Safety And Efficacy Of An Automated Telephone Call For Post-Op Follow Up After Cataract Surgery.

Demands on ophthalmology services are rising. Many parts of the care pathway are highly stereotyped, repetitive, and burnout inducing. Dora is a UKCA marked artificial intelligence (AI) enabled autonomous conversational assistant that can conduct routine telephone conversations with patients and enable greater efficiency in the delivery of care. This study aimed to assess the safety and effectiveness of using Dora for post-operative cataract follow-up.

Patients were recruited from two UK teaching hospitals with different demographic profiles: Oxford University Hospitals NHS Foundation Trust, and Imperial College Healthcare NHS Trust. Any patient having routine, uncomplicated cataract surgery was eligible for inclusion. Patients had a Dora call in addition to the standard of care.  The work was funded by an NIHR Artificial Intelligence in Health and Care Award. ClinicalTrials.gov identifier: NCT05213390, ethics approval: 21/PR/0767. 

Patients were recruited with informed consent following uncomplicated cataract surgery at Oxford Eye and Imperial College Hospitals in a prospective study between Sep 2021- Jan 2022. All participants received a Dora call 3-4 weeks postoperatively. The primary outcome measure was agreement between Dora and a supervising ophthalmologist (blinded to Dora’s decision) in terms of symptom assessment and overall management decision. The presence of unexpected management changes (UMC) during a planned review or unplanned emergency attendances were identified from examining the clinical notes within 3 months post-operatively. 

197 (98%) patients with a mean age of 74 completed the Dora call. In this study 59% of the cohort would not need further routine clinician input. For the five key symptoms, the sensitivity of Dora compared to clinician in detecting significant red eye, pain or flashing lights was 100%, but 79% for vision concerns and 89% for floaters. In terms of overall management decision, the sensitivity of Dora in identifying patients who the supervising ophthalmologist deemed to require clinical review was 94% (95% CI: 85-98) with a specificity of 86% (95% CI:80-92), Kappa agreement 0.76 (p=<0.001). There were no unplanned attendances with unexpected management changes within 2 weeks of the Dora call. 

Dora is able to autonomously elicit relevant symptoms and identify patients who require clinical review with a robust level of agreement with a human ophthalmologist. The algorithm was locked during the study period, but future improvement can be made to improve sensitivity in assessing vision concerns and floaters. Our results demonstrate the viability for routine cataract patients to be autonomously assessed by an AI conservation agent in order to improve clinical resource allocation.