How it works
Five stages. Every case moves along them. Nothing is lost between hospitals.
The five stages
- Stage 1Report arrives
A scan report arrives as free text — echo, DXA, mammogram, CT or MRI.
- Stage 2AI reads it
AI extracts the test, due date and urgency, and quotes the exact sentence. If the quote isn't in the report or confidence is under 70%, it is flagged for a careful check.
- Stage 3Clinician checks
A clinician ticks “I have checked this against the report” and confirms or corrects. Nothing moves without them.
- Stage 4Tracker watches
Every day the tracker looks for a booking at every hospital. Overdue: clinician alert. 14 days later: nurse outreach.
- Stage 5Patient is reached
A plain-language message in the patient's language, approved by a clinician. The patient can reply with one tap.
Safety rules
- A human always decides
AI only suggests. A clinician confirms every follow-up and approves every message.
- Urgent means a phone call
Suspicious or urgent findings never get an automated message. A clinician phones the patient.
- No diagnosis in messages
Messages say which test is due and how to book — nothing about results.
- Translation safety check
Dates, phone numbers and names must survive translation exactly, or sending is blocked. A back-translation is shown.
- AI never answers a patient
Patient questions are sorted by AI and drafted for a clinician. Anything serious — or anything the AI is unsure about — means a phone call today.
- Every action logged
Each AI, system, clinician and patient action is written to the case's audit trail.
How accurate is the AI?
On 10 held-out synthetic reports written after the rules were built, an AI model (Gemini, in our Streamlit prototype) found 100% of recommended follow-ups (precision 90%), 100% of urgent ones, and got the timeframe right every time and the urgency right 89% of the time. Simple keyword rules found only 22%.
- Recommended follow-ups found
- 100%
- Precision
- 90%
- Urgent follow-ups found
- 100%
- Timeframe right
- 100%
- Urgency right
- 89%
- Keyword rules found
- 22%
Small synthetic test set — real validation on real reports would be needed before any clinical use. Precision under 100% is also why a clinician checks every result.
Why one hospital's system can't see the gap
When a report written at one hospital asks for a test at another, the request leaves the first system and never enters the second. Each system is right about its own bookings — and blind to everything else.
ResultWise sits across both. It holds the follow-up itself and checks every hospital's bookings — so it can raise the alarm when nothing is booked, and stay quiet when the booking is elsewhere.