Your Result Went Up.
Did anything actually change?

Start with my own result

I have a raised lipoprotein(a), usually written Lp(a): a cholesterol-carrying particle that's mostly set by your genes and linked to heart disease risk. I'm declaring that up front, because it's the example.

Two of my tests read 183.6 and then 216.8 nmol/L. That's a rise of 18%.

When I gave those two numbers to my own AI assistant, AIdan, three times in testing, it called the change "unchanged", "a meaningful rise" and "not automatically a trend". Three answers to the same two numbers. That's not an AI problem only: plenty of people, and plenty of practitioners, would read 18% either way depending on their mood. There is a better way to decide.

Idea 1: how big does a change have to be? The reference change value

Two things make a result move even when nothing about your health has changed:

• Your own biological variation. Most markers drift around your personal set point from day to day and week to week. Laboratory scientists measure this in healthy volunteers and call it the within-person variation (CVi).
• The lab's own variation (CVa): run the same sample twice and you won't get exactly the same number.

Put those together and you get the reference change value (RCV): the smallest change between two results that's probably real rather than noise. The standard formula, at 95% confidence, is:

RCV = 2.77 × √(CVa² + CVi²)

(2.77 is √2 × 1.96, because two results each carry their own wobble.) This is mainstream clinical chemistry, not a functional-medicine idea (Fraser, 2011). The within-person variation for hundreds of markers is collected in a free European database maintained by the European Federation of Clinical Chemistry and Laboratory Medicine (Perich et al., 2015).

Grade: established laboratory method. Its main limits: the variation figures come mostly from healthy volunteers, and they can differ between populations (Ma et al., 2022).

Back to my Lp(a)

The best current estimate comes from a 2025 UK study that sampled 18 healthy people weekly for six weeks. Lp(a)'s within-person variation was 10.9%, and the reference change value worked out at +31.6% / −24.0% (Antwi et al., 2025). The up and down thresholds differ because the calculation allows for skewed data. The European database's figure is close, at about 10% (as tabulated by Bio-Rad, 2025, from the 2021 EFLM database; I couldn't read the database directly). An older study put it lower, at 8.6% (Panteghini & Pagani, 1993).

Using the simple formula with 10.9% and a lab variation of about 3%:

2.77 × √(3² + 10.9²) ≈ 31%

My rise was 18%. So that's within the normal wobble: not evidence of a real change. Even with the older, lower 8.6% figure the threshold is about 25%, and the answer is the same. Not "unchanged", not "a meaningful rise": not distinguishable from noise on two results.

Units note: Lp(a) is reported in nmol/L (particle number) or mg/dL (mass). There's no reliable fixed conversion between them, because the particle's size varies from person to person — so I give only the unit my lab reported.

A more everyday example: TSH

The large European study put the within-person variation of TSH (the thyroid-stimulating hormone on most thyroid tests) at 17.7% (Bottani et al., 2021). With a 3% lab variation, that gives an RCV of about 50%.

So a TSH going from 1.8 to 2.6 mIU/L (the same as µIU/mL) — up 44% — may be nothing more than the normal wobble, even though it looks like a big jump. I'd want a third result, taken under the same conditions, before reading anything into it.

Idea 2: does the "normal range" even fit you? The index of individuality

Some markers vary a lot between people but very little within one person. For those, the population range is so wide that your own result can move a long way — up or down — and still sit comfortably "in range".

Clinical chemists measure this with the index of individuality: within-person variation divided by between-person variation (Harris, 1974). When it's low, a population range is a poor guide to whether your result is normal for you; your own previous results are a better one.

Lp(a) is the extreme case: an index of about 0.1, with between-person variation of about 86% against about 10% within a person (EFLM figures via Bio-Rad, 2025; the 1993 study found much the same, 86% against 8.6%).

This is the honest version of an argument functional medicine often makes badly. "Normal ranges are too wide, so use optimal ranges" usually ends in a different set of unsourced numbers. Clinical chemistry's own argument is better: for some markers, your own baseline is the right comparison, and the method for using it already exists.

Grade: established laboratory concept. It tells you when to trust your own trend over a population range. It doesn't tell you what your ideal value is.

Idea 3: when the numbers and the person disagree

The third problem has no formula. Sometimes a result looks bad and the person feels fine, or the person feels awful and every result looks fine. A stool test reports a bacterium in someone with no symptoms; a scan shows a "finding" in someone with no pain.

No statistic settles that. What helps is the same thing as always: a question first, results that agree with each other and with the symptoms, and a willingness to say "this doesn't fit, so we don't act on it yet". That's the subject of The Test Isn't the Investigation.

What this means for you

• Don't act on one change between two results unless it's larger than that marker's normal wobble, or something else points the same way.
• Keep the conditions the same: same lab where possible, same time of day, same fasting state, no hard training the day before.
• For a real trend, you usually need three or more results, not two.
• Bring your old results. For many markers, your own history is a better comparison than the population range.
• Big, sudden or worrying changes still go to your GP — this is about telling small changes from noise, not about ignoring large ones.

Declared interests: I sell blood testing and the Health Audit, which reviews results across years. The same standard applies to my own results, as you've just seen.

References

1. Antwi K, Downie P, Mbagaya W. Determination of the biological variation and reference change value of lipoprotein (a). Ann Clin Biochem 2025;62(5):342–351. PMID 39947649. doi:10.1177/00045632251324063
2. Bio-Rad. Biological variation: updated values only (Q-1680-U), March 2025. Values derived from the 2021 EFLM Biological Variation Database. bio-rad.com PDF. Accessed 3 Oct 2026.
3. Bottani M, Aarsand AK, Banfi G, et al. European Biological Variation Study (EuBIVAS): within- and between-subject biological variation estimates for serum thyroid biomarkers based on weekly samplings from 91 healthy participants. Clin Chem Lab Med 2021;60(4):523–532. PMID 33561908. doi:10.1515/cclm-2020-1885
4. Fraser CG. Reference change values. Clin Chem Lab Med 2011;50(5):807–812. PMID 21958344. doi:10.1515/CCLM.2011.733
5. Harris EK. Effects of intra- and interindividual variation on the appropriate use of normal ranges. Clin Chem 1974;20(12):1535–1542. PMID 4430131.
6. Ma L, Zhang B, Luo L, et al. Biological variation estimates obtained from Chinese subjects for 32 biochemical measurands in serum. Clin Chem Lab Med 2022;60(10):1648–1660. PMID 35977427. doi:10.1515/cclm-2021-0928
7. Panteghini M, Pagani F. Pre-analytical, analytical and biological sources of variation of lipoprotein(a). Eur J Clin Chem Clin Biochem 1993;31(1):23–28. PMID 8439593. doi:10.1515/cclm.1993.31.1.23
8. Perich C, Minchinela J, Ricós C, et al. Biological variation database: structure and criteria used for generation and update. Clin Chem Lab Med 2015;53(2):299–305. PMID 25415636. doi:10.1515/cclm-2014-0739