AI Blood Comparison Tool: Spot Meaningful Lab Changes

How an AI blood comparison tool judges lab changes

Kantesti is an AI blood test interpretation platform that compares laboratory visits in context, and our story explains why we built it around patient-readable reasoning rather than isolated flags. I am Thomas Klein, MD, and in clinical practice I see the same pattern weekly: a patient panics over a new asterisk, but the old result was already drifting in that direction for 18 months.

A multiple blood test comparison should never treat a new cholesterol, creatinine or ferritin value as a courtroom verdict. A creatinine of 1.05 mg/dL may be harmless in a muscular 29-year-old, but the same value rising from 0.62 mg/dL in a 72-year-old woman over 6 months is a very different story.

Our AI first builds a blood test timeline from every uploaded PDF or photo, then normalizes units and compares like with like. The useful output is not just “higher” or “lower”; it is “higher by 23%, measured after non-fasting, while taking a new diuretic, and still inside the lab range.”

When a lab change is noise versus a real signal

Laboratory medicine uses the reference change value concept to compare two consecutive results; Harris and Yasaka described this statistical approach in Clinical Chemistry in 1983. The simplified formula is about 1.96 × √2 × √(CVa² + CVi²), where CVa is assay variation and CVi is the patient’s normal biological variation.

Kantesti AI applies this logic with a patient-friendly explanation, which is why a 6% sodium shift and a 6% ALT shift do not receive the same urgency. Our engineering approach is described in the technology guide, but the clinical idea is simple: each biomarker has its own usual wobble.

I often tell patients that the flag is the headline, not the story. A deeper review of blood test variability helps explain why platelets can move 40 × 10⁹/L between visits without danger, while calcium moving from 9.4 to 10.8 mg/dL deserves a cleaner repeat and medication review.

Why units and lab methods can fake a blood test difference

A multiple blood test comparison across countries must translate units before it interprets direction. Cholesterol in mmol/L can be converted to mg/dL by multiplying by 38.67, while triglycerides use 88.57; mixing those two conversion factors is a common source of wildly wrong conclusions.

Lab methods matter too. A direct LDL-C result and a calculated LDL-C result can differ by 10-25 mg/dL when triglycerides are high, so our AI does not treat them as identical measurements; patients can read more about different lab units when a report suddenly looks unfamiliar.

Some European laboratories use slightly different reference intervals for TSH, ferritin, vitamin D and liver enzymes than US laboratories. That is not sloppiness; reference ranges depend on the local population, assay manufacturer and calibration method, which is why the prior result from the same lab often carries more weight than a generic online range.

How fasting, timing and hydration change the comparison

Triglycerides can rise by 20-50 mg/dL after eating, and in insulin-resistant patients the post-meal rise can be larger. That is why a non-fasting triglyceride of 190 mg/dL after a heavy lunch may not mean the same thing as a fasting 190 mg/dL at 8 a.m.

Hydration alters concentration-based markers. Hematocrit, albumin, sodium, urea and creatinine may all appear higher after sweating, vomiting or low fluid intake; the comparison should ask whether body weight, urine color or exercise changed in the previous 24 hours.

The thing is, timing can be the whole diagnosis. Cortisol is normally highest in the early morning, TSH often peaks overnight, and fasting can raise bilirubin in people with Gilbert syndrome; our guide to fasting versus non-fasting explains which markers are most vulnerable.

Medication and supplement clues the AI should check

Kantesti is an AI-powered blood test analysis tool used by 2M+ people across 127 countries, and medication context is one reason multilingual interpretation matters. A patient in London may call it prednisolone, a patient elsewhere may say steroid tablets, but the lab pattern of higher neutrophils, lower eosinophils and higher glucose can look similar.

High-intensity statins often lower LDL-C by about 50%, moderate-intensity statins by about 30-49%, and ezetimibe commonly adds another 15-25% reduction. If LDL-C falls from 164 to 82 mg/dL after 10 weeks, that is not random noise; it is the expected pharmacology showing up on the page.

