In pilot, generative AI expected to reduce clinical documentation time at Baptist Health

Baptist Health South Florida faced a significant challenge in managing the vast amount of information generated from patient-provider dialogues across their 11 hospitals. A major issue was the time cardiologists had to invest in documenting patient visits, leading to burnout and less time for new patients. While commercial technologies were available, they were costly, prompting the development of an in-house solution.

The proposed solution involved leveraging generative AI, using services like AWS HealthScribe for medical audio transcription and tools such as Azure OpenAI, Snowflake, and DataRobot. Patient-clinician interactions were recorded with patient consent, transcribed, and processed by a large language model (GPT-4) to generate clinical summaries in SOAP format. This automation was expected to reduce documentation time to just two to five minutes post-visit, improving efficiency.

The technology journey included recording interactions, transcription via AWS HealthScribe, and using AWS Lambda and AWS S3 for AI services. Clinicians verified the AI-generated summaries for accuracy, and the data was integrated into the electronic health record system.

The results were expected to include time savings for physicians, enhanced clinical productivity with quicker turnaround times, and cost efficiency by avoiding commercial expenses.

Advice for other healthcare providers considering generative AI technology includes viewing AI as a tool to augment, not replace, human input in clinical documentation. A human-in-the-loop verification process is crucial to maintain patient record integrity. Organizations should also weigh the costs of vendors against in-house solutions and prioritize patient consent and data security when implementing AI.