Quality evaluation of digital voice assistants for the management of mental health conditions
Digital voice assistants (DVAs) are gaining increasing popularity as a tool for accessing online mental health information. However, the quality of information provided by DVAs is not known. This study seeks to evaluate the quality of DVA responses to mental health-related queries in relation to six quality domains: comprehension ability, relevance, comprehensiveness, accuracy, understandability and reliability.
Materials and methods:
Four smartphone DVAs were evaluated: Apple Siri, Samsung Bixby, Google Assistant and Amazon Alexa. Sixty-six questions and answers on mental health conditions (depression, anxiety, obsessive-compulsive disorder (OCD) and bipolar disorder) were compiled from authoritative sources, clinical guidelines and public search trends. Three evaluators scored the DVAs from an in-house-developed evaluation rubric. Data were analyzed by using the Kruskal-Wallis and Wilcoxon rank sum tests.
Across all questions, Google Assistant scored the highest (78.9%), while Alexa scored the lowest (64.5%). Siri (83.9%), Bixby (87.7%) and Google Assistant (87.4%) scored the best for questions on depression, while Alexa (72.3%) scored the best for OCD questions. Bixby scored the lowest for questions on general mental health (0%) and OCD (0%) compared to all other DVAs. In terms of the quality domains, Google Assistant scored significantly higher for comprehension ability compared to Siri (100% versus 88.9%, p < 0.001) and Bixby (100% versus 94.5%, p < 0.001). Moreover, Google Assistant also scored significantly higher than Siri (100% versus 66.7%, p < 0.001) and Alexa (100% versus 75.0%, p < 0.001) in terms of relevance. In contrast, Alexa scored the worst in terms of accuracy (75.0%), reliability (58.3%) and comprehensiveness (22.2%) compared to all other DVAs.
Overall, Google Assistant performed the best in terms of responding to the mental health-related queries, while Alexa performed the worst. While the comprehension abilities of the DVAs were good, the DVAs had differing performances in the other quality domains. The responses by DVAs should be supplemented with other information from authoritative sources, and users should seek the help and advice of a healthcare professional when managing their mental health conditions.