The Clinicians That Code briefing

Top signal

A safety net for the referrals nobody chased Wayne Han Lee and colleagues, writing in npj Digital Medicine on 30 October, built MR-AI, which reads primary care consultation notes and flags referral intent that never turned into an order. In a quasi-experimental study across seven clinics in Singapore it scanned about 25,000 notes a week and raised weekly missed referral detection by 4.7 percentage points (p < 0.001), an estimated sevenfold increase. Of 428 cases flagged, 225 were confirmed by a clinician, alongside 31 that patients reported, for 256 detections at 99.9 per cent specificity and 48.8 per cent sensitivity. In a blinded one-week comparison against seven physician reviewers, positive predictive value was 61.8 per cent against a human confirmation rate of 51.8 per cent across 83 flagged cases, with about 41 fewer projected staff-hours. Read the sensitivity first: the net catches roughly half of what is there, so it is a second pair of eyes, not a replacement for the first. The authors say lasting improvement needs preventive workflow changes and a cost-effectiveness study, not just a better model. npj Digital Medicine

Also in this issue: shipped this week (3), build safely (2), tools & guides (1), events (1), new in the directory (1), from us (1).

Read issue 13