Performance of a triage approach for identification of patients with inflammatory rheumatic and musculoskeletal diseases among rheumatology referrals.

OBJECTIVES: This study aims to evaluate a triage approach prioritising assessment of patients with inflammatory rheumatic and musculoskeletal disease (iRMD) among rheumatology referrals. METHODS: In this prospective study, 1180 consecutive referrals to a tertiary rheumatology centre underwent a telephone interview (step 1), followed by a 10-minute rheumatologist consultation within 4 weeks (step 2), during which patients were scheduled for comprehensive inpatient assessment (step 3) immediately, within 4 weeks or later. Diagnostic performance and wait times were analysed. Machine learning (ML) models for steps 1 and 2 were retrospectively investigated. RESULTS: Among 1180 referred patients, iRMD was suspected in 413 patients (35%) and not suspected in 767 (65%), of whom 52 (4.4%) were considered sufficiently unlikely to have an iRMD to forgo further evaluation. The remaining 1128 patients were scheduled for comprehensive assessment (step 3). Of these, 148 (13.1%) dropped out, whereas 980 patients proceeded. iRMD was diagnosed in 314 patients (32%); 666 (68%) had noninflammatory conditions. Step 2 correctly identified 211 patients with iRMD (sensitivity 67.2%) and 502 patients without iRMD (specificity 75.4%). Mean (SD) time from telephone interview to final assessment was shorter for patients with iRMD than for those without iRMD (33 [30] vs 55 [24] days). The step 1 and step 2 ML models achieved area under the receiver operating characteristic curve (AUC-ROC) values of 0.73 (95% CI: 0.65-0.80) and 0.78 (95% CI: 0.70-0.86), respectively. CONCLUSIONS: The triage approach accelerated assessment for more than two-thirds of patients with iRMD, while highlighting the necessity of a full workup to capture the one-third of cases otherwise missed. Using ML may further improve triage performance.