MyopiAI Registry
The de-identified research registry beneath every surface — the flywheel: every consented child makes the model smarter, cleanly. A growing, inclusive Australian cohort, de-identified by construction, with the international registry minimum dataset captured end-to-end.
Cohort
Enrolment growth — the flywheel
Consented data compounds: every child added to the registry sharpens the next prediction, cleanly. More inclusive data in → a better engine out → more trust → more consent.
Modelled cumulative-enrolment index (synthetic) ending at the headline +41.0%12-month growth — a stylised illustration of compounding, not observed monthly enrolment.
IMI minimum dataset
captured
The International Myopia Institute (IMI) minimum datasetis the agreed crosswalk of mandatory registry fields — year of birth, sex, refraction (sphere / cylinder / axis → SER), BCVA, axial length, keratometry, and treatment category. Capturing 100% means every subject maps cleanly into international myopia registries.
Study arms
A retrospective backbone, growing prospectively — historical depth plus protocolised, forward-collected follow-up.
- Retrospective2,040
- Prospective1,360
Cohort diversity
Six ethnic groups, including Indigenous Australians — an inclusive local cohort, not a single-population dataset imported from abroad. A model fit on Australian diversity is the honest answer to “does this generalise to ourchildren?”
Shares of the consented cohort. Representation is not a performance claim — subgroup effect sizes live on the RWE dashboard.
De-identification & governance
The registry is de-identified by construction: identifiable clinical data stays in its practice-scoped schema; a server-side bridge promotes only consented, coarsened, pseudonymous rows into the research schema.
Only children whose families consent to research are promoted from the clinical record into the registry.
Name, MRN and contact details never leave the clinical schema — the de-id bridge copies only IMI/biometry fields.
Date of birth → year of birth + age band; address → region; ethnicity → group — re-identification surface reduced.
The clinical↔registry link is a one-way salted HMAC pseudonym — stable for follow-up, never reversible to PII.
Indigenous-community data is handled under the CARE principles (Collective benefit, Authority to control, Responsibility, Ethics) with a custodianship flag carried through to the registry. Honest limit: a salted hash protects the link, not the record — k-anonymity is enforced at the reporting layer (small-cell suppression), not by the bridge alone. Governance: Australian sovereign hosting · de-identified by construction · audit-logged.