Autonomous Telephone Follow-up After Cataract Surgery

This project will apply AI technology to meet the gap between increasing demand and limited capacity of high- volume healthcare services. The project will develop evidence that will support the safe deployment of Ufonia's automated telemedicine platform to deliver calls to cataract surgery patients at two large NHS hospital trusts. The proposed study will implement DORA in addition to the current standard of care for a cohort of patients at Imperial College Healthcare Trust and Oxford University Hospitals NHS Foundation Trust. The study will evaluate the agreement of DORA's decision with an expert clinician. In addition it will test the acceptability of the solution for patients and clinicians; the sensitivity and specificity of the system in deciding if a patient requires additional review; and the health economic benefits of the solution to patients (reduced time and travel) and the local healthcare system. If successful, a proposal will be developed to roll the solution out to all patients at each site in anticipation of an application to a late phase award for wider NHS deployment.

Background Due to an ageing population and increased expectation, the demand for many services is exceeding the capacity of the clinical workforce. As a result, staff are facing a crisis of burnout from being pressured to deliver high- volume workloads, driving increasing costs for providers. Artificial intelligence, in the form of conversational agents, presents a possible opportunity to enable efficiencies in the delivery of care. Aims and Objectives This study aims to evaluate the effectiveness, usability and acceptability of DORA - an AI-enabled autonomous telemedicine call - for detection of post-operative cataract surgery patients who require further assessment. The study's objectives are: to establish efficacy of DORA's decision making in comparison to an expert human clinician; baseline sensitivity and specificity for detection of true complications; evaluation of patient acceptability; evidence for cost-effectiveness; and to capture data that may support further studies. Project plan and methods used Based on implementation science, the interdisciplinary study will be a mixed-methods phase one pilot establishing inter-observer reliability; as well as usability and acceptability. Timelines for delivery The study will last eighteen months: seven months of evaluation and intervention refinement, nine months of implementation and follow-up, and two months of post-evaluation analysis and write-up. Anticipated Impact and Dissemination The project's key contributions will be evidence on artificial intelligence voice conversational agent effectiveness, and associated usability and acceptability. Results will be disseminated in peer-reviewed journals and at international medical sciences and engineering conferences.

The study will be a multi-centre, mixed-methods clinical investigation to develop evidence regarding the feasibility, acceptability and potential effectiveness of Dora.

DORA uses a variety of AI technologies to deliver the patient follow-up call, including: speech transcription, natural language understanding, a machine-learning conversation model to enable contextual conversations, and speech generation. Together, these technologies cover the input, processing and analysis, and output needed to maintain a natural conversation. DORA is configured to deliver calls through a telephone connection as a real-time, stand-alone system: the operator inputs individual patient details to initiate the call and completes a summary in the electronic health record (EHR) afterwards. The entire conversation will be supervised by a clinician. This clinician will be able to interrupt the call at any point if the system fails, the patient struggles to interact with it, or DORA does not collect sufficient information from the patient. The clinician will record a clinical assessment which will be compared to the DORA assessment.

DORA uses a variety of AI technologies to deliver the patient follow-up call, including: speech transcription, natural language understanding, a machine-learning conversation model to enable contextual conversations, and speech generation. Together, these technologies cover the input, processing and analysis, and output needed to maintain a natural conversation. DORA is configured to deliver calls through a telephone connection as a real-time, stand-alone system: the operator inputs individual patient details to initiate the call and completes a summary in the electronic health record (EHR) afterwards.

Inter-rater reliability: the degree of agreement between DORA and the clinician on their assessments of the individual symptoms and the management plan; Whether or not the clinician had to interrupt the call to ask clarifying questions

Complications identified from patients' electronic health records up to 90 days following cataract surgery; Congruence between complications identified and management planned in DORA call and face-to-face follow up (Imperial); Comparison to data from patients attending eye casualty (Oxford)

Proportion of autonomous calls that were completed without needing any intervention from the supervising clinician; Clinician-reported reasons for asking clarifying questions

Measured using the System Usability Scale (minimum of 0, maximum of 100, higher scores indicate better usability)

Measured using the Telehealth Usability Questionnaire (minimum score of 1, maximum score of 5, averaged across 19 items; higher scores indicate better usability)

Inclusion Criteria: Willing and able to provide informed consent; Aged 18 years or older; On the waiting list for routine cataract surgery. Cataract surgery as part of a combined procedure with other ocular surgery will not be included; No history or presence of significant ocular comorbidities that would be expected to alter the risks of cataract surgery or normal post-operative follow-up schedule. Note that significant ocular comorbidities do not include stable, chronic, or inactive ocular conditions such as amblyopia, drop-controlled stable glaucoma or ocular hypertension, previous squint surgery, inactive macular pathology, previous refractive surgery, or previous vitreoretinal surgery with stable retina. Exclusion Criteria: Individuals with any condition that could preclude the ability to comply with the study or follow-up procedures; Presence of ocular or systemic uncontrolled disease (unless deemed not clinically significant by the Investigator and Sponsor); Involved in current research related to this technology or been involved in related research to this technology prior to recruitment; Cognitive difficulties, hearing impairment or non-English speakers; History of current or severe, unstable or uncontrolled systemic disease (unless deemed not clinically significant by the Investigator and Sponsor).

Milne-Ives M, de Cock C, Lim E, Shehadeh MH, de Pennington N, Mole G, Normando E, Meinert E. The Effectiveness of Artificial Intelligence Conversational Agents in Health Care: Systematic Review. J Med Internet Res. 2020 Oct 22;22(10):e20346. doi: 10.2196/20346.

de Pennington N, Mole G, Lim E, Milne-Ives M, Normando E, Xue K, Meinert E. Safety and Acceptability of a Natural Language Artificial Intelligence Assistant to Deliver Clinical Follow-up to Cataract Surgery Patients: Proposal. JMIR Res Protoc. 2021 Jul 28;10(7):e27227. doi: 10.2196/27227.