequate for all adults, which contradicts what I was told by a nurse who insisted on half an inch because of my weight. Which is right?

Measurement frequency determines how a plateau appears. Weighed weekly, a slow change is visible; weighed monthly, it looks flat. The observation interval is part of the finding and is never quoted with it.

Nobody has published what happens to intake over the course of these trials, though several must have collected it. A plateau explained by intake returning toward baseline is a different phenomenon from one explained by anything pharmacological.

Your residual-exposure table gives six per cent after four weeks. I calculate 6.25 per cent, which is a quibble, but the larger point is that it assumes steady state at the moment of interruption. Someone who stops three weeks into a new rung has less on board than your table implies.

Thank you for saying plainly that the maximum dose is not the goal. I stopped at 10 mg fourteen months ago because it was working and I was tired of arguing about it. Every article I read before yours implied I had given up early.

The comparison with food-hygiene ratings is instructive. Those are dated, they are issued by a body with no commercial relationship to the premises, and they can go down. Three properties, none of which this market’s marks possess, and each of which could be adopted independently.

Nobody expires anything. A badge granted three years ago is still on the page, still linking to a report about a lot that ran out long ago, and there is no mechanism anywhere in the chain that removes it. Decay is the missing feature.

You describe the services as competitors. From where I sit they are also each other’s only check, because a buyer who disagrees with one result has nowhere to go but another of the four. That is a peculiar market structure and it is worth naming as one.

Speaking as a former quality manager, the safeguard that matters is whether the analyst knows whose sample it is before the run. Blinding at the bench costs nothing and removes the whole class of problem your article describes. It is a question worth putting to each of the four.

The correction printed against an earlier piece in this series was handled better than most publications manage. Owning an error in the same place as the claim is the whole of the practice.

Nobody measures what they have after storage, which is the only way any of this would be settled. A determination on a solution held for four weeks would produce more information than every discussion thread on the subject combined.

Adsorption to the vial wall is the loss mechanism nobody thinks about, and it matters most at exactly the low concentrations people prepare. Material can leave a solution without degrading at all.

A general point about rate. Everything here is kinetics, not thresholds, and the popular framing of a material being fine until it suddenly is not has no basis in the chemistry. Loss is continuous and the question is how much.

Seasonal variation makes a single documented shipment close to meaningless. A supplier’s packing that performs well in March may not in July, and any claim about cold-chain performance should be dated.

Rapid weight change mobilises material stored in adipose tissue, and analytes that partition there behave accordingly. Reading such a movement as new pathology rather than as redistribution is an easy error and your piece describes it well.

Tube type and fill volume matter. An underfilled tube changes the ratio of sample to additive, and the result is affected in a direction that depends on the analyte. The laboratory notices; the person receiving the report usually does not.

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.

Correction to your HbA1c section. You write that chronic kidney disease lowers HbA1c through shortened erythrocyte survival. It can also raise measured values on some assay platforms through carbamylated haemoglobin interference. The net direction depends on the method.

Aliquoting breaks the chain silently. If a buyer decants a portion into a second tube before sending it, the laboratory receives a sample the supplier never touched, and any dispute about the result is now unresolvable in both directions.

Time is the variable custody records exist to capture, and almost every account I read gives dates rather than durations. Eleven days between purchase and injection is a fact about the result; the two dates it is derived from are usually printed and never subtracted.

An interlaboratory comparison needs a homogeneous material and a stated assigned value, which is why proficiency schemes in other sectors are run by a body that prepares the material rather than by a participant. Until somebody does that here, these exercises measure agreement rather than accuracy.

Weight loss of any origin is associated with density change, so the comparator matters enormously. Without a matched non-pharmacological arm, an observed change tells you about losing weight rather than about the agent.

Precision error on a densitometer is a published figure for each machine and it is comparable in size to the changes being reported. Any study reporting a change smaller than its own least significant change is reporting noise, and some do.

Slower escalation as an alternative to stopping is the option that most often gets skipped, and the trial protocols permitted it. Presenting the choice as continue or stop leaves out the middle that the evidence base itself used.

An effect is dose-limiting when it stops the escalation, and that is a decision rather than a symptom threshold. Your framing makes that visible, which matters because the same symptom is tolerable in one person’s circumstances and not in another’s.