RE-AIMED: Readjusted responses by use of AI in medical calls

In RE-AIMED we are exploring how the use of artificial intelligence (AI) can support medical assessments made by operators in medical call centres. We are making a service that presents the call operators with questions adjusted to the conversation with the caller via a web-interface. By choosing pairs of answers and questions, the call is automatically documented at the same time as the operator receives help to recognize medical patterns. This is especially important for acute and severe conditions, which are rare reasons for calls to medical call centres. The main challenges in the project are how to prevent the AI from misclassifying the few, severe cases, and how to design a user interface that provides information and guidance to the operator without disrupting the operator's ability to focus and communicate with the caller. The project started in April 2020. Since the start, we have built a database of questions, answers, and pairs of questions and answers that suggest acute and severe medical conditions. This database has been coupled with deterministic machine learning algorithms, and constitutes a prototype for support of documentation of and progression in the conversation with the caller. The prototype has been refined through several iterations. In 2023 it was tested in an experiment. We are currently preparing the data for analyses. Throughout the year, results from the project has been presented at national and international conferences. We have also published the first scientific papers based on work in the project.

Medical call centres are increasingly used to ease the pressure on emergency medical health care services. In their work situation, medical call centre operators have to relate to several non-integrated tools, which put significant cognitive load on them and reduces their attentiveness, efficiency and communicative ability. To enhance the quality of the decisions made, the operators use decision support tools, which steer the conversation into predefined pathways. Together, these factors interrupt the natural flow of the conversation, and have detrimental effect on the quality of the communication and how the call is handled. RE-AIMED will explore how to improve the communication with the caller and the operator's workflow by using artificial intelligence to suggest relevant questions which are sensitive to the context, and to identify medically significant patterns by analysing questions and answers. Choice of relevant questions and answers will help the operator to document the conversation in real-time. To provide context-specific suggestions, an ensemble of methods that not only analyses content, but also considers multiple characteristics like the pace of the conversation, have to be applied. As the medical urgency of reasons for contacting medical call centres are skewed, with few urgent cases, further research into new ways of overcoming imbalance-problems in prediction of minority classes is needed to ensure that rare, but serious, medical cases are not missed. Real-time documentation will demand new ways of using an intelligent user interface to require only a minimum of input to select appropriate questions, so that the operator can concentrate on the caller. Automatic documentation will generate large data sets of standardised and detail-coded descriptions of medical calls. The collected data will be a rich source for research on telephone triage, medical decision making, communication and reduction of biases in machine learning.