Abstract
Background
Shortage of health care professionals is a worldwide problem. AI powered conversational assistants can automate non-urgent routine tasks to help nurses focus on more critical aspects of patients’ care. Preparing a patient for cardiac catheterization requires critical but repetitive tasks.
Purpose
Our goal was to evaluate the effectiveness and reliability of an AI voice assistant, Sofiya, in partially automating patient pre-procedural onboarding process during a 90-day implementation period.
Methods
Sofiya called patients scheduled for the next day’s cardiac catheterization procedures to (1) provide instructions on arrival and time, directions, parking, diet, and discharge; (2) collect patient’s data on allergies and medications, and (3) answer patient’s questions or re-directs patient to a nurse. Large Language Model (LLM) were utilized to process input text sequences based on trained medical and procedural knowledge and generating contextual and accurate responses to patient queries. All call transcripts were recorded and monitored for quality assurance and data analysis by registered nurses. The primary study outcome was the mean rate of successfully completed calls over the study period. Patient satisfaction was evaluated using surveys after procedure.
Results
From January 16 to April 16, 2025, 701 patients received 806 calls from Sofiya with pre-procedural instruction and clinical questionnaire. The overall rate of successfully completed calls was 86.4%. In the completed calls group, 295 (36.6%) calls were fully automated and did not need a callback from the nurses after review, 303 (37.7%) calls required a quick protocol-defined clarifying call, and 98 (12.2%) calls were converted to a human call. One hundred and ten (13.6%) calls did not reach the end and were classified as incomplete due to either patient-related issues 62 (7.7%) or AI system errors 48 (6.0%). The effectiveness of AI automation showed significant improvement during the 3-month implementation phase and reached a weekly completion rate exceeding 90% during the last month of the pilot (Figure). Post-procedural patient satisfaction score was 94.9%.
Conclusion
A voice-based AI assistant can efficiently automate routine tasks without compromising patient-centered care. Overcoming barriers to AI adoption can further improve the system's efficiency.
Weekly trend of AI calls reaching end