Sector
Healthcare
Predict the expensive event before it happens, from data the health system already has — and explain why, so the output can support review, planning, prioritization, and further validation.
§ 01 — what we do
What we assess, model, and demonstrate.
Healthcare engagements cover data assessment, risk-model research and development, population-health analysis, explainability, model evaluation, and prototyping with an analytical roadmap. Every engagement starts by assessing whether the data you hold can support the decision you want to make.
- Risk stratification on data you already hold
- Identifying who is at risk from records a health system already collects — survey instruments, demographics, physical examination — with no new collection program.
- Scoring without laboratory data
- Risk models built and demonstrated in R&D on public survey data, from survey, demographic, and physical-exam inputs alone — the approach that would make population-level screening possible without drawing blood.
- Explainability as a deliverable
- Each score comes with the factors that drive it, ranked, and what happens to the risk when one of them moves — scoped as a deliverable in its own right, not an afterthought.
- Population layer above the individual layer
- Individual risk is one product. Where disease is spreading, where exposure concentrates geographically, and what a health authority should do about it is another.
The data problem
Data quality, sampling, missingness, and class imbalance are often the limiting factors in health-risk modeling. Thousands of candidate attributes per patient. The event you care about is, thankfully, rare. Missingness that isn't random, because who gets tested is itself a clinical decision.
A model that scores well on a convenient sample and fails on the population gives a confident wrong answer, so the engagement starts there — class balance, missingness, and how the sample was drawn — before any modeling. The ensemble methodology we apply to national-survey-class health data follows a published, peer-reviewed method; the paper is independent of Analytica.
Recent work
Our most recent healthcare delivery is an AI transformation assessment for a health-portal operator, completed in 2025 with an academic collaborator: an assessment of the operator's data and AI readiness, and a roadmap for what to build next.
A live demonstration
You can try the approach this page describes. GlucoGuard produces a risk estimate together with the factors driving it, ranked — the same explainability the engagements above deliver.
Live application
GlucoGuard ↗
Six metabolic risks, and what is driving each one
Estimates diabetes, pre-diabetes, cardiovascular, hypertension, kidney and fatty-liver risk from measurements a routine physical already produces — then shows a SHAP attribution for every prediction, so you can see which inputs drive each estimate. Each input is placed against the NHANES population distribution. A public demonstration, not a medical device: nothing it outputs is a diagnosis or a substitute for clinical advice.
This is a demonstration, not a clinical tool. It is trained on NHANES, a public US survey, and it exists to show how a model can be made explainable. It is not a medical device, it has not been through regulatory clearance, and nothing it outputs is a diagnosis or a substitute for advice from a clinician.
Discuss a healthcare data use case
Bring a risk-stratification, prediction, or measurement question. We'll assess what your existing clinical and operational data can support.