Digital biomarker co‐design with people with preclinical or prodromal dementia

Abstract

AbstractBackgroundLanguage markers can serve screening and early detection in Alzheimer’s disease (AD). Using artificial intelligence (AI) approaches enables automated assessment and analysis of speech biomarkes using chatbot technology. Reliable digital speech biomarkers require an easy‐to‐use chatbot which is adopted to the users’ needs. Currently, chatbots often are unsuitable for people with cognitive impairment and rather frustrating for their users. Here, we implemented a user centred design approach to iteratively evaluate usability of a chatbot system for automated speech assessments for people with prodromal or preclinical dementia.MethodWithin the study PROSPECT‐AD, participants (n = 53 by January 2023, 300 planned) are automatically called for six times every three months by our chatbot “Mili”. Each call takes about 15 minutes and consists of three cognitive tasks (Wordlist, Semantic Verbal Fluency, Story Telling). We applied a six stage usability check to improve Mili iteratively (see Figure 1). For deeper insights, we conduct semi‐structured interviews (n = 20). During the interviews, we focus on affinity for technology, usability, and participants’ attitude towards chatbots and artificial intelligence in medicine.ResultsTo date, Mili has completed 54 assessments (see Figure 2). Drop‐outs (11.3%) occcur around the first assessment. Reasons were e.g., lack of accessibility of participants, excessive demands through cognitive tasks or through Mili. Participants reported problems due to Mili’s high speaking pace and hangs of Mili. Additionally, Mili struggled in automatically retrieving appointments and missed calling. All problems were quickly resolved. The speaking pace was reduced and the time‐out time extented by 2 seconds. Results of the system usability scale revealed a high user‐friendliness of the calls (see Table 1). Participants (n = 6) evaluated the calls as quick to learn and easy to use. Two participants wished for more technical support for the calls. Two participants felt uncomfortable in using Mili.ConclusionOur results revealed a high usability and feasibility of the automated phone calls by Mili. Due to close colaboration, problems with Mili were solved qickly and further drop‐outs prevented. This underlines the benefit of close cooperation between participants, scientists, and IT specialists. Interviews are needed for further improvement of Mili and to understand uncertainty with Mili.