DOI: 10.4103/jpdtsm.jpdtsm_41_26 ISSN: 2949-6594

Enhancing Health Literacy for People with Cystic Fibrosis: Can Artificial Intelligence – Generated Lay Summaries Improve Readability of Scientific Abstracts?

Jaime Roberts, John E. Moore, Beverley Cherie Millar
Show PDF Cite

BACKGROUND :

Cystic fibrosis (CF) is a rare disease with complex pathophysiology. This complexity makes it difficult for healthcare professionals to articulate and explain its background, diagnosis, treatments, and therapies to people with CF (PwCF), particularly with the arrival of CFTR modulators and their associated mechanisms of action. Understanding these is important for the health literacy of PwCF and to encourage optimal compliance and adherence to support good clinical outcomes. With less face-to-face contacts with healthcare professionals, there is increased use of digital resources to help source reliable information within the lay CF community. The aim of this study was therefore to explore how AI can be used to safely prepare CF-related healthcare information in lay summaries sourced from scientific abstracts, for PwCF, their parents, family, and friends.

METHODS:

Scientific abstracts ( n = 20) published in the Journal of Cystic Fibrosis (JCF) along with scientific abstracts ( n = 20) published in other scientific journals listed in PubMed with the title word “cystic fibrosis” ystic fibrosis were selected for the analysis. ChatGPT 4.0 was employed with specific prompts to write a lay summary at three reading complexities, namely (i) lay summary, (ii) at a reading level equivalent to US 8 th Grade and (iii) at a reading level equivalent to US 6 th Grade. Readability of all was calculated employing the Flesch Reading Ease and the Flesch-Kincaid Grade Level (FKGL).

RESULTS:

The readability of scientific summaries for both JCF ( n = 20) and PubMed journals ( n = 20) did not meet the recommended reading grade levels with a mean FRES of 24.63 ± 7.27 and 20.09 ± 14.93, and a mean FKGL of 14.84 ± 1.71 and 15.33 ± 2.55, respectively. Author lay summaries in JCF showed a statistically significant increased readability with a mean FRES of 46.76 ± 8.43 and an FKGL of 11.18 ± 1.34. Similarly, all summaries (lay, 8 th grade, and 6 th grade) generated by ChatGPT 4.0 showed statistically significant increased mean FRES scores of 41.00 ± 10.13, 52.66 ± 11.54, and 64.01 ± 7.49 for JCF and 39.39 ± 11.98, 49.75 ± 8.06, and 63.44 ± 8.98 for PubMed abstracts. These results were statistically significant except for summaries produced by ChatGPT 4.0 at a US 6 th grade reading level. FKGL scores decreased across all summaries generated, indicating improved readability, with mean FKGL scores of 12.49 ± 1.36, 10.61 ± 1.77, and 8.80 ± 1.21 for JCF and 13.21 ± 2.08, 11.58 ± 1.57, and 9.07 ± 1.59 for PubMed, respectively.

CONCLUSIONS :

ChatGPT 4.0 has shown it can produce information for the lay CF community that is highly readable and complete. Therefore, it may prove to be useful in the generation of lay summaries for authors in line with the European Union Clinical Trials Regulation 536/2014 and at an appropriate reading level that will promote understanding. AI platforms may become a valuable tool in the dissemination/generation of CF-related health information that supports the health literacy level of a lay rare disease community.

More from our Archive