Conversational Agents in Healthcare: A Variability Perspective

Conversational agents in healthcare are gaining popularity, for example, in the context of eliciting medical histories. Furthermore, due to the growing diversity of use cases and stakeholders, they are becoming increasingly configurable and are often based on variability mechanisms. In this paper, we present a high-level perspective on typical variability aspects and describe common challenges based on our research and practical experience in developing and evaluating conversational agents in the healthcare domain. We introduce variability aspects that are classified into technology-related (e.g., intelligence framework, input/output mode) and user-related aspects (e.g., careflow integration, health literacy). Moreover, these aspects are described in a case study on the Digital Medical Interview Assistant (DMIA) for radiology. We highlight main challenges that arise in the context of evolution, verification, input processing, privacy and security compliance, as well as ethical considerations. Our findings are intended to help developers, researchers, and healthcare professionals understand the importance and impact of configurability and to spur further discussions on variability aspects of conversational agents.