An artificial intelligence-based first-line defence against COVID-19: digitally screening citizens for risks via a chatbot

Alistair Martin; Jama Nateqi; Stefanie Gruarin; Nicolas Munsch; Isselmou Abdarahmane; Bernhard Knapp

Above, we compared the symptom-to-disease search engine Symptoma to alternative methods on both COVID-19 and BMJderived decoy cases. We showed that Symptoma is superior to these alternative approaches in multiple aspects. First, Symptoma can correctly identify cases of COVID-19, not only against a few diseases but against more than 20,000 other potential causes. This far exceeds the 6,000 differential diagnoses provided by the second-largest symptom checker "Isabel Healthcare" 5 . Secondly, we showed that Symptoma needs only a few symptoms to highlight cases of COVID-19. Crucially, these can be inputted as free text that is semantically understood. For example, if one enters "Tiramisu", "Food Poisoning" is returned as one of the top results. In contrast, other COVID-19 questionnaires and symptom checkers require that patients only select from a predefined list of symptoms or are constrained to fixed questions. Furthermore, Symptoma is localised into 36 languages, allowing a standardised approach on disease predictions globally. Lastly, we found that Symptoma performs better than simple questionnaire-based approached in identifying COVID-19 cases, even though these alternative approaches only considered three alternative diagnoses (COVID-19, influenza, common cold, and hay fever). Moreover, as Symptoma is constantly mining the newest literature, it keeps up-to-date with the latest knowledge and alters its predictions accordingly, e.g. the recent reports of anosmia in COVID-19 patients 6, 7 . Similarly, by tracking cases of COVID-19 entered into Symptoma, due to the free-text function, new symptoms related to COVID-19 may be highlighted. On these grounds, we believe that Symptoma is a highly valuable tool in the global COVID-19 crisis.