AI System Detects More Missed Specialist Referrals in Primary Care
An artificial intelligence safety system helped identify substantially more missed specialist referrals across seven primary care clinics in Singapore, according to a study published in npj Digital Medicine on 30 October 2026. The system, called MR-AI, scans consultation notes for signs that a clinician intended to refer a patient to a specialist but that no corresponding referral order was placed.
The researchers evaluated the system in a quasi-experimental study involving around 25,000 consultation notes each week. They estimated that its introduction produced a sevenfold increase in weekly detection of missed referrals, an absolute increase of 4.7 percentage points. The findings suggest that AI can provide an additional safety check when a referral discussed during a consultation does not appear in the clinical record.
How the AI System Identified Missed Specialist Referrals
MR-AI searches primary care notes for referral intent and checks whether a referral order exists. If the record suggests that a referral was intended but not ordered, the system flags the case for review. It does not independently make clinical decisions or replace the clinician responsible for patient care.
During the evaluation, the system flagged 428 cases. Clinicians verified 225 of them, while a further 31 cases were reported by patients, bringing the total number of detections to 256. The researchers reported 99.9% specificity and 48.8% sensitivity. This means the system rarely flagged cases that were not missed referrals, but it also failed to identify a substantial proportion of actual cases.
AI Detection Compared With Physician Review
The study also included a blinded one-week comparison involving seven physician reviewers. Across 83 flagged cases, MR-AI recorded a positive predictive value of 61.8%, compared with a 51.8% human confirmation rate. The researchers estimated that using the system in this comparison required about 41 fewer staff-hours than the physician review process.
These results indicate that automated screening may help clinical teams focus their attention on records that need checking. However, the comparison was limited to the cases and period examined. It does not establish that AI review will always outperform clinicians or reduce staffing requirements in other healthcare settings. The results also do not show that the system can independently resolve missed specialist referrals.
Why Missed Specialist Referrals Still Need Workflow Changes
The study highlights a gap between identifying a missed referral and ensuring that the patient receives the required care. MR-AI can alert staff to a possible omission after a consultation, but detection alone does not guarantee that a referral is completed or that a patient attends a specialist appointment.
The researchers say lasting improvements will require preventive changes to clinical workflows alongside AI-based monitoring. They also identify cost-effectiveness as an area for future assessment. The evaluation took place across seven clinics within one Singapore healthcare system, so further work is needed to understand how well the approach transfers to other health systems.
The findings offer a practical example of AI being used as a safety layer in routine healthcare administration. Its value will depend not only on how accurately it identifies potential omissions, but also on whether clinics can turn those alerts into timely follow-up for patients with missed specialist referrals.
