‘Educating RITTA’: evaluation of an artificial intelligence programme in opioid prescribing - a pilot project and needs assessment

Taubert, Mark ORCID: https://orcid.org/0000-0003-0454-5609 and Webb, Phil 2021. ‘Educating RITTA’: evaluation of an artificial intelligence programme in opioid prescribing - a pilot project and needs assessment. Presented at: 17th World Congress of the EAPC, Virtual, 5-8 Oct 2021. Abstracts from the 17th World Congress of the EAPC 2021. Palliative Medicine. , vol.35 (1S) Palliative Medicine: SAGE Publications, 10.1177/02692163211035909

PDF - Accepted Post-Print Version Download (520kB)

Abstract

Background/aims: Through a person centred, design thinking process, our cancer hospital palliative care team in conjunction with IBM Watson developed an Artificial Intelligence (AI) enabled virtual assistant, trained in giving basic advice on opioids. This dialogue agent is currently trained to answer a limited number of patient generated queries to demonstrate capability. Our patient/carer group suggested a hospital virtual chatbot, that could answer queries at any time of day or night. Methods: Patients, carers and healthcare professionals were tasked with creating common queries and answers around opioid prescribing. Questions and answers were programmed into the IBM Watson machine learning appliance ‘RITTA’ (Realtime Information Technology Towards Activation) with help from IBM IT engineers. Results: Phase 1 testing results: 10 patients in a palliative care outpatient clinic who had recently been prescribed opioids, were invited to write down questions on the topic of these medications in palliative care. These queries were put to RITTA after the first programming phase. 50% of questions were answered well, with further programming needs identified due to language specifics, human misspellings, dialects, jargon and variations. Programming weaknesses were also identified. Conclusions: A key theme in the development of AI is the time, care and resources required to develop Machine Learning (ML) layers. Technical work included expanding patient generated queries and machine learning in areas like palliative opioid prescribing, where a lot of repetition occurs and human medication errors or omissions can happen repeatedly and cause harm. Machine learning in palliative care has potential, but will require significant time commitment to enter thousands of question/answer variations, even for small topic areas. We identified a need for local language, slang/dialect programmes, as well as check systems on how up to date clinical guidance remains.

Item Type:	Conference or Workshop Item (Lecture)
Date Type:	Publication
Status:	Published
Schools:	Medicine
Publisher:	SAGE Publications
ISSN:	0269-2163
Funders:	Pfizer
Date of First Compliant Deposit:	26 October 2021
Date of Acceptance:	2021
Last Modified:	09 Nov 2023 02:10
URI:	https://orca.cardiff.ac.uk/id/eprint/144524

Actions (repository staff only)

Edit Item