AI in medical writing

The use of AI in medical communications has been an area of active discussion in the industry during the last two years. The period has been characterised by extensive investigation of how the technology can drive efficiency in content creation. Whilst the area has spawned multiple start-up companies and corporate initiatives, full scale rollouts are still uncommon. Whilst AI looks set to improve productivity, the need for human oversight is universally acknowledged across the board. 

Differing approaches for different outputs

At Springer Health+ we have been closely following the use and potential of generative AI to enhance the work of our medical writers. In 2023 we successfully developed a process for article summarisation which allowed us to deliver a variety of summary formats including Plain Language Summaries (PLS). Subsequently to that we have looked at the use of GenAI to create a draft or outline from a clinical study report (CSR). This has shown some promise, though there are also a number of hurdles to overcome to make the technology more generally applicable in this area. When given the right prompts by the medical writer, we found that the system could pick up study endpoints and produce accurate narrative however it became clear that different therapy areas and study types required adaptation. Once more, the need to ensure an accurate and ethical approach is the responsibility of the writer overseeing the process. 

Balancing data and narrative 

Whilst a CSR is a potential starting point for the AI, a more useful approach would be able to generate descriptive content from the starting point of data tables. This aspect of the use of AI in medical writing follows on from the fact that numerical doesn't follow narrative patterns. A table isn't a narrative story, it’s a map of relations where correlations can be more important than sequences. This presents the possibility that other forms of AI based around mathematical reasoning may be more suitable for creating initial interpretations of data sets with LLMs fulfilling a downstream role in creating a full narrative. As the field matures, we are likely to see a range of different approaches and techniques emerge to tackle different scenarios. Subject matter experts will ultimately drive the choice of specific technologies and external independent review will always be required prior to publication. Moreover, AI assistance must always be disclosed to authors and readers. 

We are continuing to evaluate the use of all aspects of AI in medical communications and publication, and are working closely with the wider AI and ethics teams  group at Springer Nature to explore the application of AI across scientific research within Springer Nature’s AI governance principles. As we explore deeper into the research process, we will quite possibly find areas where LLMs may need to be supplemented by other AI approaches. As with all our work in this field, the guiding principles of ethics and sustainability will always be paramount. 

Bryce McMurray

VP Scientific, Medical Communications and Education