enesru
Get the app

Science

Three layers, and none of them is a guess

Labless separates what can be calculated, what can be learned from data, and what must be measured. The language model only explains the result.

Layer 1

Validated formulas

Indices with decades of clinical use, computed in code and unit-tested. Sex, age and units are handled before any rule runs.

  • HOMA-IR, QUICKI, HOMA-B
  • FIB-4
  • eGFR (CKD-EPI 2021)
  • Mentzer index
  • TyG and TG/HDL
Layer 2

Models trained on population data

Large public datasets such as NHANES measured routine and specialized tests in the same people. That lets a model learn what one predicts about the other, and lets us measure how well it does.

  • Fatty liver from routine labs
  • Iron stores from the blood count
  • Biological age (PhenoAge)
Layer 3

A rules graph that connects results

Clinician-reviewed rules link markers, conditions and context. Each rule carries a source and an evidence grade. The engine looks for the single explanation that fits the most findings.

  • Sourced to guidelines
  • Evidence grade on every rule
  • Urgent values bypass everything

How we label confidence

Strong signalPublished formula or model with good accuracy.
Moderate signalUseful estimate. Confirm if it changes a decision.
Screening onlyTells you whether to look closer.
Needs a real testBlood work can't answer this well. We say so.

What published research already shows

0.88

AUC for predicting fatty liver from routine check-up data, ahead of the standard fatty liver index

0.84

AUC for flagging low ferritin from a standard blood count

~80%

of people with sleep apnea are estimated to be undiagnosed

Figures from independent studies listed below, not from Labless.

How we'll prove it

Planned

Held-out population data

Every model is scored on people it never saw during training, and we publish the results by age and sex.

Planned

Blinded comparison

Real anonymized cases reviewed by Labless, by physicians and by general AI assistants, then graded by experts who don't know which is which.

Planned

Looking forward in time

Using a person's older reports only, did Labless flag what was diagnosed years later? This is the hardest test, and the one that matters most.

References

  1. Matthews DR et al. Homeostasis model assessment: insulin resistance and β-cell function from fasting plasma glucose and insulin concentrations in man. Diabetologia, 1985. doi.org/10.1007/BF00280883
  2. Sterling RK et al. Development of a simple noninvasive index to predict significant fibrosis in patients with HIV/HCV coinfection (FIB-4). Hepatology, 2006. doi.org/10.1002/hep.21178
  3. Inker LA et al. New creatinine- and cystatin C–based equations to estimate GFR without race (CKD-EPI 2021). NEJM, 2021. doi.org/10.1056/NEJMoa2102953
  4. Levine ME et al. An epigenetic biomarker of aging for lifespan and healthspan (PhenoAge). Aging, 2018. doi.org/10.18632/aging.101414
  5. Chung F et al. STOP questionnaire: a tool to screen patients for obstructive sleep apnea. Anesthesiology, 2008. doi.org/10.1097/ALN.0b013e31816d83e4
  6. Machine-learning algorithm for predicting fatty liver disease in a Taiwanese population. www.ncbi.nlm.nih.gov/pmc/articles/PMC9317783/
  7. Advancing non-alcoholic fatty liver disease prediction: machine learning with SHAP interpretability and multi-cohort validation, including NHANES. www.ncbi.nlm.nih.gov/pmc/articles/PMC11493712/
  8. Beyond prediction: explainable machine learning for ferritin estimation from complete blood count. BMC Medical Informatics and Decision Making, 2026. pubmed.ncbi.nlm.nih.gov/42215930/
  9. National Health and Nutrition Examination Survey (NHANES). CDC, National Center for Health Statistics. www.cdc.gov/nchs/nhanes/