How do I interpret a magnesium trend across three draws?

The clinician who ordered the panel was not concerned; I would still like to understand it.

I would like to know the limits of what can be inferred from this.

What I am trying to avoid is over-reading a single result, which I have done before.

How should I read this, and where are the traps?

Answer first: decide what you would do differently for each possible result before you order the panel. Anything that fails that test is a number you will worry about and not act on.

Same laboratory, same method, same time of day, same fasting state. Between-laboratory differences on several common analytes are larger than the changes people are trying to detect.

A twenty-analyte panel run on a healthy person will produce, on average, one out-of-range result purely from how reference intervals are constructed. That is arithmetic rather than pathology.

Reference intervals are conventionally the central ninety-five per cent of a reference population, which is the direct cause of the one-in-twenty out-of-range rate on a healthy panel.

One out-of-range value on a twenty-analyte panel is expected. Two on a repeat is a finding.

edited 16 Jun 2025 by zeynep_arslan — corrected a unit error in the worked example

The relevant statistical point is that a ninety-five per cent reference interval means one analyte in twenty will read out of range in a healthy person by construction.

Timing matters per analyte: cortisol and testosterone are diurnal, triglycerides are postprandial, and creatinine responds to hydration and to recent training. Fixing the conditions removes most of the noise.

Repeat before you react. A single abnormal value has a substantial probability of being within the combined biological and analytical variation of a normal one.

Biological variation data are published per analyte and are the basis for the reference change value — the difference between two results that is larger than noise.

Baseline first, then a repeat under identical conditions. Everything else is secondary.

Before reacting to any single value, check whether it is outside the interval by an amount larger than the assay's own variation.

A sensible core for this population is a full blood count, renal function with electrolytes, liver enzymes with bilirubin, a fasting lipid panel with apolipoprotein B, HbA1c and thyroid-stimulating hormone.

External quality assurance schemes document between-laboratory differences on common analytes that routinely exceed the size of clinically interesting changes.

Research-use compounds are not approved for human use, and no panel makes that safer.

Keep the full report, not the number. You will need the units and the interval later.

edited 15 Jul 2025 by g_paskevicius — fixed an arithmetic slip in the third paragraph

Start with a baseline. A result taken before anything started converts most later ambiguity into a simple comparison, and it cannot be obtained retrospectively.

Delta checks — comparing against your own previous value — are far more sensitive than comparing against a population interval, which is the argument for keeping a series rather than a snapshot.

SELECT reported a hazard ratio of 0.80 (95% CI 0.72–0.90) for the primary composite major adverse cardiovascular event endpoint with semaglutide 2.4 mg in overweight or obese adults with established cardiovascular disease and without diabetes[1].

Same laboratory, same time, same fasting state, or the comparison is not a comparison.