Most software is built to give an answer. In clinical reasoning, that is a risk. When the information is incomplete or the guidance doesn't cover a case, a confident-sounding answer can mislead more than no answer at all.
Abstaining by design
Our platform, which is in development, is designed to say when there isn't enough evidence to support a suggestion. We call this abstention. Instead of filling the gap with a guess, it tells the clinician what is missing: a test result, a part of the history, or a question the guidance doesn't answer.
Why this matters
Clinicians already know the difference between "this is clear" and "I need more information". A good senior colleague will say "I'm not sure, let's get the bloods back first." We want the platform to behave the same way, so clinicians can calibrate how much weight to put on what it shows them.
How we build it
- Grounded in guidance. Suggestions are tied to UK clinical guidance. If the guidance doesn't support a suggestion, the platform doesn't make one.
- Missing information is shown. Where a pathway needs a test or a detail that isn't in the record, the platform lists it rather than assuming.
- Tested on purpose. Our clinician-written test cases include deliberately incomplete and ambiguous cases, so we can check that the system abstains when it should.
Measuring it
We track how often the platform abstains, and whether it abstains in the right cases. A system that never says "I don't know" is as much a concern to us as one that says it all the time. We'll publish what we learn, including what didn't work. See Evidence & fairness.