Ask Rosa – The making of a digital genetic conversation tool, a chatbot, about hereditary breast and ovarian cancer. Issue 6 (June 2022)

• Record Type: Journal Article
• Authors: Siglen, Elen; Vetti, Hildegunn Høberg; Lunde, Aslaug Beathe Forberg; Hatlebrekke, Thomas Akselberg; Strømsvik, Nina; Hamang, Anniken; Hovland, Sigrid Tronsli; Rettberg, Jill Walker; Steen, Vidar M.; Bjorvatn, Cathrine
• Is Part Of: Patient education and counseling. Volume 105:Issue 6(2022)
• Page Start: 1488
• Page End: 1494
• Publication Date: 2022-06
• Subjects: Genetic counseling -- Genetic information -- Genetic communication -- Information tool -- Support tool -- Patient educationChatbot -- Chatbot technology -- Artificial intelligence -- Hereditary cancer -- Breast cancer -- Ovarian cancer; Patient education -- Periodicals; Health counseling -- Periodicals; Health education -- Periodicals; Counseling -- Periodicals; Patient Education -- Periodicals; Éducation des patients -- Périodiques; Counseling -- Périodiques; Éducation sanitaire -- Périodiques; 615.5071
• DOI: 10.1016/j.pec.2021.09.027
• Languages: English
• ISSNs: 0738-3991

Abstract

Highlights: Participatory methodology and iterative approaches should be chosen when developing health care chatbots. Narrow the content of the database and upscale carefully. Specifically ask for negative feedback as test persons have high tolerance for errors. Due to limitations in AI, a high level of manual work in building and maintaining the database is needed. Every answer must contain all information needed to understand the content correctly. Abstract: Objective: We aimed at developing a pilot version of an app (Rosa) that can perform digital conversations with breast or ovarian cancer patients about genetic BRCA testing, using chatbot technology, to identify best practices for future patient-focused chatbots. Methods: We chose a commercial chatbot platform and participatory methodology with a team of patient representatives, IT engineers, genetic counselors and clinical geneticists, within a nationwide collaboration. An iterative approach ensured extensive user and formal usability testing during the development process. Results: The development phase lasted for two years until the pilot version was completed in December 2019. The iteration steps disclosed major challenges in the artificial intelligence (AI)-based matching of user provided questions with predefined information in the database, leading initially to high level of fallback answers. We therefore developed strategies to reduce potential language ambiguities (e.g. BRCA1 vs BRCA2 ) and overcome dialogue confusion. The first prototype contained a database with 500 predefined questions and 67 corresponding predefined answers, while the final version included 2257 predefined questions and 144 predefined answers. Despite the limited AI functionality of the chatbot, the testing revealed that the users liked the layout and found the chatbot trustworthy and reader friendly. Conclusions: Building a health chatbot is challenging, expensive and time consuming with today's technology. The users had a positive attitude to the chatbot, and would use it in a real life setting, if given to them by health care personnel. Practice implications: We here present a framework for future health chatbot initiatives. The participatory methodology in combination with an iterative approach ensured that the patient perspective was incorporated at every level of the development process. We strongly recommend this approach in patient-centered health innovations.