All studies
Research·November 8, 2026·7 min read

Can Claude Compare My Last Two Blood Tests and Show What Changed

Claude can compare two blood tests you share and show what changed and by how much. Many differences between draws are noise, so treat the answer as a question for your doctor.

Yes, Claude can compare your last two blood tests and list which markers moved and by how much. With the HealthOSX connector, Claude can read the blood tests you chose to share. If you prefer not to connect anything, in the Claude apps you can attach a report as a PDF or photo instead. A change is not automatically meaningful, though, because many differences between two draws are smaller than the normal wobble of the marker itself.

What Claude needs to compare two blood tests accurately

The minimum Claude needs for each result is a record: marker name, number, unit and date. With that it can line up ferritin against ferritin and LDL against LDL and do the subtraction.

How the results reach Claude matters. If you attach a PDF or photo, reading can go wrong. A 5.4 can come out as 5.1, a decimal point can slip, and small print such as units or the collection date can be missed. Transcription errors are possible and worth checking. Data that comes through the connector may not carry those reading errors, because nothing has to be read off a page.

Some questions need more than the basic record. Flagging out-of-range values requires the reference range, saying whether the lab changed requires the lab name, and judging a triglyceride change requires knowing whether you fasted. The HealthOSX pages do not specify whether reference ranges, lab name or fasting status come through the connector. Before you trust a comparison, ask Claude to list exactly what it received for each result. If a field is missing, Claude cannot use it, and you can add it yourself from the report.

What Claude does well, and what it can't know

It can do the mechanical part well: match markers across two dates and calculate the absolute and percentage difference. If the reference ranges are included, it can also flag which values sit outside the range on each date.

What it cannot do is know why anything moved. It does not know that you fasted for the first draw and not the second, or that you had a cold that week. Unless you tell it, or the data carries it, the comparison is arithmetic with no context.

A change between draws is not automatically meaningful

Two things make a repeat measurement differ from the first even when nothing in your body has changed. The lab method has its own imprecision (analytical variation). And your body genuinely fluctuates from day to day for most markers (within-person biological variation). Laboratory medicine combines the two into the reference change value (RCV), the smallest difference between two results that is likely to be real at 95% confidence:

RCV = 2.77 × √(analytical² + biological²)

Here is the arithmetic for a marker whose combined variation is 6%: 2.77 × 6 = 16.6, so the value would need to move by about 17% before the change is likely to be real. Markers differ a lot. Sodium varies far less than 6% from day to day in most people, so a small shift in sodium can matter more than a much larger one in triglycerides. Triglycerides vary well above that, which puts their threshold for a real change well above 17%. This is why a percentage change on its own tells you almost nothing.

The worked example below uses made-up numbers to show how the checks play out.

Worked example: four changes, two that clear the bar
MarkerBefore → afterChange vs assumed RCVVerdict
LDL cholesterol (mmol/L)3.4 → 3.1-8.8% vs 20%Within expected noise
Ferritin (µg/L)80 → 40-50% vs 35%Beyond noise, ask your doctor
Sodium (mmol/L)140 → 134-4.3% vs 3%Beyond noise, despite a small percentage
Triglycerides (mmol/L)1.2 → 1.9+58% vs 65%Within expected noise, despite a large percentage

Scroll the table sideways for the rest →

Illustrative numbers only. The RCVs are assumed for this example and are not reference values for any lab or person.

Variation figures for individual markers are published in resources such as the EFLM biological variation database from the European Federation of Clinical Chemistry and Laboratory Medicine.

There is a second trap, regression to the mean. If your first result was unusually high, a repeat is likely to come back closer to your usual level even if you did nothing. People credit the supplement they started in between.

What else shifts between two draws

Before reading anything into a difference, check these:

  • Fasting status. Triglycerides and glucose respond to your last meal.
  • Time of day. Cortisol, iron and testosterone are higher in the morning and drift down.
  • Lab and method. Different labs use different assays and print different reference ranges. A move from one lab to another can shift a value with no change in you.
  • Recent hard exercise, illness or dehydration. Any of them can nudge muscle enzymes, inflammatory markers and kidney markers.
  • Biotin supplements. High doses are known to interfere with some immunoassays, including some thyroid and hormone tests.
  • What the marker measures. HbA1c reflects roughly the past two to three months, so a gap of a few weeks between draws cannot show a full change.

What a good before-and-after question looks like

Specific questions get answers you can use.

  • Compare my last two blood tests for LDL, HbA1c and ferritin. Show both values, units and dates in one list.
  • Which markers changed by more than normal day-to-day variation would explain, and which are probably noise?
  • If both reports show the lab and the reference ranges, tell me whether either changed between the two.
  • Which values crossed in or out of the reference range, and by how much?
  • Write me three questions to bring to my doctor about the changes that look real.

A caveat on the second question: the HealthOSX pages do not say that variation figures come with your results. Claude would be drawing on general figures that may be approximate and are not specific to your lab or to you. Ask it which figure it used for each marker and where that figure comes from. Treat the answer as a rough screen, not a verdict, and use it to decide what to ask your doctor. The last two questions only work if the ranges and lab details are in the data Claude has.

Tell it the context it cannot see: fasted or not, time of draw, new supplements, illness, a change in training or diet in between. The more of that you give it, the less it has to guess.

If a marker clears the noise bar

A marker that moves by more than its RCV is worth a closer look, not a conclusion. Write down a short note for each one:

  • The date of both draws.
  • Whether you fasted for each.
  • Any supplements, medicines, illness or hard training in the days before.

Bring that list, with both reports, to your clinician and ask what they make of the change.

Can I trust the answer, and what do I do with it?

Check it. Ask Claude to show both raw values next to each other and glance at them against your reports; if a digit is wrong you will see it in ten seconds. Ask it to say what it is unsure of. It can be wrong, and it can sound confident while being wrong.

Two draws also make a very thin trend. A line needs more points, and a clinician can see context that Claude cannot, such as your history, your medicines and an examination. Some of this is still argued about: how tightly to apply population-based variation figures to one individual, since your own variation may be larger or smaller than the average. Anyone with a known condition, or on medication that is monitored with blood tests, should leave interpretation to the clinician who manages it.

HealthOSX can be added in Claude as a custom connector with one link. In Claude, go to Settings → Connectors → Add custom connector, add the link and sign in. Then tick blood tests in the share list, because Claude can only see the items you tick. Start a new chat and switch HealthOSX on from the tools menu. On Team and Enterprise plans, an owner may have to add the connector first. The HealthOSX page's own example is the question you came here with: Summarise my last blood test and what changed since the one before. HealthOSX is wellness information, not a medical device. An assistant can be wrong, so check anything important with a doctor. Setup steps are on the HealthOSX for Claude page.

Get your blood testedThe HealthOSX Blood Test: a finger-prick kit at home in Germany, a nurse at home in Saudi Arabia, a partner lab elsewhere. Results right here.Blood Test

Written by

HealthOSX Research

Keep reading