AB232. SOH26AB_0431. Automating referral intake and triage

Background: Irish hospitals receive an estimated 1.8 million referrals each year, creating substantial workload and variation in triage quality. Manual review is slow, inconsistent, and risks missed urgent cases. High referral volumes demand tools that support clinicians to make safer and faster decisions. Bloom is a clinician-engineered platform that automates referral data extraction and applies clinician-defined triage rules to support faster, safer decision-making. We aimed to evaluate Bloom.

Methods: Ethics approval was obtained, and all patients were consented. The study analysed 200 outpatient referrals to the Beacon Hospital Breast Clinic, comprising Healthlink and free-text formats. Clinical staff triaged referrals using standard hospital guidelines. Bloom independently extracted clinical data from referrals and applied predefined triage rules, with performance compared against human triage for accuracy. Agreement with initial clinical staff triage was deemed correct; disagreements were adjudicated by blinded consultant review.

Results: Bloom achieved 95.9% accuracy, correctly identifying all urgent cases. Data extraction succeeded for 98.4% of referrals, including 100% of Healthlink referrals. A total of 15 referrals were excluded for not meeting the inclusion criteria. Bloom upgraded 17 referrals from “soon” to “urgent”.

Conclusions: Bloom demonstrated high-accuracy clinician-verified triage and reliable referral data extraction on our sample, supporting its value in improving safety and reducing delays to outpatient assessment. These findings support progression to larger live pilots and position Bloom for scalable national deployment. With structured, rule-based triage outputs consistently generated, future work can extend this foundation into downstream operational automation, including the allocation of outpatient appointments based on triage category and clinic capacity.

Keywords: Outpatient referral triage; clinical decision support; rules-based automation; data extraction; electronic referrals