Cite it or do not say it.

Evidence review · Research direction

The problem

Clinical teams are asked for answers from a literature that grows faster than anyone can read, and the sources vary from strong trials to weak reports. A fluent summary that hides that variation is worse than no summary.

The question

How should a system find, read and report evidence so that a clinician can see how strong each claim is and check it against the paper in under a minute?

Cite it or do not say it

On evidence summaries a clinician can check in a minute.

A clinician asks a question that has a literature behind it. The literature is large, uneven and moving. They want an answer they can stand behind, and they have a few minutes.

A fluent summary is the wrong thing to hand them. Fluency hides the thing they most need to see, which is how strong each claim is. A line drawn from a large randomised trial and a line drawn from a small case series read the same in a paragraph of confident prose.

The design rule is simple to state and demanding to keep: every statement links to the passage that supports it, and carries the type of study it came from. If a clinician can open the passage and see in under a minute whether the statement holds, the summary is useful. If they cannot, it is decoration.

The second rule is the more important one. The system must say when the evidence is thin, conflicting or missing, and must be allowed to give no answer. A confident reply to a question that has no good evidence is the most dangerous thing such a tool can produce.

We also keep two kinds of record apart. Our own reading notes are ours and may be wrong. A cited paper is someone else's work. Neither is a Mynd publication, and neither becomes a recommendation.

We publish no clinical guidance. This area produces summaries that can be checked, and the method for checking them.

Test design

• Each statement links to a passage in a named source, with the type of study visible.
• The system says when the evidence is thin, conflicting or absent.
• Reading notes and published conclusions are different records and are labelled so.

Protocol: Citation checkability test