AI Accelerates Preterm Birth Research, Outperforming Traditional Methods
Researchers at UC San Francisco and Wayne State University have demonstrated that generative artificial intelligence can dramatically accelerate medical data analysis, in this case, identifying patterns related to preterm birth. In a direct comparison, AI systems not only processed data far faster than traditional computer science teams – who had spent months on the same task – but in some instances, delivered superior results.
The study, published in Cell Reports Medicine on , involved challenging both human experts and AI to predict preterm birth using data from over 1,000 pregnant women. This research is particularly significant given that preterm birth is the leading cause of newborn death and a major contributor to long-term cognitive and motor challenges in children, affecting roughly 1,000 babies each day in the United States.
A key finding was the ability of even a junior research duo – Reuben Sarwal, a UCSF master’s student and Victor Tarca, a high school student – to develop viable prediction models with AI assistance. The AI generated functioning computer code in a matter of minutes, a task that would typically require experienced programmers hours or even days to complete. This speed-up is crucial, according to Marina Sirota, PhD, a professor of Pediatrics and interim director of the Bakar Computational Health Sciences Institute (BCHSI) at UCSF. “These AI tools could relieve one of the biggest bottlenecks in data science: building our analysis pipelines,” Sirota said. “The speed-up couldn’t come sooner for patients who need help now.”
The Challenge of Preterm Birth Data
Understanding the causes of preterm birth remains a significant challenge for medical researchers. Sirota’s team had compiled a large dataset of microbiome data from approximately 1,200 pregnant women, collected across nine separate studies, to investigate potential risk factors. Analyzing this complex and vast dataset proved to be a major hurdle.
To overcome this, the researchers leveraged a crowdsourcing competition called DREAM (Dialogue on Reverse Engineering Assessment and Methods). Sirota co-led one of three DREAM pregnancy challenges, specifically focusing on vaginal microbiome data. Over 100 teams worldwide participated, developing machine learning models to detect patterns linked to preterm birth. While the competition itself concluded within three months, consolidating the findings and publishing the results took nearly two years.
AI Steps In: A New Approach
Recognizing the potential for AI to accelerate this process, Sirota’s group collaborated with researchers led by Adi L. Tarca, PhD, a professor at Wayne State University’s Center for Molecular Medicine and Genetics. Tarca had previously led the other two DREAM challenges, concentrating on improving methods for estimating pregnancy stage – a process that is often an estimate and impacts the type of care a pregnant woman receives.
The researchers instructed eight different AI systems to independently generate algorithms using the datasets from the three DREAM challenges, without direct human coding intervention. The AI chatbots were given carefully crafted natural language instructions, similar to the prompts used with systems like ChatGPT, guiding them to analyze the health data in a manner comparable to the original DREAM participants.
The AI’s objectives mirrored those of the earlier challenges: analyze vaginal microbiome data to identify indicators of preterm birth and examine blood or placental samples to estimate gestational age. Accurate pregnancy dating is critical for appropriate prenatal care, as inaccurate estimates can complicate labor preparation.
Results and Limitations
After running the AI-generated code using the DREAM datasets, the researchers found that four of the eight AI tools produced models that matched the performance of the human teams. In some cases, the AI models even outperformed their human counterparts. Remarkably, the entire generative AI effort – from initial instruction to paper submission – was completed in just six months.
However, the researchers emphasize that AI is not a replacement for human expertise. They caution that these systems can produce misleading results and require careful oversight. “Scientists emphasize that AI still requires careful oversight,” the research indicates. Despite this, the ability of AI to rapidly sift through massive health datasets allows researchers to dedicate more time to interpreting results and formulating meaningful scientific questions.
“Thanks to generative AI, researchers with a limited background in data science won’t always need to form wide collaborations or spend hours debugging code,” Tarca said. “They can focus on answering the right biomedical questions.”
The study authors include Reuben Sarwal, Claire Dubin, Sanchita Bhattacharya, and Atul Butte from UCSF, along with Victor Tarca (Huron High School), Nikolas Kalavros and Gustavo Stolovitzky (New York University), Gaurav Bhatti (Wayne State University), and Roberto Romero (National Institute of Child Health and Human Development).
This work was funded by the March of Dimes Prematurity Research Center at UCSF and by ImmPort, with data generated in part with support from the Pregnancy Research Branch of the NICHD.
