Aiding clinical triage with text classification

Abstract

SNS24 is a telephone service for triage, counselling, and referral service provided by the Portuguese National Health Service. Cur-rently, following the predefined 59 Clinical Pathways, the selection of themost appropriate one is manually done by nurses. This paper presents astudy on using automatic text classification to aid on the clinical path-way selection. The experiments were carried out on 3 months calls data containing 269,669 records and a selection of the best combination often text representations and four machine learning algorithm was pursued by building 40 different models.Then, fine-tuning of the algorithm parameters and the text embedding model were performed achieving afinal accuracy of 78.80% and F1 of 78.45%. The best setup was then used to calculate the accuracy of the top-3 and top-5 most probable clinical pathways, reaching values of 94.10% and 96.82%, respectively. These results suggest that using a machine learning approach to aid the clinical triage in phone call services is effective and promising.