Supplements can mislead the comparison as well. Biotin at 5,000-10,000 mcg daily can interfere with several immunoassays, iron can raise ferritin over 6-12 weeks, and creatine can raise measured creatinine without true kidney injury; our medication timeline is useful when dates are fuzzy.

Why trend direction beats one abnormal result

The slope matters. Ferritin rising from 9 to 24 ng/mL after 8 weeks of iron is usually a good early response, while ferritin falling from 58 to 22 ng/mL over a year suggests ongoing loss or under-replacement even if the lab still says “normal.”

Different markers have different clocks. HbA1c reflects roughly 8-12 weeks of glycaemic exposure, TSH should usually be rechecked 6-8 weeks after a thyroid dose change, and CRP may fall by half within 24-72 hours after an acute infection starts settling.

A good lab trend graph shows slope, spacing and swings, not just dots. I get more concerned when three values move in the same direction across 6-18 months than when one value jumps once after a marathon, fever or poor night of sleep.

Lipid panels need fasting status, risk context and pattern checks

The 2018 AHA/ACC cholesterol guideline, published in Circulation in 2019, treats ApoB ≥130 mg/dL and triglycerides ≥175 mg/dL as risk-enhancing factors in selected adults (Grundy et al., 2019). That matters because a patient with LDL-C 118 mg/dL and ApoB 142 mg/dL may carry more particle-related risk than the LDL alone suggests.

A fasting triglyceride under 150 mg/dL is generally considered desirable, while 150-499 mg/dL is elevated and ≥500 mg/dL raises concern for pancreatitis risk. Non-fasting triglycerides are often acceptable for screening, but a big jump between visits should be repeated fasting before anyone labels it a true worsening.

For patients comparing statin response, I like to see the percentage change from baseline. A drop from LDL-C 190 to 122 mg/dL is a 36% reduction, which may be decent but not the expected 50% response to high-intensity therapy; the difference between lipid profiles and panels becomes very practical here.

Glucose, HbA1c and insulin change on different clocks

Fasting glucose of 100-125 mg/dL is commonly classified as impaired fasting glucose, and ≥126 mg/dL on repeat testing supports diabetes diagnosis. HbA1c of 5.7-6.4% suggests increased diabetes risk, while ≥6.5% supports diabetes diagnosis when confirmed in the right clinical context.

I see a classic mismatch after holidays or steroid bursts: fasting glucose jumps to 132 mg/dL, but HbA1c remains 5.6% because the exposure was too recent. The reverse happens after weight loss, where fasting glucose improves quickly while HbA1c takes another 6-10 weeks to fully reflect the change.

The most useful multiple blood test comparison combines glucose, HbA1c, triglycerides, HDL-C, ALT and sometimes fasting insulin. If HbA1c is normal but triglycerides are 240 mg/dL and fasting insulin is high, our guide to A1c versus fasting sugar explains why risk can still be present.

Kidney and electrolyte changes need fast safety sorting

The KDIGO 2024 CKD guideline defines chronic kidney disease by kidney abnormalities present for at least 3 months, including eGFR below 60 mL/min/1.73 m² or urine ACR ≥30 mg/g, which is about 3 mg/mmol (KDIGO, 2024). A single eGFR of 58 after dehydration is not the same as three eGFR values under 60 across a year.

Creatinine is muscle-sensitive. A bodybuilder may sit at 1.25 mg/dL with normal filtration, while an older frail patient can have a “normal” creatinine of 0.75 mg/dL despite reduced kidney reserve; urine ACR often finds earlier risk, as explained in our urine ACR guide.

Urea, BUN and the BUN/creatinine ratio are especially vulnerable to hydration, protein intake and gastrointestinal fluid loss. If BUN rises from 14 to 31 mg/dL while creatinine barely changes, the BUN creatinine ratio can point toward dehydration or high protein intake rather than intrinsic kidney injury.

