P289 The initial development of a Chat-Bot for inflammatory bowel disease (IBD) patients for use in e-health applications
A. Zand*1, A. Sharma1, Z. Stokes1, C. Reynolds1, D. Hommes1
1University of California, Los Angeles, Vatche and Tamar Manoukian Division of Digestive Diseases, Center for Inflammatory Bowel Diseases, Los Angeles, USA
The emergence of Chat-Bots in healthcare through mobile applications is fast-approaching. A Chat-Bot is a natural language processor that attempts to simulate a conversation with a human user. While there have been many attempts to develop Chat-Bots that interpret and triage common symptoms and ailments, data on the feasibility of Chat-Bot development for chronic diseases like inflammatory bowel diseases (IBD) is scarce. In this study, we attempt to explore the feasibility of creating a Chat-Bot specifically for patients with IBD by categorising retrospective electronic dialogue data between patient and healthcare providers (HCP).
We used electronic dialogue data collected between 2013 and 2018 from a care management platform (eIBD) at a tertiary referral centre for IBD at the University of California, Los Angeles (UCLA). The platform includes a portal for providers and a mobile application with messaging functionality for patients. We focussed on patient to HCP dialogues only. A sample of the data were manually reviewed and an algorithm for categorisation was established. This algorithm was applied to the entire set via programming code. We successfully placed all relevant dialogues into a number of categories. Additionally, we tested the accuracy of our program by having three independent doctors evaluate the appropriateness of the categorisation by manually categorising 100 lines of randomly picked dialogue and comparing it to the categorisation of our algorithm.
In total, 16453 lines of dialogue from 1712 patients interacting with 3 IBD physicians, 3 nurses and 3 administrative assistants were collected. 8324 of these were patient to HCP interactions and we determined that 6193 were relevant for our categorisation. Ultimately, we were able to categorise the messages into seven categories, there was overlap in these categories, so we measured their frequencies independently into: symptoms (32.8%), medications (38.7%), appointments (24.5%), labs (34%), finance/insurance (7.2%), communications (34.9%), procedures (10%), and miscellaneous (10%). Additionally, our algorithm showed 94% similarity in categorisation compared with our three independent physicians.
With increased adaptation of electronic health (e-health) technologies, Chat-Bots could have great potential in interacting with patients, collecting data, and increasing efficiency. Our categorisation showcases the feasibility of using large amounts of electronic dialogue for the development of a Chat-Bot algorithm. Text-based Chat-Bot interventions in healthcare for chronic diseases such as IBD would allow for the monitoring of patients beyond consultations and potentially empower and educate patients and improve clinical outcomes.