Addressing the healthcare workforce crisis with Dora: A multilingual, accessible, telephone-based AI conversational agent for cataract surgery follow up

Research question Can an artificial intelligence, automated telephone agent deliver regulated clinical consultations in multiple languages?

Background An estimated 1 million people in the UK feel that they have either no or limited English proficiency, and face language barriers to accessing healthcare. Language barriers have been correlated with reduced engagement and worse health outcomes. Patients rely on translation services, which aren t always available.

Ufonia Limited have developed an AI voice-based software-as-a-medical device, Dora , that can conduct clinical-grade conversations over the telephone.

Dora has automated over 15,000 consultations that were previously delivered by clinicians, benefitting patients by providing accessible, timely care. At present, Dora is only regulated in English.

Aims and Objectives To use state of the art AI technology to develop a clinical-grade machine-learning translator to deliver live AI conversations in a patient s preferred non-English language To validate the conversation in a trial setting to assess the safety and effectiveness when delivering cataract follow-up To build a dynamic safety case that supports regulatory clearance To evaluate the cost, efficiency, and broader social benefits (triple bottom line) Methods This is a single-arm, prospective trial of Dora used in up to 10 of the most common languages at Moorfields Eye Hospital, London, a site chosen for its demographic and linguistic diversity. 800 consented patients will receive a cataract follow up call from Dora in their preferred language.

To validate the safety and efficacy of these conversations in each language the Dora call assessment will be compared to the understanding of a masked bilingual clinician listening to the call and findings at standard face-to-face clinic review.

In addition to directly validating these 10 languages, the novel technical approach is to have an AI-based supervisor to capture any potential safety issues within a call.

This in combination with the development of a dynamic safety case will support Dora s regulatory clearance and generate a safe and reproducible governance framework for scaling new language and medical use-cases in the future.

Fundamental to this work is the patient and public involvement workstream which will include lay-member evaluation of a subset of the conversations, and focus groups for non-English speakers to understand how AI can address the cultural and linguistic nuances and inform product development.

The triple bottom line assessment will include a cost-comparison analysis comparing use of Dora with the pre-existing pathway.

Data for assessment of the environmental and social impact will come from travel distances, time for patients/friend/relative/carer to attend hospital and mapping to the index of multiple deprivation to model inequalities.

Timelines: Development (6-months) of product and ethics approval Deployment (12 -months) in a clinical trial setting Dissemination (6-months) via data analysis, regulatory submission and business case for wider adoption Anticipated Impact Multilingual capability will accelerate Dora s national and international growth and impact beyond Ufonia s current growing pipeline of 25 NHS organisations.

Dissemination will be supported by the PPI groups, Health Innovation Networks and NHS England. This will include peer-reviewed papers and multi-language electronic media.