Clusters beat isolated liver, CBC and inflammation flags

ALT is often more liver-specific than AST, but AST can rise after heavy exercise, muscle injury or alcohol exposure. A 52-year-old marathon runner with AST 89 IU/L and normal ALT may need CK and training context before anyone assumes liver disease.

CRP under 3 mg/L on a high-sensitivity assay can be used for cardiovascular risk stratification, but CRP above 10 mg/L usually suggests an acute inflammatory process and should not be read as routine heart-risk data. White cells, neutrophils, lymphocytes and platelets help decide whether the CRP direction fits infection, inflammation, medication effect or recovery.

CBC indices are remarkably useful over time. RDW above about 14.5% varies by lab but often signals mixed cell sizes; when RDW rises with falling MCV and ferritin, an iron-deficiency timeline becomes much more plausible, and our RDW guide pairs well with a full panel review.

Personal baseline matters more than a generic range

Typical adult hemoglobin ranges are roughly 13.5-17.5 g/dL for men and 12.0-15.5 g/dL for women, but the patient’s own history matters. A woman whose hemoglobin falls from 14.2 to 12.1 g/dL in 9 months may be developing iron loss even while still inside many lab ranges.

Creatinine is another baseline marker. A 0.25 mg/dL rise may be trivial in one person and significant in another, especially when body size is small; our article on sex-based lab ranges explains why one-size ranges are blunt tools.

Hormone timing is one of those areas where context matters more than the number. Progesterone, estradiol, FSH and LH can be uninterpretable without cycle day or menopause status, which is why our women’s health guide focuses on timing notes as much as lab values.

Data quality checks protect patients from false comparisons

Kantesti is an AI biomarker interpretation platform that reads uploaded blood test PDFs and photos, but our system still treats OCR as a clinical safety step rather than a magic trick. A potassium of 4.8 mmol/L misread as 48 mmol/L is not a small typo; it completely changes urgency.

The most common patient-upload problems are cropped reference intervals, unclear decimal points, old reports mixed into new folders and country-specific unit swaps. Our PDF upload checklist is deliberately boring because boring checks prevent bad interpretation.

Hemolysis is the classic pre-analytical trap. A hemolyzed specimen can falsely raise potassium by roughly 0.5-1.5 mmol/L, and sometimes more, so a potassium jump without symptoms or kidney change should trigger a sample-quality question; our methodology and clinician review standards are described in medical validation.

How to use the comparison before your doctor visit

A useful doctor-visit summary should state the marker, the old value, the new value, the percent change and the likely context. “ALT rose from 28 to 73 IU/L over 4 months after starting a supplement” is much more actionable than “my liver test is high.”

Same-day medical advice is sensible for severe electrolyte shifts, chest pain with abnormal troponin, hemoglobin near or below 7 g/dL, platelets under 20 × 10⁹/L, or glucose above 300 mg/dL with dehydration, vomiting or confusion. AI can highlight these thresholds, but a human clinician must decide what they mean for the person in front of them.

If you want to test the workflow, upload one old and one new report to try analysis and then turn the result into a doctor checklist. Most patients find that a one-page timeline lowers anxiety because it replaces a pile of red flags with 3-5 concrete questions.

Research, privacy and clinical oversight behind comparison tools

Kantesti AI supports 75+ languages across 127+ countries, which makes unit handling, terminology and patient context unusually important. A ferritin result reported in ng/mL, µg/L or pmol/L can look familiar or foreign depending on the country, but the medical question stays the same: did the patient’s iron story actually change?

Our research publications include Figshare DOI records on gastrointestinal symptom interpretation and women’s health timing, both relevant to lab timelines because fasting, stool symptoms, cycles and hormonal phase can alter what a result means. Thomas Klein, MD reviews this material with the clinical team because comparison tools need physician skepticism, not just pattern detection.

No AI should pretend to diagnose from a laboratory report alone. Kantesti’s medical reviewers help define when the platform should reassure, recommend a repeat test, or tell a patient to seek urgent care; in my experience, that boundary is where trust is either earned or lost.

Frequently Asked Questions