Which blood panels are actually worth ordering on a GLP-1, and at what interval?

I have been offered a "metabolic optimisation panel" by a private lab that runs to 58 analytes and costs about the same as a month of compounded semaglutide. My GP will run a much shorter list for free but wants to do it annually. Neither of those feels like the right answer.

What I want is a defensible minimum: the tests whose results could actually change a decision, at an interval matched to how fast the underlying thing moves. I can see the shape of the problem — HbA1c cannot meaningfully change in three weeks, so ordering it monthly is theatre, whereas creatinine can move in a fortnight — but I do not know the time constants well enough to build the schedule myself.

Specifically:

• Which tests belong in a baseline draw, taken before the first dose?
• Which of those are worth repeating, and how often?
• Which are one-and-done for life?
• Which are on the 58-analyte panel purely because the lab can bill for them?

I am not asking anyone to manage my care. I am asking how to construct a monitoring schedule that a reasonable clinician would not laugh at, so that when something does look wrong I have a trend to show them rather than a single number and a panic.

Adding the case against the long panel, quantitatively, because "it cannot hurt to test more" is false and the reason is arithmetic rather than opinion.

The multiple-comparisons problem on a routine panel

A reference interval is conventionally the central 95% of results in a reference population. That means a healthy person has, by construction, a 5% chance of falling outside the interval on any single analyte. If analytes were independent, the probability that all of them come back inside the interval is 0.95 raised to the power of the number of tests:

• 10 tests: 0.95^10 = 0.599. So a 40% chance of at least one flag.
• 20 tests: 0.95^20 = 0.358. A 64% chance of at least one flag.
• 58 tests: 0.95^58 = 0.0517. A 95% chance of at least one flag.

Read that last line again. On a 58-analyte panel, a perfectly healthy person has roughly a 95% probability of receiving at least one out-of-range result, and an expected count of 58 × 0.05 = 2.9 flagged values. The panel is designed, mathematically, to find something.

Two honest caveats on the arithmetic. Analytes are not independent — sodium and chloride move together, the liver enzymes correlate, and clustering reduces the effective number of comparisons somewhat. And some reference intervals are not true central-95% intervals; a few are outcome-derived thresholds where the 5% logic does not apply. Both effects pull the numbers down a little. Neither rescues the 58-item panel.

What the flags then cost

The cost is not the retest fee. It is that a flagged value with no baseline and no prior probability generates a diagnostic cascade: repeat, then imaging, then a specialist referral, then in the worst case a biopsy of something that was never going to matter. The asymptomatic lipase elevation is the canonical example in this drug class — elevations above the reference limit are common on GLP-1 receptor agonists without any pancreatic pathology, and pooled analyses of the liraglutide programme found that these elevations did not predict acute pancreatitis [1]. If you did not order the lipase, you lose nothing. If you did, you now have a number that will worry you at every subsequent draw.

The tests that are worse than useless

• Fasting insulin and HOMA-IR. Insulin immunoassays are not harmonised between platforms, cross-react with proinsulin to varying degrees, and have within-person day-to-day variation of the order of 20 to 30%. There is no threshold at which a clinician does something different. It is a number that generates a feeling.
• Random cortisol. Cortisol has a several-fold diurnal swing and responds to the venepuncture itself. A single value is uninterpretable outside a formal protocol with a defined sampling time.
• IgG food-sensitivity panels. IgG against food antigens reflects exposure, not intolerance. Multiple professional bodies have advised against these specifically.
• Serial thyroid autoantibodies. Once positive, positive. There is no serial information; the titre does not guide anything.
• Serial hs-CRP during active weight loss. It falls with adiposity, which you already know is happening, and rises with any minor infection, which you cannot exclude. The signal-to-noise is poor and the result changes nothing.

The exception worth stating: if you are the sort of person who will order the long panel anyway, order it once, at baseline, and then never again. A single wide snapshot is at least interpretively useful later. It is the annual repeat that turns it into a machine for manufacturing worry.

Practical addendum on the logistics, which turn out to matter as much as the test selection.

Use one lab

Reference intervals are method-specific, and so are the numbers. Two labs running different immunoassays for the same analyte can report results that differ by more than the change you are trying to detect. This is worst for the analytes without a good international reference method: insulin, free T4 by some platforms, ferritin, vitamin D, testosterone at female concentrations. It is least bad for creatinine, HbA1c, glucose and cholesterol, all of which have properly standardised reference methods behind them.

If you switch labs mid-series, treat it as starting a new series. Not because either lab is wrong, but because the step change you will see is a method difference and you will spend three months trying to explain it biologically.

Standardise the draw conditions

Things that move results by more than the drug does:

• Posture. Twenty minutes upright versus supine changes plasma volume by several per cent, and every concentration-based analyte with it. Standing draws read higher.
• Tourniquet time. Prolonged stasis raises potassium and haemoconcentrates everything.
• Hydration. A dehydrated draw during a bad GI week raises creatinine, urea, albumin, haemoglobin and haematocrit together. If four unrelated analytes all rose by a similar fraction, suspect the water, not the organs.
• Time of day. TSH is around 50% higher in the small hours than in mid-afternoon. Iron is higher in the morning. Cortisol, obviously.
• Recent exercise. Raises creatine kinase, ALT and AST for days, and can push urine albumin into the abnormal range transiently.
• Alcohol. GGT and triglycerides. A single heavy evening is visible.

The cheap fix is to always book the same slot: same lab, same time of day, same fasting state, no alcohol for 72 hours, no hard training for 48, and drink normally rather than either fasting dry or loading water. You will not eliminate variation, but you will stop adding to it.

Keep the numbers yourself

Ask for numerical results rather than "normal", and keep them in one file with the date, the lab, the dose you were on, and your weight that week. Portal access is revoked when you change provider, and results older than a few years are often archived out of easy reach. The whole value of the schedule above is that it produces a series; a series you cannot retrieve is not a series.