Pregnancy puts higher oxygen demands on the body, so expectant mothers breathe significantly more air over the course of a day. When air quality is bad, they're also breathing in more harmful pollutants.
Many studies have linked individual air pollutants, such as fine particulate matter (PM2.5) or ozone, to early preterm birth-delivery before 34 weeks' gestation, when infants face higher risks of health complications. But those studies don't reflect the reality that people are often exposed to mixtures of pollutants simultaneously.
In a new University of Utah study of 44,874 first-time mothers in Utah, researchers found that exposure to even moderate levels of multiple air pollutants may increase the risk of preterm birth when experienced together-especially during critical stages of pregnancy.
The findings raise questions about whether current air quality index (AQI) ratings-typically based on the single pollutant posing the greatest harm-may be missing serious health risks to the public.
"People are exposed to multiple things at once, over multiple times," said Brenna Kelly, lead author on the paper. Kelly is a recent graduate of the University of Utah's Population Health Sciences PhD program and an incoming Responsible AI Postdoctoral Fellow. "It may matter when pollution is slightly elevated for multiple chemicals."
Using AI to untangle complex exposures
Growing evidence suggests that environmental mixtures may have a greater impact on health than individual pollutants alone. But identifying which combinations to study remains a major challenge.
"Finding these patterns is like searching for a needle in a haystack-without the machine learning component, testing all combinations of mixtures would have been intractable," Kelly said.
This study focused on air pollution and preterm births in Utah, a state that intermittently has some of the worst air quality in the world. The researchers developed an epidemiologic machine learning framework using a self-organizing map, a type of neural network that finds patterns in data. Similar models have assessed environmental mixtures, but this is the first to link them to health outcomes over time.
The scientists provided the model with high-resolution air pollution data from Utah between 2013 and 2016, focusing on temperature and three common pollutants: nitrogen dioxide (NO2), ozone (O3) and fine particulate matter PM2.5. The algorithm identified 12 distinct mixtures, and the researchers then modeled the effect of these exposures on early preterm birth during each week of pregnancy.
Early pregnancy: A critical window
Exposure to a mixture of O3 and PM2.5 in the late first trimester had the strongest relationship to early preterm birth. Because the AQI does not take multiple exposures into account, the EPA would have considered these levels "safe" air quality, although it may still pose risks to health. Women exposed to this mixture in week 11 of pregnancy had 53% greater odds of a preterm birth later in pregnancy. Additionally, those exposed repeatedly over weeks 9-14 had almost three-times greater odds of preterm birth.
"Many different pathways could lead to early preterm birth, including inflammation, infection or problems with placental development," said Michelle Debbink, associate professor of obstetrics and gynecology at the U and coauthor of the study. "Early pregnancy is a critical period because the placenta and arteries that supply blood and oxygen to the fetus are still developing. Exposure to pollutants at this juncture could impair this process, increasing the risk of complications such as preeclampsia, which can require preterm delivery. Repeated exposures could also cause inflammation and damage that accumulates over time, further increasing the risk of preterm birth."
Rethinking air quality standards
The authors hope that scientists will use this framework to better understand how other types of environmental hazards may impact human health.
"Public health policy definitely oversimplifies real-world exposures, but research into the health effects of complex mixtures is the first step in improving these policies," Kelly said.
"Brenna's work helps demonstrate the potential of machine learning and artificial intelligence in tackling complex environmental problems and in the assessment of their impacts," said Simon Brewer, professor in the U's School of Environment, Society & Sustainability and coauthor of the study. "U has been key in supporting this work, both through the DELPHI initiative, and more broadly across campus as thematic areas within the U's Responsible AI Initiative."
Other authors include Robert Silver of the U's Department of Obstetrics and Gynecology; Heather Holmes of the U's Department of Chemical Engineering; Heidi Hanson of both the U's Department of Population Health Sciences (DPH) and Oak Ridge National Laboratory; and Jennifer Doherty of the U's DPH and Huntsman Cancer Institute; and Joel Schwartz of the Harvard T.H. Chan School of Public Health.
The University of Utah's Center for High Performance Computing hosted the study data and provided the computational resources for model training.
The research was supported by a University of Utah DELPHI Data Science Initiative Seed Grant.
Header image: Wintertime inversion over Salt Lake City in December, 2010. Credit: TimeScience via Flickr
