Artificial intelligence (AI) can help simplify complex medical information and make it easier to understand. This is evident from research conducted by Dutch Maastricht UMC+ university hospital, published in Human Reproduction. The study shows that the use of AI does not have to compromise the accuracy of the information.
The scientists involved developed an online AI tool that allows healthcare providers to simplify texts and check their comprehension. They believe this will make a significant contribution to reducing health inequalities. "Making information more understandable is essential so that everyone – regardless of education level – can make good choices about their health," says lead researcher Masoud Zamani Esteki.
The research itself focused on genetic information surrounding pregnancy. Genetic testing can provide parents with insight into the risk of hereditary conditions in their child or even prevent genetic conditions in their child. However, information about this is often difficult to understand. This can lead some people to undergo unnecessary testing, while others forgo useful testing. For example, young women undergoing IVF treatment sometimes receive an additional test for chromosome numbers in embryos, even though this isn't always useful.
Rewriting complex texts
The researchers tested four AI models—GPT-3.5, GPT-4, Copilot, and Gemini—on their ability to rewrite complex texts about heredity and pregnancy. Examples include information about NIPT (non-invasive prenatal testing), chorionic villus sampling, amniocentesis, and embryo selection.
A total of 30 information leaflets were collected and simplified using the AI models. The readability of the texts improved significantly in all cases, reaching a grade 8 reading level — the level recommended by Dutch medical associations. GPT-4 proved most effective at making the texts more comprehensible without losing important details.
Zamani Esteki: “Our research shows that AI can help make complex medical information more accessible. However, we saw that some models omitted essential details, which can be detrimental to comprehension and decision-making. GPT-4 found a good balance.”
Online tool for healthcare providers
The researchers have also developed an online tool that can help healthcare providers simplify texts and make them more readable: the Health Literacy Evaluator. This tool works with GPT-4 and helps healthcare providers automatically simplify texts. The tool then measures the comprehensibility of the text, providing healthcare providers with immediate feedback on readability. This allows them to better tailor patient information for people with low health literacy.
The tool was developed using English-language texts. A Dutch version has since been tested by four experts, who indicate that the tool works well. Furthermore, the tool has broader applications than just simplifying pregnancy-related information. The developers hope this will contribute to making information more understandable and increasing patient trust. They consider this just as important as access to care.
"AI can contribute to health equity by better informing patients and enabling them to make informed choices about genetic testing," says Zamani Esteki. This can lead to better healthcare outcomes and reduced inequality, especially for vulnerable groups with low health literacy.
Follow-up studies
In follow-up studies, the researchers want to determine how patients experience the simplified texts and what this means for their knowledge and choices. Zamani Esteki's group is also developing an approach to investigate whether increased knowledge improves patients' mental well-being during their treatment.
The article "Artificial intelligence-simplified information to advance reproductive genetic literacy and health equity" by Marjan Naghdi and colleagues has been published in the Human Reproduction journal. doi:10.1093/humrep/deaf135.
