Abstract
OBJECTIVES: Increasing demand for haematological specialist care makes the optimisation of referrals and outpatient workflow a priority. Automated placing of standardised test orders prior to the first appointment may provide haematologists with necessary information to reach diagnoses and initiate treatment at the first patient encounter, reducing low-value follow-up appointments. We aimed to evaluate rates of patient participation in an initiative using artificial intelligence to place standardised test orders as well as reasons for non-participation, differences in the number of participants and non-participants discharged back to primary care with a diagnosis or appropriate treatment plan, and potentially avoidable referrals. DESIGN: A retrospective, multicentric cohort study. SETTING: Four academic hospitals in Madrid, Spain. PARTICIPANTS: 18 190 patients referred for a first haematologist appointment for 11 included presenting complaints. INTERVENTION: Referral notes from primary care were classified using natural language processing and automated placement of standardised test order sets was carried out prior to first appointment for participating patients. OUTCOME MEASURES: We compared demographic differences between participants and non-participants, the main motives for not participating, and the number of patients discharged back to primary care at first appointment with a diagnosis and treatment plan. Most frequent International Classification of Diseases, tenth revision codes for each of the included presenting complaints were described. RESULTS: During the study period, 18 190 (41%) patients were referred for a first haematologist appointment for presenting complaints included in the intervention ('eligible patients'), of which 612 (3.3%) patients agreed to participate in the intervention. Participants were significantly younger than non-participants. Most common motives for not participating were administrative reasons (6268, 76.9%). Only 122 (1.5%) patients expressed explicit unwillingness to participate. A significant increase in the number of patients discharged upon first appointment was observed for participants (146 (23.9%) vs 3375 (19.36%); p=0.041), signifying a 22% relative reduction in avoidable follow-up. The diagnosis 'haematological disorders ruled out' was constantly observed as one of the ten most common diagnoses made by the haematology specialist for all but one of the included presenting complaints. CONCLUSION: Natural language processing of referrals from primary to specialist haematology care with automated placing of standardised test orders can decrease low-value follow-up appointments. Explicit refusal to participate was low. Participants tended to be younger than non-participants, underlining the importance of designing strategies to target the older population in order to improve participation.