About UveitisDDx

Clinical status. Educational decision support for ophthalmologists and ophthalmology residents working under supervision. It is not a diagnostic device, does not replace examination and clinical judgement, and must never be used to exclude a sight- or life-threatening diagnosis.

Purpose

UveitisDDx ranks the differential diagnosis of uveitis from the findings you enter, using a literature-weighted percentage matrix (a naive-Bayes-style score) rather than a black-box model: every number in the app traces back to a published percentage, a regional prior, or a documented safety rule. It also runs the published classification criteria (SUN 2021, ICBD and others) against the same case, suggests the next question or test worth asking, and flags emergencies and organisms that steroids alone would make worse.

How to read the relative weights

The bar next to each diagnosis is a relative weight of evidence, not a calibrated probability - the underlying data are literature percentages of varying quality, not a validated epidemiological model. Treat the ranking as "which diagnoses this picture favours and by roughly how much", not as a percentage chance of disease.
Each finding shifts every diagnosis's score by how typical the finding is for that diagnosis compared with its background rate across all diseases - a nonspecific finding barely moves anything, a rare and typical one moves a great deal. Marking a finding absent counts against diagnoses that usually have it; a finding you never asked about never counts against anything.
Coverage is how much of what you entered a diagnosis actually explains. A leading diagnosis that leaves an alarming finding unexplained triggers the unexplained-finding banner - that finding needs its own differential.
Must not miss is ranked separately, by urgency times plausibility, so an emergency stays visible even when it is not the top-ranked explanation.
Ask / examine / test next suggests whichever question, sign or test would most reduce uncertainty among the leading diagnoses, per unit of cost.
Open why this rank on any diagnosis for the full breakdown: every term - the recorded value, its percentage, the background rate, the resulting contribution and its source - sums exactly to the shown score.

Evidence grades and citations

Every literature value carries a grade: A a large, well-designed series; B a smaller or single-centre series; C consensus, a review, or an unsourced estimate used only to keep the matrix complete. C-grade cells are always marked, in "why this rank" and in the admin editor, so they can be prioritised for better sourcing. A cell with no literature value at all is shown as "not reported: default rate" and never treated as zero. Citations link out - PMID: to PubMed, DOI: to the publisher - so every number can be checked against its source.

Disease pages link out to EyeWiki (American Academy of Ophthalmology). EyeWiki content is copyright AAO and is never copied into this app.

Privacy

The differential runs entirely on your device from the offline knowledge base. Nothing is sent anywhere unless you sign in and save a case, generate a PDF, or opt a case into the research pool.
No identifiers are ever collected: no name, MRN, date of birth or calendar date. Cases store an age in years (90 stands for "90 or older"), sex, region and free-text tags/notes - notes are scrubbed of anything that looks like a date, phone number, e-mail address or record number before saving, and a warning is shown before you save free text. Visits are dated only as a day offset from the case's first visit.
The private reference field (for example a hospital number, so you can find the real patient again) is stored only in this browser's local storage on this device - it is never sent to the server and never appears in any export.
Sharing a case with the research pool is opt-in, per case, and reversible. Only opted-in cases with a final diagnosis are ever included in de-identified, aggregate analysis used to improve the knowledge base; your identity is never part of that data.
Share links encode only variable and level codes, never free text.
Saved cases record the knowledge-base version that produced them, so past results can always be reproduced exactly.
Cases you create or edit while offline are queued on your device and sent when you are back online; nothing leaves the device before that.

Knowledge base

Loading knowledge base counts…

Clinical owner: Dr. Shehab Al-Abed. Built as an ASP.NET Core application with an offline JavaScript engine.