Hayat Health catalogues, models and training
health.hayaterp.com ↗For the Hayat team (health.admin) — verify independent doctors' PMDC registrations; maintain the test catalogue (aliases, units, conversions, reference ranges) and the generic medicine catalogue (contraindications, interactions, allergy groups); train prediction models from opt-in de-identified data only, check AUC and calibration, activation thresholds, activate or retire; platform settings and statistics.
These screens are for the platform operators of Hayat Health: the Hayat team, or staff of the platform business with the permission health.admin. Subscribers' admins do not see them.
Verify doctors
The doctors list filters by status (pending, verified, rejected, suspended) and kind (hospital, independent). For a pending independent doctor:
- Open the doctor and their PMDC certificate.
- Check the name and PMDC number against the PMDC register online, and that the registration is current.
- Verify — the doctor gets an SMS and can now be added by patients. Or reject with a reason they will see (e.g. certificate unreadable — please upload again).
Suspend a doctor (reason required) to stop all their access at once — e.g. a complaint or an expired registration; reinstate when resolved. Hospital doctors are verified by their own hospital's permission, not here.
The test catalogue
Report reading and charts depend on the test catalogue — 43 tests are provided (HbA1c, glucose, creatinine, eGFR, lipids, CBC, thyroid, liver, vitamin D, CRP, and every reading kind). For each test:
| Field | Example |
|---|---|
| Code, LOINC, Name, Urdu name, Category | hba1c, 4548-4, HbA1c, its Urdu name, Diabetes |
| Unit, Decimals | %, 1 |
| Aliases | Spellings labs print: Glycated Hb, HbA1C, A1c |
| Conversions | Other units with factor / offset: mmol/mol → % |
| Ranges | By sex and age: low, high, critical low, critical high (and for a second value) |
| About (EN / UR) | One line for the patient |
When reports from a lab keep failing to match a test, add the lab's spelling as an alias. A test is never deleted — switch it off.
The medicine catalogue
26 generics are provided — the first-line drugs for diabetes, hypertension, lipids, thyroid, anaemia and vitamin D, plus common drugs kept only for interaction and allergy checks (NSAIDs, clarithromycin, warfarin, spironolactone, potassium). For each:
- Generic, class, form, strengths, for conditions and line (first / second / not suggested);
- Kidney: avoid below / reduce below an eGFR, with a note; liver avoid with a note; potassium avoid above; pregnancy (avoid / caution / OK);
- Allergy groups and interactions (with a medicine or a class, severity avoid / caution, note);
- Default dose, frequency, days and the guideline;
- Brands — informational examples only. Check availability and DRAP registration before relying on them.
Changes apply to the next suggestion and safety check; existing prescriptions are not changed.
Models and training
Rule flags work without training. Models add risk estimates for three families: diabetes onset, CKD progression and uncontrolled BP.
- Training data shows the rows per family and label, and how many patients have opted in. Rebuild refreshes it.
- Train (one family or all). Training also runs weekly by itself (setting).
- Each run makes a model version with its training rows, positive rows, test rows, AUC, Brier score, log-loss and accuracy on a held-out quarter of the data.
- A model is activated only when it passes all thresholds (below). With auto activate on, that happens by itself; otherwise activate it by hand. A model below the thresholds is marked below threshold and cannot be activated.
- Retire a model to stop its estimates; the rule flags continue.
Settings
| Setting | Default | Means |
|---|---|---|
| Minimum training rows | 300 | Rows needed to activate a model |
| Minimum positive rows | 30 | Positive labels needed |
| Minimum AUC | 0.75 | Discrimination needed |
| Maximum Brier | 0.20 | Calibration needed |
| Auto activate | On | Activate a passing model by itself |
| Show probability | 0.5 | Model estimates below this are not shown to doctors |
| Training day / hour | Sunday, 21:00 UTC | Weekly training |
| Default missed days | 3 | Missed-reading reminder default |
| Baseline SDs / minimum readings | 2.0 / 10 | Personal-baseline alerts |
| Default consent days | 365 | Length of a new consent |
| Report size / files | 15 MB / 10 | Upload limits |
| Low confidence | 0.80 | Report values below this are highlighted for checking |
Statistics
The statistics page shows counts only — patients, active patients, opted-in patients, doctors by status, active consents, reports by status, readings in 30 days, open alerts by severity, predictions by status, training rows and active models. No patient data is shown, and Hayat Health has no Hayat Reports datasets.