At Baptist Well being South Florida, which incorporates 11 hospitals in Miami-Dade, Broward and Palm Seaside Counties, suppliers had been grappling with an awesome problem of managing giant quantities of knowledge emanating from patient-provider dialogues.
THE PROBLEM
One main ache level was the period of time cardiologists needed to put money into documenting affected person visits. This guide documentation course of was not solely time-consuming but in addition exacerbated supplier burnout and fatigue. As a number of the docs identified, the longer time spent on scientific documentation meant clinicians had fewer alternatives to take care of new sufferers.
“Whereas business applied sciences had been accessible available on the market, they got here with a hefty price ticket,” stated Douglas Davila-Pestana, Baptist Well being’s technical supervisor of AI – one other instance of a current development in AI-specific job titles in healthcare organizations.
“Given the difficult monetary local weather many healthcare organizations, together with ours, had been navigating, there was an pressing want for a cheaper and environment friendly system to streamline scientific documentation.
“We labored to have generative AI built-in into an AI-assisted documentation app that seamlessly blended medical transcription know-how with superior AI, particularly giant language fashions,” he continued. “This distinctive mixture meant the AI might swiftly generate scientific notes from transcribed affected person conversations.”
“Earlier than investing, organizations ought to weigh the prices of accessible distributors towards the feasibility of creating an in-house resolution, retaining in thoughts the distinctive wants and experience accessible inside their establishments.”
Douglas Davila-Pestana, Baptist Well being South Florida
The good thing about the system was its immediacy: Clinicians might have entry to complete scientific notes shortly after concluding a affected person go to, slicing down the lag time historically related to guide documentation processes, he added.
PROPOSAL
“To counter this guide documentation dilemma, a groundbreaking proposition was made: the usage of generative AI, leveraging companies corresponding to AWS HealthScribe for medical audio transcription and different instruments like Azure OpenAI, Snowflake and DataRobot,” stated Jaymin Patel, supervisor of knowledge platform and engineering at Baptist Well being South Florida.
“The concept was to report patient-clinician interactions with the affected person’s consent, transcribe these recordings into textual content, after which use a big language mannequin to generate scientific summaries in a scientific SOAP format,” he defined.
“This automation course of was anticipated to drastically cut back the documentation time to simply two to 5 minutes post-visit. Furthermore, by integrating with Snowflake, which helps our information lake, information warehouse and reporting layer, it was proposed that affected person and appointment information may very well be simply pulled from our information warehouse platform to additional improve and personalize these summaries.”
GPT-4, backed by analysis printed within the Journal of the American Medical Affiliation, was highlighted as a promising LLM given its demonstrated proficiency in suggesting medical diagnoses. All these measures aimed to enhance the affected person expertise, enhance scientific productiveness, improve operational effectivity, and result in important value financial savings.
“As well as,” Patel stated, “the information generated by the appliance was deliberate to be saved within the Snowflake information lake layer for additional evaluation and software enhancements.”
MEETING THE CHALLENGE
The know-how journey started with the recording of patient-clinician interactions. These recordings had been despatched to the AWS HealthScribe service for transcription. As soon as transcribed, the textual content was fed into the big language mannequin, which incorporates GPT-4. This mannequin, by deciphering the transcription, generated concise summaries in scientific SOAP format, which then had been verified by clinicians for accuracy.
“Cardiology physicians, together with their beta tester group of physicians, had been the first customers,” Davila-Pestana famous. “For the tech stack, AWS Lambda and AWS S3 had been pivotal in constructing the AI service. The second an audio transcription file was generated, the AI Lambda Service, constructed on AWS, received activated.
“This automation course of was anticipated to drastically cut back the documentation time to simply two to 5 minutes post-visit.”
Jaymin Patel, Baptist Well being South Florida
“Integration with Snowflake was important to extract affected person appointment information,” he continued. “Moreover, as soon as the abstract was clinician-edited and authorised, it could get exported and built-in with our digital well being report system. This whole course of ensures well timed, correct and standardized scientific documentation.”
RESULTS
This use of AI is anticipated to supply important leads to three main areas, Davila-Pestana defined:
-
Time financial savings. The group of physicians observe a discount of a number of minutes per affected person interplay, with regard to the general time spent on scientific documentation. This newfound effectivity supplies them with extra bandwidth to take care of new sufferers, thus enhancing the affected person care expertise.
-
Enhanced scientific productiveness. With the AI-assisted summaries, clinicians now not have to attend for hours as they did with earlier instruments. The summaries now will be generated in mere minutes, leading to faster turnaround instances and extra streamlined operations.
-
Price effectivity. By adopting this strategy, Baptist Well being South Florida sidestepped the appreciable bills of commercially accessible choices, tapping into the group’s mental sources and know-how and capitalizing on the efficiencies of AI.
ADVICE FOR OTHERS
“For healthcare suppliers considering the incorporation of generative AI know-how, it is essential to view AI as a software that augments, relatively than replaces, human enter in scientific documentation,” Davila-Pestana suggested. “AI can result in super efficiencies, however blind reliance can result in oversight.
“Making certain a human-in-the-loop verification course of, the place clinicians evaluation and validate the accuracy of AI-generated summaries, is paramount to take care of the integrity of affected person information,” he added. “Moreover, earlier than investing, organizations ought to weigh the prices of accessible distributors towards the feasibility of creating an in-house resolution, retaining in thoughts the distinctive wants and experience accessible inside their establishments.”
Lastly, any AI implementation ought to prioritize affected person consent and information safety to take care of belief and cling to healthcare rules, he stated.
Comply with Invoice’s HIT protection on LinkedIn: Invoice Siwicki
Electronic mail him: bsiwicki@himss.org
Healthcare IT Information is a HIMSS Media publication.
