Building Generative AI Solutions for the Health Care Industry

The health care industry is committed to high-quality care and great outcomes for patients—and along with that mission come important but time-consuming documentation tasks. Helios, a division of Fresenius and a leading hospital chain operating throughout Germany, had already begun to work toward a comprehensive digital ambition when its team decided to also explore using Generative AI to simplify and improve some of that documentation work. One logical place to start was with discharge letters, because of their sheer volume and importance for the course of treatment, as well as Helios’ relationship with external healthcare professionals. Besides, streamlining the process could improve medical care and allow doctors and nurses to spend more of their valuable time treating and caring for patients.

Jointly with BCG, Helios wanted to pioneer GenAI by working on a proof of concept to automate this complex task, while leveraging learnings for their overall digital strategy.

In all, about 20 stakeholders from Helios, parent company Fresenius, and BCG engaged on solving the issue.

“The collaboration with BCG allowed us to gain hands-on experience with GenAI, unlock previously unavailable data, learn what it takes to use GenAI in a hospital setting, and sharpen our digital strategy going forward,” said Helios CEO Robert Möller.

BCG closely collaborated with Helios health care professionals in Berlin and Leipzig to identify how to best support the discharge letter creation process. This has long been necessarily complex, given that it involves several people and the responsible doctor needs to eventually synthesize the entire hospital stay, documented across various files and systems to inform subsequent care. BCG’s thorough examination of hospital processes found over 50 friction points that could be addressed.

BCG worked with Helios’ various IT teams to find ways to acquire patient data that doctors would consider before generating a discharge letter from the hospital information system, all of which was previously difficult to retrieve. Recognizing that this was a sensitive environment, and that patient privacy was essential, BCG developed the proof of concept for the cardiology department in an on-the-premises environment.

Creating this proof of concept required painstakingly exporting, pseudonymizing, cleaning, and structuring several thousand patient cases and more than 120,000 documents. Based on real patient data and historic discharge letters, BCG fine-tuned two open-source large language models (LLMs) for the specific task in several iterations, leveraging state-of-the-art data science practices. Using the fine-tuned LLMs, discharge letters were generated and then validated by health care professionals, who provided very promising feedback supporting the use case’s potential to improve patient care.

Given this success, BCG detailed a clear plan on how to expand the proof of concept to a medical-grade tool that allows doctors to generate discharge letter drafts with just the touch of a button and to quickly review its contents—saving hours of work per day and ensuring that all important information is covered by the letter.

Beyond that, BCG helped Helios identify how to scale GenAI as part of its broader digital strategy going forward, all of which could lead to an excellent prognosis for improving medical care and allowing medical personnel to focus on the most valuable part of their work—caring for their patients.