We are representing our team today here at GIN Confrence!
It's a great opportunity to ask us questions and learn more about our products and solutions.
Come by our booth, grab your gift, and try delicious Polish candies.
We are excited to meet you!
Would you like to learn more about the customization of our tools? Discover how and why to implement our solutions in your organization. Talk to Artur, and you will know more about the possibilities of getting our solutions in your office!
Everything about GRADEpro and more! You can ask Justyna about the insights of various GRADEpro features. She will gladly show you the DEMO of our tool as well.
Are you interested in our events and news? Would you like to present your project in one of our webinars? Talk to Klaudia and share your ideas and feedback with us.
Would you like to ask anything about the methodology of GRADE and systematic reviews? Gain more insights into various research projects and features of Laser AI by talking to Jan!
Background: Plain language summaries help understand recommendations by providing more accessible and easy-to-read statements. However, creating and keeping these summaries up to date is a time-consuming task.
Objective: Evaluate whether AI can be applied to help with writing plain language summaries.
Methods: We randomly sampled 500 recommendations in English from the GRADEpro database that provided either Subgroup or Implementation Considerations. We extracted the recommendation statement and this additional information into two separate fields. We generated plain language versions of both fields using an AI system we built based on the GPT-3 model. We assessed the results using the Flesch Reading Ease formula. Three reviewers (ES, EB, JB) then manually evaluated 40 results for readability and factuality.
Results: The plain language summaries were successfully generated for 444 recommendations. The mean Flesch Reading Ease score increased from 17.86 to 46.71 (mean difference: 28.85, CI: (26.69, 31.01)) for recommendations and from 40.38 to 56.34 (MD: 15.96, CI: (12.7, 19.22)) for additional information. During the qualitative assessment of 40 recommendations, 25 were rated as improved, including three assessed as much improved. 9 results were evaluated as having the same level of readability. The evaluators also noted the good factuality of the results, although in 4 cases, the generated summary was inconsistent with the original statement.
Discussion: AI has a great promise to help maintain plain language summaries, especially if used in an interactive way. Therefore, we started the integration of AI into GRADEpro, and we plan to report further results of this work.
Nowak A 1 , Borowiack E 1 , Sadowska E 1 , Brozek J 1,2
Evidence Prime creates innovative tools for EBHC. GRADEpro (over 100k users) is the world’s most popular healthcare guideline development tool. Laser AI is a groundbreaking systematic reviews system, based on the award-winning AI model and reinvented user experience, used already by US National Toxicology Program, and German IQWIG and GBA.