Artificial intelligence can accelerate healthcare software design

ANDArtificial intelligence (AI) has helped clinicians speed up the design of diabetes prevention software, a new study finds.

Posted online March 6 at Journal of Medical Internet Research, the study examined the capabilities of generative artificial intelligence, or GenAI, which predicts likely options for the next word in any given sentence based on how billions of people have used words in context online. A side effect of this next word prediction is that GenAI chatbots like ChatGPT can generate answers to questions in realistic language and produce clear summaries of complex texts.

Led by researchers at NYU Langone Health, this paper explores the application of ChatGPT to the design of a software program that uses text messaging to combat diabetes by encouraging patients to eat healthier and exercise. The team tested whether AI-enabled exchanges between doctors and software engineers could accelerate the development of such a personalized automated messaging system (PAMS).

In the current study, 11 evaluators in fields ranging from medicine to computer science successfully used ChatGPT to develop a diabetes version of the tool over 40 hours. The original, non-AI-enabled effort required more than 200 programming hours.

“We found that ChatGPT improves communication between technical and non-technical team members to accelerate the design of computational solutions to medical problems,” said study author Danissa Rodriguez, PhD, MS, assistant professor in the Department of Population Health at NYU Langone and a member of its Laboratory for research, informatics and design (HiBRID) for bridging innovations in healthcare. “Chatbot has accelerated progress through the software development lifecycle, from capturing original ideas, to deciding which features to include, to generating computer code. If it proves effective on a large scale, it could revolutionize the design of healthcare software.”

Artificial intelligence as a translator

Generative AI tools are sensitive, the study authors say, and asking a question about a tool in two subtly different ways can yield different answers. The skill required to shape the questions asked by chatbots in a way that elicits the desired response, called rapid engineering, combines intuition and experimentation. Physicians and nurses, with their understanding of nuanced medical contexts, are well positioned to devise strategic instructions that improve communication with engineers, and they can do so without learning to write computer code.

However, these design efforts—in which care providers, the potential users of new software, seek to advise engineers on what it must include—can be compromised by attempts to converse using “different” technical languages. In the current study, clinical team members could type their ideas in plain English, enter them into ChatGPT, and ask the tool to convert their input into the type of language needed to guide the coding work of the team’s software engineers. AI could only drive software design so far before human software developers were needed to finally generate the code, but the overall process was greatly accelerated, the authors say.

“Our study found that ChatGPT can democratize healthcare software design by allowing doctors and nurses to drive its creation,” said senior study author Devin Mann, MD, director of the HiBRID Lab and strategic director of digital health innovation at NYU Langone Medical Center. Information Technology (MCIT). “GenAI-assisted development promises to deliver computing tools that are usable, reliable and conform to the highest coding standards.”

Along with Dr. Rodriguez and Dr. Mann, study authors from NYU Langone’s Department of Population Health were Katharine Lawrence, MD, MPH; Beatrix Brandfield-Harvey; Lynn Xu, MPH; Sumaiya Tasneem, MPH; and Defne Levine, MPH. Javier Gonzalez, technical manager at HiBRID Lab, was also an author of the study. This work was supported by National Institute of Diabetes and Digestive and Kidney Diseases grant 1R18DK118545-01A1.