Healthcare data workflows
From clinical dataset to results you can trace.
Med-AId helps clinicians and researchers move from clinical datasets to publication-ready insight with no-code statistics, machine learning, and explainability, while every result stays traceable to the computation behind it.
Demonstration
A synthetic study runs end to end: data from a file or from the EHR through SQL, then preprocessing, missing-data imputation, statistics and models, and an analytics dashboard.
All names, values, and records in this demonstration are invented. It illustrates the workflow and is not a screenshot of a deployed system.
Computed in code, not by a language model
Study question
Which factors predict ICU admission within 48 hours of an emergency admission?
1. Data source
Done
SELECT a.admission_id, p.age, p.sex,
v.resp_rate, v.spo2, l.lactate,
l.crp, l.wbc, a.icu_within_48h
FROM ed_admissions a
JOIN patients p USING (patient_id)
LEFT JOIN first_vitals v
USING (admission_id)
LEFT JOIN first_labs l
USING (admission_id)
WHERE a.admitted_at >= '2024-01-01'
AND a.admitted_at < '2026-01-01';ed_admissions_2024_2025.csv
2. Preprocessing
Done3. Missing data
DoneMissing values by variable
Multiple imputation by chained equations (MICE), 5 datasets. The method and seed are recorded with the results.
BMI is 22% missing. Check whether it should stay in the model.
4. Analysis and models
Done5. Analytics dashboard
DoneGradient boosting AUC 0.84Logistic regression AUC 0.79
| ICU | No ICU | p | |
|---|---|---|---|
| Age, median | 71 | 63 | <0.001 |
| Respiratory rate | 24 | 18 | <0.001 |
| Lactate, mmol/L | 2.8 | 1.4 | <0.001 |
| Female | 44% | 51% | 0.002 |
Capabilities
Automated detection of variable types, distributions, and quality issues before any analysis runs.
Imputation strategies chosen for clinical data realities, with the choice recorded alongside the result.
Descriptive, inferential, and survival analysis guided step by step, from research question to method.
Train, tune, and benchmark predictive models, including triage-style risk models, without writing code.
SHAP values and feature importance show what drives each prediction, patient by patient.
Workflows, queries, and models are recorded, so tables, figures, and methods text can be reproduced and checked.
State the research or operational question and the population it concerns.
Bring extracts from EHR, laboratory, and SQL sources into one reviewed dataset.
Run statistics or compare models. Deterministic computation does the maths.
Inspect the drivers of each prediction and check them against clinical knowledge.
Export tables, figures, and a methods trail that another team can reproduce.
Good test-set performance is only the first of four checks. A practical guide to the others, using the reporting guidelines the field already agrees on.
A prediction without a sense of how sure it is invites both over-trust and dismissal. Why calibration, case-level uncertainty, and the option to abstain belong in clinical AI.
Claims that AI saves hours are easy to make and hard to check. A practical approach to measuring what actually changes before, during, and after rollout.
Request a demo
Med-AId is already selected. Change it if you’d like to see something else.
